chore(ci): Attempt to optimize CI build time

chore: bump fff-mcp release artifacts to v0.10.4
chore: release 0.10.4
2026-08-15 17:33:05 -07:00 · 2026-08-15 22:16:40 +00:00 · 2026-08-15 14:35:20 -07:00 · 2026-08-14 22:11:24 -07:00 · 2026-08-14 21:47:10 -07:00 · 2026-08-14 21:46:46 -07:00
91 changed files with 4632 additions and 1701 deletions
@@ -55,7 +55,7 @@ jobs:

      - name: Build Rust binary
        shell: bash
-        run: make build
+        run: make build-e2e

      - name: Verify Windows DLL has no unexpected dependencies
        if: matrix.os == 'windows-latest'
@@ -200,6 +200,10 @@ jobs:
          { "name": "@ff-labs/fff-bin-linux-x64-musl", "version": "0.0.0" }
          JSON

+      # fff-node builds with bun; the npm `bun` package ships musl binaries
+      - name: Install bun
+        run: npm install -g bun
+
      - name: Build fff-node
        working-directory: packages/fff-node
        run: npm run build
@@ -15,6 +15,8 @@ on:
 env:
  CARGO_TERM_COLOR: always
  MACOSX_DEPLOYMENT_TARGET: "13.0"
+  CARGO_PROFILE_RELEASE_LTO: thin
+  CARGO_PROFILE_RELEASE_CODEGEN_UNITS: 16

 jobs:
  test:
@@ -31,7 +33,11 @@ jobs:
          persist-credentials: false

      - name: Install Rust
-        uses: dtolnay/rust-toolchain@stable
+        uses: actions-rust-lang/setup-rust-toolchain@v1.15.4
+        with:
+          cache: true
+          cache-on-failure: true
+          cache-key: "v1-rust-python"

      - name: Setup uv
        uses: astral-sh/setup-uv@v5
@@ -43,6 +49,8 @@ jobs:
        working-directory: packages/fff-python
        shell: bash
        run: |
-          uv sync --all-extras
+          # --no-install-project: uv would otherwise build the wheel via
+          # maturin's PEP 517 backend, then `maturin develop` builds it again.
+          uv sync --all-extras --no-install-project
          uv run maturin develop --release
          uv run pytest -v
@@ -612,7 +612,7 @@ jobs:

            Update mcp via:
            ```sh
-            curl -fsSL https://raw.githubusercontent.com/dmtrKovalenko/fff.nvim/main/install-mcp.sh | bash
+            curl -fsSL https://raw.githubusercontent.com/dmtrKovalenko/fff/main/install-mcp.sh | bash
            ```

      - name: Bump Homebrew formula (uses local checksums)
@@ -710,6 +710,8 @@ jobs:
          node-version: "25"
          registry-url: "https://registry.npmjs.org"

+      - uses: oven-sh/setup-bun@v2
+
      - name: Determine version
        id: version
        run: lua scripts/determine-version.lua
@@ -747,6 +749,8 @@ jobs:
          make set-npm-version PKG=packages/fff-bun VERSION="$VERSION"

          cd packages/fff-bun
+          bun install --frozen-lockfile
+          bun run build
          npm publish --tag "$TAG" --access public --provenance

      - name: Publish Node.js package
@@ -47,7 +47,8 @@ jobs:
          components: rustfmt, clippy

      - name: Run tests
-        run: cargo test --no-default-features --features zlob --workspace --exclude fff-nvim
+        # fff-python requires full python o3 machinery which is very slow
+        run: cargo test --no-default-features --features zlob --workspace --exclude fff-nvim --exclude fff-python

  stress-test:
    name: Fuzz Tests
@@ -1,4 +1,4 @@
-- luacheck configuration for fff.nvim
+-- luacheck configuration for fff
 -- https://luacheck.readthedocs.io/en/stable/config.html

 -- Neovim globals
@@ -633,7 +633,7 @@ checksum = "37909eebbb50d72f9059c3b6d82c0463f2ff062c9e95845c43a6c9c0355411be"

 [[package]]
 name = "fff-c"
-version = "0.10.1"
+version = "0.10.4"
 dependencies = [
 "fff-query-parser",
 "fff-search",
@@ -643,7 +643,7 @@ dependencies = [

 [[package]]
 name = "fff-grep"
-version = "0.10.1"
+version = "0.10.4"
 dependencies = [
 "bstr",
 "memchr",
@@ -651,7 +651,7 @@ dependencies = [

 [[package]]
 name = "fff-mcp"
-version = "0.10.1"
+version = "0.10.4"
 dependencies = [
 "clap",
 "fff-query-parser",
@@ -662,8 +662,10 @@ dependencies = [
 "schemars",
 "serde",
 "serde_json",
+ "tempfile",
 "tokio",
 "tracing",
+ "windows-sys 0.60.2",
 ]

 [[package]]
@@ -682,7 +684,7 @@ dependencies = [

 [[package]]
 name = "fff-nvim"
-version = "0.10.1"
+version = "0.10.4"
 dependencies = [
 "ahash",
 "chrono",
@@ -703,7 +705,7 @@ dependencies = [

 [[package]]
 name = "fff-python"
-version = "0.10.1"
+version = "0.10.4"
 dependencies = [
 "fff-query-parser",
 "fff-search",
@@ -713,7 +715,7 @@ dependencies = [

 [[package]]
 name = "fff-query-parser"
-version = "0.10.1"
+version = "0.10.4"
 dependencies = [
 "criterion",
 "zlob",
@@ -721,7 +723,7 @@ dependencies = [

 [[package]]
 name = "fff-search"
-version = "0.10.1"
+version = "0.10.4"
 dependencies = [
 "ahash",
 "aho-corasick",
@@ -3235,9 +3237,9 @@ dependencies = [

 [[package]]
 name = "zlob"
-version = "1.6.2"
+version = "1.6.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "57e7ca1588981ea66f5ac470915c4cb44a47ecfb5797576f65fca1bafacedf4e"
+checksum = "f0f010d9b10495cc0037cb47cedc143c39c7fa2d59f752dca19036fd0234a210"
 dependencies = [
 "bindgen",
 "bitflags 2.11.0",
@@ -15,8 +15,8 @@ resolver = "2"
 module_inception = "allow"

 [workspace.dependencies]
-fff-grep = { version = "0.10.1", path = "crates/fff-grep" }
-fff-query-parser = { version = "0.10.1", path = "crates/fff-query-parser", default-features = false }
+fff-grep = { version = "0.10.4", path = "crates/fff-grep" }
+fff-query-parser = { version = "0.10.4", path = "crates/fff-query-parser", default-features = false }

 # Shared dependencies
 ahash = "0.8"
@@ -36,7 +36,7 @@ ignore = "0.4.22"
 memmap2 = "0.9"
 mimalloc = "0.1.47"
 signal-hook-registry = "1.4"
-zlob = { version = "=1.6.2" }
+zlob = { version = "=1.6.3" }

 mlua = { version = "0.11.1", features = ["module", "luajit"] }
 neo_frizbee = { version = "0.11.0", features = ["match_end_col"] }
@@ -2,34 +2,34 @@
 # Maintained in-repo; auto-bumped by .github/workflows/release.yaml on stable releases.
 class FffMcp < Formula
  desc "Fast file search toolkit for AI agents (MCP server)"
-  homepage "https://github.com/dmtrKovalenko/fff.nvim"
+  homepage "https://github.com/dmtrKovalenko/fff"
  license "MIT"
-  version "0.10.1"
+  version "0.10.4"

-  LIVECHECK_REPO = "dmtrKovalenko/fff.nvim".freeze
-  RELEASE_BASE = "https://github.com/dmtrKovalenko/fff.nvim/releases/download".freeze
+  LIVECHECK_REPO = "dmtrKovalenko/fff".freeze
+  RELEASE_BASE = "https://github.com/dmtrKovalenko/fff/releases/download".freeze

  on_macos do
    on_arm do
      url "#{RELEASE_BASE}/v#{version}/fff-mcp-aarch64-apple-darwin"
-      sha256 "eff6660a9c48e3e1972d557c1003e057e5ff99d6039f5f810671f2123093fdfc"
+      sha256 "f55ce2945cd30fd252a4b1e7d8674ff7ce41b104cc009a8d3d52d08bfccca2c0"
    end

    on_intel do
      url "#{RELEASE_BASE}/v#{version}/fff-mcp-x86_64-apple-darwin"
-      sha256 "d3c8d70d47ab80afa2287e1b4546b23f494bb4604b80e5a706a0af714ee25674"
+      sha256 "821aedd5873292841f32d7584eb68fea085b248e5c970ff5d4f6dca790e9b18c"
    end
  end

  on_linux do
    on_arm do
      url "#{RELEASE_BASE}/v#{version}/fff-mcp-aarch64-unknown-linux-gnu"
-      sha256 "ed628d57200e48298fcbeffc0f6d85fb572de02bb900bcc6fc7cc05e33ba2777"
+      sha256 "3f266c0eded51a106badf1d47448a7f963762ccf3a12c5a4f06ee2b7adaf82d1"
    end

    on_intel do
      url "#{RELEASE_BASE}/v#{version}/fff-mcp-x86_64-unknown-linux-gnu"
-      sha256 "3541dad4663fc9e5b8b97ce8bd20084ddae08a3c6cb828fcfbcbb6ca8c411513"
+      sha256 "b5dd3182028c8a7bc34ee964cf2d30694f8121d2837a8516cf4a99bcfa822127"
    end
  end

@@ -14,7 +14,7 @@ SHELL := bash
 # string rather than the literal `-o` / `pipefail` tokens.
 .SHELLFLAGS := -o pipefail -euc

-.PHONY: build build-c-lib install uninstall test test-rust test-c-smoke test-c-api test-lua test-lua-snap test-version test-bun test-node prepare-bun prepare-bun-packaged prepare-node set-npm-version header test-stress test-stress-seeded test-stress-random test-stress-regressions test-stress-repos test-node-stress sync-js-api sync-js-api-check bump-homebrew-formula bump-install-mcp-sh test-bun-compile
+.PHONY: build build-c-lib install uninstall test test-rust test-rescan test-rescan-known-defects rescan-probe test-c-smoke test-c-api test-lua test-lua-snap test-version test-bun test-node prepare-bun prepare-bun-packaged prepare-node set-npm-version header test-stress test-stress-seeded test-stress-random test-stress-regressions test-stress-repos test-node-stress sync-js-api sync-js-api-check bump-homebrew-formula bump-install-mcp-sh test-bun-compile

 all: format test lint

@@ -45,6 +45,11 @@ sync-js-api-check:
 build:
 	cargo build --release --no-default-features --features zlob

+# Only the crates the e2e suites load (nvim lua tests + C/bun/node FFI tests),
+# skipping fff-python (pyo3) and fff-mcp (tokio/rmcp) which e2e never touches.
+build-e2e:
+	cargo build --release -p fff-nvim -p fff-c --no-default-features --features zlob
+
 build-c-lib:
 	cargo build --release -p fff-c --no-default-features --features zlob

@@ -90,7 +95,26 @@ test-setup:
 	fi

 test-rust:
-	cargo test --workspace --no-default-features --features zlob --exclude fff-nvim
+	cargo test --workspace --no-default-features --features zlob --exclude fff-nvim --exclude fff-python
+
+# Watcher rescan harness: asserts that editing, build output, git activity and
+# preview reads all stay on the incremental path instead of re-walking the tree.
+test-rescan:
+	cargo test -p fff-search --no-default-features --features zlob \
+		--lib --test rescan_regression -- rescan
+
+# Live probe for watcher rescan requests and their causes.
+# Usage: make rescan-probe DIR=~/some/repo [SECONDS=120]
+rescan-probe:
+	cargo run --release -p fff-nvim --bin rescan_probe \
+		--no-default-features --features zlob,rescan-stats -- \
+		$(or $(DIR),.) $(if $(SECONDS),--seconds $(SECONDS),)
+
+# The same harness, restricted to cases that currently fail on purpose. Each
+# `#[ignore]` reason names the code that causes the unnecessary rescan.
+test-rescan-known-defects:
+	cargo test --no-fail-fast -p fff-search --no-default-features --features zlob \
+		--lib --test rescan_regression -- --ignored --nocapture

 CC ?= cc
 CFLAGS ?= -O0 -g -Wall -Wextra -std=c99
@@ -286,7 +310,7 @@ lint: lint-rust lint-lua lint-ts

 check: format lint

-FFF_RELEASE_REPO ?= dmtrKovalenko/fff.nvim
+FFF_RELEASE_REPO ?= dmtrKovalenko/fff
 FFF_FORMULA_PATH ?= Formula/fff-mcp.rb
 FFF_INSTALL_SCRIPT_PATH ?= install-mcp.sh

@@ -37,7 +37,7 @@ curl -L https://dmtrkovalenko.dev/install-fff-mcp.sh | bash
 Windows (PowerShell):

 ```powershell
-irm https://raw.githubusercontent.com/dmtrKovalenko/fff.nvim/main/install-mcp.ps1 | iex
+irm https://raw.githubusercontent.com/dmtrKovalenko/fff/main/install-mcp.ps1 | iex
 ```

 The scripts live at [`install-mcp.sh`](./install-mcp.sh) and [`install-mcp.ps1`](./install-mcp.ps1) if you want to read them first. They print the exact wiring instructions for your client.
@@ -49,7 +49,7 @@ brew install dmtrKovalenko/fff/fff-mcp
 brew upgrade fff-mcp   # after new stable releases
 ```

-Formula lives in [`Formula/fff-mcp.rb`](./Formula/fff-mcp.rb) in this repo and is **auto-bumped on every stable release** (see `bump-homebrew-formula` in [`.github/workflows/release.yaml`](./.github/workflows/release.yaml)). Installs the prebuilt `fff-mcp` binary from [GitHub releases](https://github.com/dmtrKovalenko/fff.nvim/releases).
+Formula lives in [`Formula/fff-mcp.rb`](./Formula/fff-mcp.rb) in this repo and is **auto-bumped on every stable release** (see `bump-homebrew-formula` in [`.github/workflows/release.yaml`](./.github/workflows/release.yaml)). Installs the prebuilt `fff-mcp` binary from [GitHub releases](https://github.com/dmtrKovalenko/fff/releases).

 ### Codex setup

@@ -153,8 +153,9 @@ https://github.com/user-attachments/assets/5d0e1ce9-642c-4c44-aa88-01b05bb86abb
 #### lazy.nvim

 ```lua
+-- Package name changed from `fff.nvim` to `fff`. If you installed fff.nvim before, clean with `:Lazy clean`
 {
-  'dmtrKovalenko/fff.nvim',
+  'dmtrKovalenko/fff',
  build = function()
    -- downloads a prebuilt binary or falls back to cargo build
    require("fff.download").download_or_build_binary()
@@ -187,13 +188,14 @@ https://github.com/user-attachments/assets/5d0e1ce9-642c-4c44-aa88-01b05bb86abb
 #### vim.pack

 ```lua
-vim.pack.add({ 'https://github.com/dmtrKovalenko/fff.nvim' })
+-- Package name changed from `fff.nvim` to `fff`. If you installed fff.nvim before, clean with `:packdel fff.nvim`
+vim.pack.add({ 'https://github.com/dmtrKovalenko/fff' })

 vim.api.nvim_create_autocmd('PackChanged', {
  callback = function(ev)
    local name, kind = ev.data.spec.name, ev.data.kind
-    if name == 'fff.nvim' and (kind == 'install' or kind == 'update') then
-      if not ev.data.active then vim.cmd.packadd('fff.nvim') end
+    if name == 'fff' and (kind == 'install' or kind == 'update') then
+      if not ev.data.active then vim.cmd.packadd('fff') end
      require('fff.download').download_or_build_binary()
    end
  end,
@@ -614,7 +616,7 @@ cargo build --release -p fff-c --features zlob

 The output is a `cdylib` (`libfff_c.so` / `libfff_c.dylib` / `fff_c.dll`). The header lives at [`crates/fff-c/include/fff.h`](./crates/fff-c/include/fff.h).

-Prebuilt binaries for every version, including every commit on main, are on the [releases page](https://github.com/dmtrKovalenko/fff.nvim/releases). The same binaries also ship inside the `@ff-labs/fff-bin-*` npm packages.
+Prebuilt binaries for every version, including every commit on main, are on the [releases page](https://github.com/dmtrKovalenko/fff/releases). The same binaries also ship inside the `@ff-labs/fff-bin-*` npm packages.

 ### Install

@@ -1,6 +1,6 @@
 [package]
 name = "fff-c"
-version = "0.10.1"
+version = "0.10.4"
 edition = "2024"
 description = "Raw C api of FFF file finder"
 license = "MIT"
@@ -19,6 +19,6 @@ zlob = ["fff/zlob", "fff-query-parser/zlob"]
 [dependencies]
 git2.workspace = true

-fff = { package = "fff-search", path = "../fff-core", version = "0.10.1", default-features = false }
-fff-query-parser = { path = "../fff-query-parser", version = "0.10.1", default-features = false }
+fff = { package = "fff-search", path = "../fff-core", version = "0.10.4", default-features = false }
+fff-query-parser = { path = "../fff-query-parser", version = "0.10.4", default-features = false }
 serde_json = "1.0"
@@ -729,19 +729,12 @@ pub unsafe extern "C" fn fff_multi_grep(
        }
    };

-    let is_ai = picker.mode().is_ai();
-
    // Parse constraints from the optional string (e.g. "*.rs /src/")
-    let parsed_constraints = constraints_str.map(|c| {
-        if is_ai {
-            fff::QueryParser::new(fff_query_parser::AiGrepConfig).parse(c)
-        } else {
-            fff::grep::parse_grep_query(c)
-        }
-    });
+    let parsed_constraints = constraints_str
+        .map(|c| fff::QueryParser::new(fff_query_parser::AiGrepConfig).parse_constraints(c));

    let constraint_refs: &[fff::Constraint<'_>] = match &parsed_constraints {
-        Some(q) => &q.constraints,
+        Some(constraints) => constraints,
        None => &[],
    };

@@ -1,6 +1,6 @@
 [package]
 name = "fff-search"
-version = "0.10.1"
+version = "0.10.4"
 edition = "2024"
 license = "MIT"
 authors = ["Dmitriy Kovalenko <dmtr.kovalenko@outlook.com>"]
@@ -41,6 +41,9 @@ harness = false
 default = ["ripgrep"]
 # Enable C FFI exports
 ffi = []
+# Count full rescans and their causes. Always on in debug builds; enable this
+# to keep the accounting in a release build (used by the rescan_probe binary).
+rescan-stats = []
 # Enables POC definition classification for grep result matched lines
 definitions = []
 # Pure-Rust filesystem walker + glob matcher (ignore + globset crates).
@@ -3,6 +3,15 @@ fn main() {
    // used by tests/fuzz_git_watcher_stress.rs
    println!("cargo::rustc-check-cfg=cfg(stress)");

+    // Full-rescan accounting. Debug builds get it for free; a release build has
+    // to opt in with `--features rescan-stats` (what the rescan_probe needs).
+    println!("cargo::rustc-check-cfg=cfg(rescan_stats)");
+    if std::env::var("DEBUG").is_ok_and(|debug| debug != "false")
+        || std::env::var("CARGO_FEATURE_RESCAN_STATS").is_ok()
+    {
+        println!("cargo::rustc-cfg=rescan_stats");
+    }
+
    // When the `zlob` feature is enabled (Zig-compiled C library):
    // On Windows MSVC, explicitly link the C runtime libraries.
    // Zig-compiled static libraries don't emit /DEFAULTLIB directives for the
@@ -12,16 +12,25 @@ pub const MAX_INDEXABLE_FILE_SIZE: usize = 2 * 1024 * 1024;
 pub const MAX_CACHED_CONTENT_BYTES: u64 = 512 * 1024 * 1024;

 /// Files below one page waste the remainder when mmapped, so the cache skips
-/// them and falls back to chunked reads. Unused on Windows (no content cache).
+/// them and falls back to chunked reads. Unused on Windows (no content cache)
 #[cfg(all(not(target_os = "windows"), target_arch = "aarch64"))]
 pub const MMAP_THRESHOLD: u64 = 16 * 1024;
 #[cfg(all(not(target_os = "windows"), not(target_arch = "aarch64")))]
 pub const MMAP_THRESHOLD: u64 = 4 * 1024;

-/// Capacity reserved for files the watcher discovers after the initial scan;
-/// exceeding it forces a full rescan.
+/// Watcher overflow capacity reserved after the initial scan
 pub const MAX_OVERFLOW_FILES: usize = 1024;

+/// Minimum delay between watcher-initiated rescans.
+pub const RESCAN_MIN_INTERVAL: std::time::Duration = std::time::Duration::from_secs(30);
+
+/// Rescan delay for large indexes.
+pub const RESCAN_MIN_INTERVAL_LARGE_INDEX: std::time::Duration =
+    std::time::Duration::from_secs(5 * 60);
+
+/// Live-file count at which [`RESCAN_MIN_INTERVAL_LARGE_INDEX`] takes over.
+pub const LARGE_INDEX_FILE_COUNT: usize = 1_000_000;
+
 /// Fresh-mmap threshold: files at or above this size get mmapped directly on
 /// cache miss instead of chunked reads into Vec. Empirically tuned per-platform.
 /// Only referenced on Unix; Windows uses the `std::fs::read` fallback so this
@@ -0,0 +1,250 @@
+use heed::{Env, EnvOpenOptions};
+use std::collections::HashMap;
+use std::fs;
+use std::ops::Deref;
+use std::path::{Path, PathBuf};
+use std::sync::atomic::{AtomicBool, Ordering};
+use std::sync::{Arc, LazyLock, Mutex, MutexGuard, PoisonError, Weak};
+use std::thread;
+use std::time::Duration;
+
+use crate::error::{Error, Result};
+use crate::lmdb::DbHealth;
+
+pub(crate) struct EnvSpec {
+    pub label: &'static str,
+    pub map_size: usize,
+    pub max_dbs: u32,
+    pub size_cap_bytes: u64,
+}
+
+pub(crate) struct PooledEnv {
+    env: Env,
+    key: PathBuf,
+    /// lmdb's env spec label
+    label: &'static str,
+    map_size: usize,
+    max_dbs: u32,
+    health: DbHealth,
+    gc_started: AtomicBool,
+    dbi_lock: Mutex<()>,
+}
+
+impl Drop for PooledEnv {
+    fn drop(&mut self) {
+        let mut pool = POOL.lock().unwrap_or_else(PoisonError::into_inner);
+        // Only remove a dead entry: begin_exclusive_destroy may have removed ours.
+        if pool.get(&self.key).is_some_and(|w| w.strong_count() == 0) {
+            pool.remove(&self.key);
+        }
+        // heed closes the env right after this body; a concurrent reopen of the
+        // same path rides out that gap via env_closing_event in get_or_open.
+    }
+}
+
+// Cloneable handle to a process-shared LMDB env, derefs to `heed::Env`.
+#[derive(Clone)]
+pub(crate) struct SharedEnv(Arc<PooledEnv>);
+
+impl Deref for SharedEnv {
+    type Target = Env;
+    fn deref(&self) -> &Env {
+        &self.0.env
+    }
+}
+
+impl std::fmt::Debug for SharedEnv {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_tuple("SharedEnv").field(&self.0.env).finish()
+    }
+}
+
+impl SharedEnv {
+    pub(crate) fn get_or_open(db_path: &Path, spec: &EnvSpec) -> Result<Self> {
+        fs::create_dir_all(db_path).map_err(Error::CreateDir)?;
+        let path = fs::canonicalize(db_path).map_err(|e| Error::EnvOpen {
+            db: spec.label,
+            source: heed::Error::Io(e),
+        })?;
+
+        let mut close_waits = 0u32;
+        let mut transient_retries = 0u32;
+
+        loop {
+            let mut open_failed = false;
+
+            {
+                let mut pool = POOL.lock().unwrap_or_else(PoisonError::into_inner);
+                if let Some(existing) = pool.get(&path).and_then(Weak::upgrade) {
+                    drop(pool);
+                    if existing.label != spec.label
+                        || existing.map_size != spec.map_size
+                        || existing.max_dbs != spec.max_dbs
+                    {
+                        return Err(Error::EnvSpecMismatch {
+                            path,
+                            open_as: existing.label,
+                            requested_as: spec.label,
+                        });
+                    }
+                    return Ok(Self(existing));
+                }
+
+                erase_if_oversized(&path, spec);
+                let result = unsafe {
+                    let mut opts = EnvOpenOptions::new();
+                    opts.map_size(spec.map_size);
+                    if spec.max_dbs > 0 {
+                        opts.max_dbs(spec.max_dbs);
+                    }
+                    opts.open(&path)
+                };
+
+                match result {
+                    Ok(env) => {
+                        let entry = Arc::new(PooledEnv {
+                            env,
+                            key: path.clone(),
+                            label: spec.label,
+                            map_size: spec.map_size,
+                            max_dbs: spec.max_dbs,
+                            health: DbHealth::new(),
+                            gc_started: AtomicBool::new(false),
+                            dbi_lock: Mutex::new(()),
+                        });
+                        pool.insert(path.clone(), Arc::downgrade(&entry));
+                        drop(pool);
+                        let shared = Self(entry);
+
+                        match shared.clear_stale_readers() {
+                            Ok(cleared_count) if cleared_count > 0 => {
+                                tracing::info!(
+                                    cleared_count,
+                                    db = spec.label,
+                                    "reclaimed stale LMDB reader slots at open"
+                                );
+                            }
+                            Ok(_) => {}
+                            Err(e) => {
+                                tracing::debug!("clear_stale_readers at open failed: {e}")
+                            }
+                        }
+
+                        return Ok(shared);
+                    }
+                    Err(heed::Error::EnvAlreadyOpened) => open_failed = true,
+                    // special handling cause we know this happens randomly
+                    Err(e)
+                        if is_transient_env_open_error(&e)
+                            && transient_retries < MAX_TRANSIENT_RETRIES =>
+                    {
+                        transient_retries += 1;
+                        tracing::debug!(
+                            path = %path.display(),
+                            transient_retries,
+                            error = ?e,
+                            "transient LMDB env open error, retrying"
+                        );
+                    }
+                    Err(e) => {
+                        return Err(Error::EnvOpen {
+                            db: spec.label,
+                            source: e,
+                        });
+                    }
+                }
+            }
+
+            if open_failed {
+                close_waits += 1;
+                if close_waits > MAX_CLOSE_WAITS {
+                    return Err(Error::EnvOpen {
+                        db: spec.label,
+                        source: heed::Error::EnvAlreadyOpened,
+                    });
+                }
+
+                match heed::env_closing_event(&path) {
+                    Some(event) => {
+                        event.wait_timeout(CLOSE_WAIT);
+                    }
+                    None => thread::sleep(Duration::from_millis(2)),
+                }
+            } else {
+                thread::sleep(TRANSIENT_RETRY_SLEEP);
+            }
+        }
+    }
+
+    pub(crate) fn health(&self) -> &DbHealth {
+        &self.0.health
+    }
+
+    // First caller wins: GC runs once per opened env, not once per tracker.
+    pub(crate) fn try_start_gc(&self) -> bool {
+        !self.0.gc_started.swap(true, Ordering::AcqRel)
+    }
+
+    // LMDB forbids mdb_dbi_open from concurrent txns in the same process.
+    pub(crate) fn lock_dbi_open(&self) -> MutexGuard<'_, ()> {
+        self.0
+            .dbi_lock
+            .lock()
+            .unwrap_or_else(PoisonError::into_inner)
+    }
+
+    pub(crate) fn destroy(&self) -> Result<Option<heed::EnvClosingEvent>> {
+        let mut pool = POOL.lock().unwrap_or_else(PoisonError::into_inner);
+        let holders = Arc::strong_count(&self.0);
+
+        if holders > 1 {
+            return Err(Error::DbInUse {
+                db: self.0.label,
+                path: self.0.key.clone(),
+                holders: holders - 1,
+            });
+        }
+
+        pool.remove(&self.0.key);
+        Ok(heed::env_closing_event(&self.0.key))
+    }
+}
+
+static POOL: LazyLock<Mutex<HashMap<PathBuf, Weak<PooledEnv>>>> = LazyLock::new(Mutex::default);
+
+const CLOSE_WAIT: Duration = Duration::from_millis(100);
+const MAX_CLOSE_WAITS: u32 = 100;
+const TRANSIENT_RETRY_SLEEP: Duration = Duration::from_millis(50);
+const MAX_TRANSIENT_RETRIES: u32 = 8;
+
+// Concurrent mdb_env_open calls on the same path can race on macOS
+// this is for some reason fixable by simple retry of the open
+fn is_transient_env_open_error(err: &heed::Error) -> bool {
+    match err {
+        heed::Error::Io(io) => matches!(
+            io.kind(),
+            std::io::ErrorKind::InvalidInput | std::io::ErrorKind::NotFound
+        ),
+        _ => false,
+    }
+}
+
+fn erase_if_oversized(db_path: &Path, spec: &EnvSpec) {
+    let data = db_path.join("data.mdb");
+    let Ok(meta) = fs::metadata(&data) else {
+        return;
+    };
+
+    if meta.len() <= spec.size_cap_bytes {
+        return;
+    }
+
+    tracing::error!(
+        path = %db_path.display(),
+        size = meta.len(),
+        cap = spec.size_cap_bytes,
+        "LMDB db exceeds size cap, erasing"
+    );
+    let _ = fs::remove_file(&data);
+    let _ = fs::remove_file(db_path.join("lock.mdb"));
+}
@@ -1,10 +1,11 @@
 use super::db_healthcheck::DbHealthChecker;
-use super::lmdb::{DbHealth, LmdbStore, is_map_full};
+use super::env_pool::SharedEnv;
 use crate::error::{Error, Result};
 use crate::file_picker::FFFMode;
 use crate::git::is_modified_status;
+use crate::lmdb::{DbHealth, LmdbStore, is_map_full};
+use heed::Database;
 use heed::types::{Bytes, SerdeBincode};
-use heed::{Database, Env};
 use std::time::{SystemTime, UNIX_EPOCH};
 use std::{collections::VecDeque, path::Path};

@@ -19,7 +20,7 @@ const AI_MAX_HISTORY_DAYS: f64 = 7.0; // Only consider accesses within 7 days

 #[derive(Debug)]
 pub struct FrecencyTracker {
-    env: Env,
+    env: SharedEnv,
    db: Database<Bytes, SerdeBincode<VecDeque<u64>>>,
    health: DbHealth,
 }
@@ -77,7 +78,7 @@ impl LmdbStore for FrecencyTracker {
    // MAP_SIZE so we don't hit MDB_MAP_FULL before the open-time erase fires.
    const SIZE_CAP_BYTES: u64 = 12 * 1024 * 1024;

-    fn env(&self) -> &Env {
+    fn shared_env(&self) -> &SharedEnv {
        &self.env
    }

@@ -85,7 +86,7 @@ impl LmdbStore for FrecencyTracker {
        &self.health
    }

-    fn purge_stale_data(env: &Env) -> Result<()> {
+    fn purge_stale_data(env: &SharedEnv) -> Result<()> {
        let (deleted, pruned) = Self::purge_stale_entries(env)?;
        if deleted > 0 || pruned > 0 {
            tracing::info!(deleted, pruned, "Frecency GC purged entries");
@@ -121,7 +122,7 @@ impl FrecencyTracker {
    /// Removes entries where all timestamps are older than MAX_HISTORY_DAYS,
    /// and prunes stale timestamps from entries that still have recent ones.
    /// Returns (deleted_count, pruned_count).
-    fn purge_stale_entries(env: &Env) -> Result<(usize, usize)> {
+    fn purge_stale_entries(env: &SharedEnv) -> Result<(usize, usize)> {
        let now = SystemTime::now()
            .duration_since(UNIX_EPOCH)
            .unwrap()
@@ -1,12 +1,11 @@
-use heed::{Database, Env, EnvOpenOptions};
-use std::fs;
+use heed::{Database, Env};
 use std::path::Path;
 use std::sync::Arc;
 use std::sync::RwLock;
 use std::sync::atomic::{AtomicU8, Ordering};
 use std::thread;
-use std::time::Duration;

+use super::env_pool::{EnvSpec, SharedEnv};
 use crate::error::{Error, Result};

 pub(crate) fn is_map_full(err: &heed::Error) -> bool {
@@ -85,9 +84,13 @@ pub(crate) fn spawn_lmdb_gc<T: LmdbStore>(shared: Arc<RwLock<Option<T>>>) {
            let Some(ref tracker) = *guard else {
                return; // destroyed before we started
            };
-            let env = tracker.env();
+            // Trackers attaching to an already-pooled env must not repeat the
+            // GC; the first opener's run flips the shared health flag.
+            if !tracker.shared_env().try_start_gc() {
+                return;
+            }

-            if let Err(e) = T::purge_stale_data(env) {
+            if let Err(e) = T::purge_stale_data(tracker.shared_env()) {
                tracing::debug!("purge_stale_data failed: {e}");
            }

@@ -105,18 +108,6 @@ pub(crate) fn spawn_lmdb_gc<T: LmdbStore>(shared: Arc<RwLock<Option<T>>>) {
    }
 }

-// Concurrent `mdb_env_open` calls on the same path can race on macOS
-// this is for some reason fixabtly by simple retry of the open
-fn is_transient_env_open_error(err: &heed::Error) -> bool {
-    match err {
-        heed::Error::Io(io) => matches!(
-            io.kind(),
-            std::io::ErrorKind::InvalidInput | std::io::ErrorKind::NotFound
-        ),
-        _ => false,
-    }
-}
-
 pub(crate) trait LmdbStore: Sized + Send + Sync + 'static {
    /// Short label used to defferintiate different instances of this trait
    const LABEL: &'static str;
@@ -127,85 +118,51 @@ pub(crate) trait LmdbStore: Sized + Send + Sync + 'static {
    /// Hard cap on `data.mdb` size.
    const SIZE_CAP_BYTES: u64;

-    /// Borrow the env in the read lock
-    fn env(&self) -> &Env;
+    /// Borrow the pooled env handle shared by every tracker of this path.
+    fn shared_env(&self) -> &SharedEnv;
    /// Borrow the health flag from the tracker.
    fn health(&self) -> &DbHealth;

+    /// Borrow the raw heed env.
+    fn env(&self) -> &Env {
+        self.shared_env()
+    }
+
    /// Override to purge stale rows, compact, etc. Default no-op. Runs on
    /// the GC thread while a read lock is held against the shared handle,
    /// so destroy / re-init naturally wait for it.
-    fn purge_stale_data(_env: &Env) -> Result<()> {
+    fn purge_stale_data(_env: &SharedEnv) -> Result<()> {
        Ok(())
    }

-    /// Open the LMDB env. Returns env + a `DbHealth` starting in Pending;
-    /// the GC thread spawned by `spawn_gc` flips it to Healthy. Write
-    /// paths flip it to Degraded on MDB_MAP_FULL.
+    /// Open (or join) the process-shared LMDB env for `db_path`. The health
+    /// flag is per-env: the GC of the first opener flips it for everyone.
    #[tracing::instrument]
-    fn open_env(db_path: &Path) -> Result<(Env, DbHealth)> {
-        Self::erase_if_oversized(db_path);
-        fs::create_dir_all(db_path).map_err(Error::CreateDir)?;
-        let db = Self::LABEL;
-
-        const MAX_ATTEMPTS: u32 = 8;
-        let mut attempt = 0u32;
-        let env = loop {
-            let result = unsafe {
-                let mut opts = EnvOpenOptions::new();
-                opts.map_size(Self::MAP_SIZE);
-                if Self::MAX_DBS > 0 {
-                    opts.max_dbs(Self::MAX_DBS);
-                }
-                opts.open(db_path)
-            };
-
-            match result {
-                Ok(env) => break env,
-                Err(e) if is_transient_env_open_error(&e) && attempt + 1 < MAX_ATTEMPTS => {
-                    attempt += 1;
-                    tracing::debug!(
-                        path = %db_path.display(),
-                        attempt,
-                        error = ?e,
-                        "transient LMDB env open error, retrying"
-                    );
-
-                    thread::sleep(Duration::from_millis(50));
-                }
-                Err(e) => return Err(Error::EnvOpen { db, source: e }),
-            }
-        };
-
-        // Reclaim reader slots left behind by prior processes that died
-        // without cleanup. Must run before we start any read txns (which
-        // open_database_safe does) — otherwise we may hit MDB_READERS_FULL
-        // on a fresh env just because lock.mdb still has stale entries
-        // from a previous crash.
-        //
-        // This is the one LMDB maintenance call we run on the caller's
-        // thread. If the lock file is genuinely wedged this will block
-        // forever, but the alternative — never getting past init — is
-        // worse and the bg-thread trick doesn't solve it anyway.
-        match env.clear_stale_readers() {
-            Ok(cleared) if cleared > 0 => {
-                tracing::warn!(cleared, "reclaimed stale LMDB reader slots at open");
-            }
-            Ok(_) => {}
-            Err(e) => tracing::debug!("clear_stale_readers at open failed: {e}"),
-        }
-
-        Ok((env, DbHealth::new()))
+    fn open_env(db_path: &Path) -> Result<(SharedEnv, DbHealth)> {
+        let shared = SharedEnv::get_or_open(
+            db_path,
+            &EnvSpec {
+                label: Self::LABEL,
+                map_size: Self::MAP_SIZE,
+                max_dbs: Self::MAX_DBS,
+                size_cap_bytes: Self::SIZE_CAP_BYTES,
+            },
+        )?;
+        let health = shared.health().clone();
+        Ok((shared, health))
    }

    /// Open or create a database without blocking on the LMDB writer mutex
    /// when the database already exists.
-    fn open_database_safe<KC, DC>(env: &Env, name: Option<&str>) -> Result<Database<KC, DC>>
+    fn open_database_safe<KC, DC>(env: &SharedEnv, name: Option<&str>) -> Result<Database<KC, DC>>
    where
        KC: 'static,
        DC: 'static,
    {
        let db = Self::LABEL;
+        // mdb_dbi_open must not run from concurrent txns in this process.
+        let _dbi_guard = env.lock_dbi_open();
+
        let rtxn = env
            .read_txn()
            .map_err(|source| Error::DbStartReadTxn { db, source })?;
@@ -237,23 +194,4 @@ pub(crate) trait LmdbStore: Sized + Send + Sync + 'static {
            }
        }
    }
-
-    fn erase_if_oversized(db_path: &Path) {
-        let data = db_path.join("data.mdb");
-        let Ok(meta) = fs::metadata(&data) else {
-            return;
-        };
-        if meta.len() <= Self::SIZE_CAP_BYTES {
-            return;
-        }
-
-        tracing::error!(
-            path = %db_path.display(),
-            size = meta.len(),
-            cap = Self::SIZE_CAP_BYTES,
-            "LMDB db exceeds size cap, erasing"
-        );
-        let _ = fs::remove_file(&data);
-        let _ = fs::remove_file(db_path.join("lock.mdb"));
-    }
 }
@@ -1,3 +1,4 @@
+pub(crate) mod env_pool;
 pub(crate) mod lmdb;

 pub mod db_healthcheck;
@@ -1,6 +1,7 @@
 use super::db_healthcheck::DbHealthChecker;
-use super::lmdb::{DbHealth, LmdbStore, is_map_full};
+use super::env_pool::SharedEnv;
 use crate::error::Error;
+use crate::lmdb::{DbHealth, LmdbStore, is_map_full};
 use heed::types::{Bytes, SerdeBincode};
 use heed::{Database, Env};
 use serde::{Deserialize, Serialize};
@@ -27,7 +28,7 @@ struct HistoryEntry {

 #[derive(Debug)]
 pub struct QueryTracker {
-    env: Env,
+    env: SharedEnv,
    // Database for (project_path, query) -> QueryMatchEntry mappings
    query_file_db: Database<Bytes, SerdeBincode<QueryMatchEntry>>,
    // Database for project_path -> VecDeque<HistoryEntry> mappings (file picker)
@@ -92,7 +93,7 @@ impl LmdbStore for QueryTracker {
    const MAX_DBS: u32 = 16;
    const SIZE_CAP_BYTES: u64 = 8 * 1024 * 1024;

-    fn env(&self) -> &Env {
+    fn shared_env(&self) -> &SharedEnv {
        &self.env
    }

@@ -34,6 +34,22 @@ pub enum Error {
        #[source]
        source: heed::Error,
    },
+    #[error(
+        "LMDB env at {path} is already open as the '{open_as}' database with different options; requested by '{requested_as}'. Use a distinct path per database."
+    )]
+    EnvSpecMismatch {
+        path: std::path::PathBuf,
+        open_as: &'static str,
+        requested_as: &'static str,
+    },
+    #[error(
+        "The {db} database at {path} is still used by {holders} other tracker(s) in this process"
+    )]
+    DbInUse {
+        db: &'static str,
+        path: std::path::PathBuf,
+        holders: usize,
+    },
    #[error("Failed to create {db} database: {source}")]
    DbCreate {
        db: &'static str,
@@ -629,6 +629,10 @@ impl FilePicker {
        &self.base_path
    }

+    pub fn has_git_repo(&self) -> bool {
+        self.sync_data.git_workdir.is_some()
+    }
+
    /// Ignore rules the walker assembled during the last scan (zlob backend
    /// only). The background watcher uses these to filter events without
    /// libgit2. `None` when the backend doesn't surface rules or no ignore
@@ -229,9 +229,19 @@ pub(crate) fn grep_search<'a>(
        return result;
    }

+    // Keep any explicit FilePath scope (AI mode `path/to/file.ext` prefix) so the
+    // fallback can't leak matches outside the file the user pinned. Only the
+    // swallowed operator/glob tokens are dropped. See issue #756.
+    let scoped_constraints: fff_query_parser::ConstraintVec<'_> = query
+        .constraints
+        .iter()
+        .filter(|c| matches!(c, fff_query_parser::Constraint::FilePath(_)))
+        .cloned()
+        .collect();
+
    let literal_query = FFFQuery {
        raw_query: query.raw_query,
-        constraints: Vec::new(),
+        constraints: scoped_constraints,
        fuzzy_query: fff_query_parser::FuzzyQuery::Text(raw),
        location: None,
    };
@@ -461,3 +461,58 @@ fn test_grep_no_duplicates_with_overflow_trailing_bits() {
        result.matches.len()
    );
 }
+
+/// Issue #756: an AI-mode regex query with an inline FilePath scope and
+/// top-level alternation. The regex fragments are swallowed as bogus Glob
+/// constraints, the constrained search finds nothing, and the literal/regex
+/// fallback must NOT drop the FilePath scope — otherwise the `|` branch leaks
+/// matches into files outside the pinned path.
+#[test]
+fn regex_fallback_keeps_file_path_scope_issue_756() {
+    use fff_query_parser::{AiGrepConfig, QueryParser};
+    let dir = tempfile::tempdir().unwrap();
+    let base = crate::path_utils::canonicalize(dir.path()).unwrap();
+    std::fs::create_dir(base.join("scope")).unwrap();
+    std::fs::write(
+        base.join("scope").join("target.css"),
+        "/* ---------- target ---------- */\n",
+    )
+    .unwrap();
+    std::fs::write(
+        base.join("outside.css"),
+        "/* ---------- outside ---------- */\n",
+    )
+    .unwrap();
+
+    let mut picker = FilePicker::new(FilePickerOptions {
+        base_path: base.to_str().unwrap().into(),
+        watch: false,
+        ..Default::default()
+    })
+    .unwrap();
+    picker.collect_files().unwrap();
+
+    let options = crate::GrepSearchOptions {
+        mode: super::GrepMode::Regex,
+        smart_case: true,
+        max_matches_per_file: 80,
+        page_limit: 100,
+        ..Default::default()
+    };
+
+    let raw = r"scope/target.css ^/\* |^\s*/\* ----------";
+    let query = QueryParser::new(AiGrepConfig).parse(raw);
+    let result = picker.grep(&query, &options);
+    let mut paths: Vec<String> = result
+        .files
+        .iter()
+        .map(|f| f.relative_path(&picker))
+        .collect();
+    paths.sort();
+
+    assert_eq!(
+        paths,
+        vec!["scope/target.css"],
+        "regex fallback must not leak outside the FilePath scope"
+    );
+}
@@ -1,5 +1,5 @@
 use super::grep::{GrepContext, perform_grep};
-use super::prefilter::prefilter_with_filepath_retry;
+use super::prefilter::prefilter_files;
 use super::sink::{SinkState, debug_assert_newline_terminator};
 use super::types::{GrepResult, GrepSearchOptions};
 use crate::index::{BigramFilter, BigramOverlay, bigram_boundary, literal_candidates};
@@ -121,7 +121,8 @@ pub(crate) fn multi_grep_search<'a>(
    let bigram_candidates = literal_candidates(bigram_index, bigram_overlay, patterns);
    let base_file_count = bigram_boundary(bigram_overlay, files.len());

-    let (files_to_search, filtered_file_count) = prefilter_with_filepath_retry(
+    // Constraints are separate from patterns, so a miss must not broaden the search.
+    let (files_to_search, filtered_file_count) = prefilter_files(
        files,
        constraints,
        bigram_candidates.as_deref(),
@@ -50,7 +50,7 @@ pub(super) fn prefilter_with_filepath_retry<'a>(
 /// Single pass prefilter that doesn't involve file reading
 /// allocates only amount of memory required for storing references of the FileItems have to be
 /// opened for grepping unaviodably, in the worst case allocates N * <word> memory if no prefilter needed
-fn prefilter_files<'a>(
+pub(crate) fn prefilter_files<'a>(
    files: &'a [FileItem],
    constraints: &[Constraint<'_>],
    bigram_candidates: Option<&[u64]>,
@@ -3,28 +3,53 @@ use std::path::Path;
 /// Directories excluded when walking a non-git root. Entries are `cfg`-gated
 /// so a single iteration covers standard + platform-specific overrides.
 pub(crate) const IGNORED_DIRS: &[&str] = &[
+    // various dev tools that can be meet in the developer app
    "node_modules",
    "__pycache__",
    "venv",
    ".venv",
-    // Rust (glob-only patterns for non_git_repo_overrides; is_non_code_directory
-    // matches the "target" component separately).
    "target/debug",
    "target/release",
    "target/rust-analyzer",
    "target/criterion",
+    // Language package caches in non-git roots.
+    "go/pkg/mod",
+    ".cargo/registry",
+    ".rustup/toolchains",
+    ".gradle/caches",
+    ".m2/repository",
+    ".npm/_cacache",
+    ".pub-cache",
+    #[cfg(not(target_os = "windows"))]
+    ".local/state", // this contains tons of logs which generate too much watcher noise
    #[cfg(target_os = "macos")]
    "Library/Application Support",
    #[cfg(target_os = "macos")]
    "Library/Caches",
-    // App-group sandbox storage — used by iMessage, Photos, Notes, Calendar,
-    // Electron apps, etc. for SQLite-WAL, LevelDB, protobuf files. These are
-    // almost entirely extension-less binary files (~80k on a typical $HOME)
-    // that never need to appear in a fuzzy or grep search.
    #[cfg(target_os = "macos")]
-    "Library/Group Containers",
+    "Library/Containers", // sandboxed apps data
    #[cfg(target_os = "macos")]
-    "Library/Containers",
+    "Library/Group Containers", // random application data and networking
+    #[cfg(target_os = "macos")]
+    "Library/pnpm",
+    #[cfg(target_os = "macos")]
+    "Library/Metadata",
+    #[cfg(target_os = "macos")]
+    "Library/Developer/CoreSimulator",
+    #[cfg(target_os = "macos")]
+    "Library/Android",
+    #[cfg(target_os = "macos")]
+    "Library/Logs",
+    #[cfg(target_os = "macos")]
+    "Library/Daemon Containers",
+    #[cfg(target_os = "macos")]
+    "Library/Trial",
+    #[cfg(target_os = "macos")]
+    "Library/Preferences",
+    #[cfg(target_os = "macos")]
+    "Library/Messages",
+    #[cfg(target_os = "macos")]
+    "Library/IdentityServices",
    #[cfg(target_os = "windows")]
    "bin/Debug",
    #[cfg(target_os = "windows")]
@@ -57,6 +82,10 @@ pub(crate) fn non_git_repo_overrides(base_path: &Path) -> Option<ignore::overrid
 pub(crate) fn is_non_code_directory(path: &Path) -> bool {
    let path_str = path.as_os_str().to_str().unwrap_or("");
    IGNORED_DIRS.iter().any(|&dir| {
+        // Entries are gitignore patterns for the walkers; here they are matched
+        // as substrings, so a leading `*` wildcard has to come off first.
+        let dir = dir.strip_prefix('*').unwrap_or(dir);
+
        #[cfg(target_os = "windows")]
        let dir = dir.replace('/', std::path::MAIN_SEPARATOR_STR);
        #[cfg(target_os = "windows")]
@@ -66,3 +95,42 @@ pub(crate) fn is_non_code_directory(path: &Path) -> bool {
        path_str.contains(dir)
    })
 }
+
+#[cfg(all(test, target_os = "macos"))]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn home_machine_state_is_excluded_but_source_trees_are_not() {
+        // Representative machine state from a home index.
+        for rel in [
+            "Library/pnpm/store/v3/files/00/abcdef",
+            "Library/Preferences/com.apple.finder.plist",
+            "Library/Messages/prewarm.db-shm",
+            "Library/IdentityServices/TetraDB-identityservicesd.db-wal",
+            "Library/Developer/CoreSimulator/Devices/X/data/f",
+            "go/pkg/mod/github.com/x/y@v1/main.go",
+            ".cargo/registry/src/index.crates.io-1/serde-1.0/src/lib.rs",
+            "Library/Android/sdk/platforms/android-34/data/x",
+            ".local/state/nvim/fff+123+456.log",
+        ] {
+            assert!(
+                is_non_code_directory(Path::new(rel)),
+                "{rel} must not reach the index"
+            );
+        }
+
+        // Source trees under $HOME stay searchable.
+        for rel in [
+            "dev/chromium/third_party/blink/renderer/core/dom/node.cc",
+            "dev/fff/crates/fff-core/src/lib.rs",
+            "Documents/notes/todo.md",
+            "dev/myproj/pkg/mod/thing.go",
+        ] {
+            assert!(
+                !is_non_code_directory(Path::new(rel)),
+                "{rel} must stay searchable"
+            );
+        }
+    }
+}
@@ -1,6 +1,6 @@
 //! # FFF Search — High-performance file finder core
 //!
-//! This crate provides the core search engine for [FFF (Fast File Finder)](https://github.com/dmtrKovalenko/fff.nvim).
+//! This crate provides the core search engine for [FFF (Fast File Finder)](https://github.com/dmtrKovalenko/fff).
 //! It includes filesystem indexing with real-time watching, fuzzy matching powered
 //! by [frizbee](https://docs.rs/neo_frizbee), frecency scoring backed by LMDB,
 //! and multi-mode grep search.
@@ -135,6 +135,12 @@ pub use types::*;

 pub mod constants;

+/// Watcher rescan request accounting.
+pub mod rescan_stats;
+pub use rescan_stats::{RESCAN_STATS_ENABLED, RescanReason, RescanStats};
+
+mod rescan_throttle;
+
 // ==================================
 // these are public only for benchmarks, no backward compatibility guaranteed
 #[doc(hidden)]
@@ -0,0 +1,215 @@
+#[cfg(rescan_stats)]
+use std::sync::atomic::{AtomicUsize, Ordering};
+
+/// Whether rescan accounting is compiled in.
+pub const RESCAN_STATS_ENABLED: bool = cfg!(rescan_stats);
+
+/// Cause recorded for a filesystem rescan request.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub enum RescanReason {
+    /// Requested through the public API (refresh, directory change).
+    Explicit,
+    /// The kernel dropped events and asked us to re-read the subtree.
+    KernelEventLoss,
+    /// A `.gitignore`/`.ignore` changed, so the cached ignore rules are stale.
+    IgnoreFileChanged,
+    /// A single debounce batch touched more paths than we apply incrementally.
+    EventBatchOverflow,
+    /// The picker refused an incremental insert/update.
+    IndexUpdateRejected,
+    /// The post-scan overflow region ran out of slots.
+    OverflowCapacity,
+}
+
+impl RescanReason {
+    pub const ALL: [RescanReason; 6] = [
+        RescanReason::Explicit,
+        RescanReason::KernelEventLoss,
+        RescanReason::IgnoreFileChanged,
+        RescanReason::EventBatchOverflow,
+        RescanReason::IndexUpdateRejected,
+        RescanReason::OverflowCapacity,
+    ];
+
+    pub const fn as_str(self) -> &'static str {
+        match self {
+            RescanReason::Explicit => "explicit",
+            RescanReason::KernelEventLoss => "kernel_event_loss",
+            RescanReason::IgnoreFileChanged => "ignore_file_changed",
+            RescanReason::EventBatchOverflow => "event_batch_overflow",
+            RescanReason::IndexUpdateRejected => "index_update_rejected",
+            RescanReason::OverflowCapacity => "overflow_capacity",
+        }
+    }
+
+    const fn slot(self) -> usize {
+        match self {
+            RescanReason::Explicit => 0,
+            RescanReason::KernelEventLoss => 1,
+            RescanReason::IgnoreFileChanged => 2,
+            RescanReason::EventBatchOverflow => 3,
+            RescanReason::IndexUpdateRejected => 4,
+            RescanReason::OverflowCapacity => 5,
+        }
+    }
+}
+
+impl std::fmt::Display for RescanReason {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.write_str(self.as_str())
+    }
+}
+
+/// Snapshot of rescan requests grouped by reason.
+#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
+pub struct RescanStats {
+    pub total: usize,
+    /// Requests suppressed during the cooldown.
+    pub throttled: usize,
+    counts: [usize; RescanReason::ALL.len()],
+    throttled_counts: [usize; RescanReason::ALL.len()],
+}
+
+impl RescanStats {
+    pub fn count(&self, reason: RescanReason) -> usize {
+        self.counts[reason.slot()]
+    }
+
+    pub fn count_throttled(&self, reason: RescanReason) -> usize {
+        self.throttled_counts[reason.slot()]
+    }
+
+    /// Admitted requests originating from watcher fallbacks.
+    pub fn watcher_triggered(&self) -> usize {
+        self.total - self.count(RescanReason::Explicit)
+    }
+
+    /// Per-reason delta against an earlier snapshot.
+    pub fn since(&self, earlier: &RescanStats) -> RescanStats {
+        let mut counts = [0usize; RescanReason::ALL.len()];
+        let mut throttled_counts = [0usize; RescanReason::ALL.len()];
+        for slot in 0..RescanReason::ALL.len() {
+            counts[slot] = self.counts[slot].saturating_sub(earlier.counts[slot]);
+            throttled_counts[slot] =
+                self.throttled_counts[slot].saturating_sub(earlier.throttled_counts[slot]);
+        }
+
+        RescanStats {
+            total: self.total.saturating_sub(earlier.total),
+            throttled: self.throttled.saturating_sub(earlier.throttled),
+            counts,
+            throttled_counts,
+        }
+    }
+}
+
+impl std::fmt::Display for RescanStats {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{} rescan(s)", self.total)?;
+        let mut first = true;
+        for reason in RescanReason::ALL {
+            let count = self.count(reason);
+            if count == 0 {
+                continue;
+            }
+            f.write_str(if first { " [" } else { ", " })?;
+            write!(f, "{reason}={count}")?;
+            first = false;
+        }
+        if !first {
+            f.write_str("]")?;
+        }
+        if self.throttled > 0 {
+            write!(f, ", {} throttled", self.throttled)?;
+        }
+        Ok(())
+    }
+}
+
+#[cfg(rescan_stats)]
+#[derive(Default)]
+pub(crate) struct RescanCounters {
+    counters: [AtomicUsize; RescanReason::ALL.len()],
+    throttled: [AtomicUsize; RescanReason::ALL.len()],
+}
+
+#[cfg(rescan_stats)]
+impl RescanCounters {
+    pub(crate) fn record(&self, reason: RescanReason) {
+        self.counters[reason.slot()].fetch_add(1, Ordering::Relaxed);
+    }
+
+    pub(crate) fn record_throttled(&self, reason: RescanReason) {
+        self.throttled[reason.slot()].fetch_add(1, Ordering::Relaxed);
+    }
+
+    pub(crate) fn snapshot(&self) -> RescanStats {
+        let mut stats = RescanStats::default();
+        for reason in RescanReason::ALL {
+            let count = self.counters[reason.slot()].load(Ordering::Relaxed);
+            stats.counts[reason.slot()] = count;
+            stats.total += count;
+
+            let throttled = self.throttled[reason.slot()].load(Ordering::Relaxed);
+            stats.throttled_counts[reason.slot()] = throttled;
+            stats.throttled += throttled;
+        }
+        stats
+    }
+
+    pub(crate) fn reset(&self) {
+        for counter in self.counters.iter().chain(self.throttled.iter()) {
+            counter.store(0, Ordering::Relaxed);
+        }
+    }
+}
+
+// Release builds retain the API without counter storage.
+#[cfg(not(rescan_stats))]
+#[derive(Default)]
+pub(crate) struct RescanCounters;
+
+#[cfg(not(rescan_stats))]
+impl RescanCounters {
+    pub(crate) fn record(&self, _reason: RescanReason) {}
+
+    pub(crate) fn record_throttled(&self, _reason: RescanReason) {}
+
+    pub(crate) fn snapshot(&self) -> RescanStats {
+        RescanStats::default()
+    }
+
+    pub(crate) fn reset(&self) {}
+}
+
+#[cfg(all(test, rescan_stats))]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn counters_attribute_and_diff_per_reason() {
+        let counters = RescanCounters::default();
+        counters.record(RescanReason::Explicit);
+        let baseline = counters.snapshot();
+
+        counters.record(RescanReason::IgnoreFileChanged);
+        counters.record(RescanReason::IgnoreFileChanged);
+        counters.record(RescanReason::OverflowCapacity);
+
+        let stats = counters.snapshot();
+        assert_eq!(stats.total, 4);
+        assert_eq!(stats.watcher_triggered(), 3);
+
+        let delta = stats.since(&baseline);
+        assert_eq!(delta.total, 3);
+        assert_eq!(delta.count(RescanReason::Explicit), 0);
+        assert_eq!(delta.count(RescanReason::IgnoreFileChanged), 2);
+        assert_eq!(
+            delta.to_string(),
+            "3 rescan(s) [ignore_file_changed=2, overflow_capacity=1]"
+        );
+
+        counters.reset();
+        assert_eq!(counters.snapshot(), RescanStats::default());
+    }
+}
@@ -0,0 +1,124 @@
+use std::sync::atomic::{AtomicU64, Ordering};
+use std::time::Instant;
+
+use crate::constants::{
+    LARGE_INDEX_FILE_COUNT, RESCAN_MIN_INTERVAL, RESCAN_MIN_INTERVAL_LARGE_INDEX,
+};
+
+const NEVER: u64 = u64::MAX;
+
+// Drops watcher rescan requests inside the cooldown after the last scan.
+// A slightly stale index is fine: the next admitted event rescans everything.
+pub(crate) struct RescanThrottle {
+    epoch: Instant,
+    last_admitted: AtomicU64,
+}
+
+impl Default for RescanThrottle {
+    fn default() -> Self {
+        Self {
+            epoch: Instant::now(),
+            last_admitted: AtomicU64::new(NEVER),
+        }
+    }
+}
+
+impl RescanThrottle {
+    /// Returns `true` if a rescan may start now and records it as the last scan
+    pub(crate) fn admit(&self, live_files: usize, has_git_repo: bool) -> bool {
+        let min_interval = if !has_git_repo && live_files >= LARGE_INDEX_FILE_COUNT {
+            RESCAN_MIN_INTERVAL_LARGE_INDEX
+        } else {
+            RESCAN_MIN_INTERVAL
+        };
+
+        let min_ms = min_interval.as_millis() as u64;
+        let now = self.elapsed_ms();
+
+        loop {
+            let last = self.last_admitted.load(Ordering::Acquire);
+            if last != NEVER && now.saturating_sub(last) < min_ms {
+                return false;
+            }
+            // CAS so two concurrent requests cannot both start a walk.
+            if self
+                .last_admitted
+                .compare_exchange(last, now, Ordering::AcqRel, Ordering::Acquire)
+                .is_ok()
+            {
+                return true;
+            }
+        }
+    }
+
+    /// Records an explicit (unthrottled) scan so watcher requests right after
+    /// it are dropped: the index is already fresh.
+    pub(crate) fn note_explicit_scan(&self) {
+        self.last_admitted
+            .store(self.elapsed_ms(), Ordering::Release);
+    }
+
+    fn elapsed_ms(&self) -> u64 {
+        self.epoch.elapsed().as_millis() as u64
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::time::Duration;
+
+    fn throttle_at(ms_ago: u64) -> RescanThrottle {
+        let now = Instant::now();
+        RescanThrottle {
+            epoch: now
+                .checked_sub(Duration::from_millis(ms_ago))
+                .expect("monotonic clock older than the rewind"),
+            last_admitted: AtomicU64::new(0),
+        }
+    }
+
+    #[test]
+    fn first_request_is_always_admitted() {
+        let throttle = RescanThrottle::default();
+        assert!(throttle.admit(100, true));
+    }
+
+    #[test]
+    fn requests_inside_the_cooldown_are_dropped() {
+        let throttle = throttle_at(1_000);
+        assert!(!throttle.admit(100, false));
+        assert!(!throttle.admit(100, false));
+    }
+
+    #[test]
+    fn a_large_index_outside_a_git_repo_uses_the_slower_cadence() {
+        // A minute is past the normal cooldown but not the large-index one.
+        let throttle = throttle_at(60_000);
+        assert!(throttle.admit(100, false));
+
+        let throttle = throttle_at(60_000);
+        assert!(!throttle.admit(LARGE_INDEX_FILE_COUNT, false));
+    }
+
+    #[test]
+    fn a_git_repo_keeps_the_normal_cadence_at_any_size() {
+        let throttle = throttle_at(60_000);
+        assert!(throttle.admit(LARGE_INDEX_FILE_COUNT, true));
+    }
+
+    #[test]
+    fn cooldown_expiry_admits_again() {
+        let throttle = throttle_at(RESCAN_MIN_INTERVAL.as_millis() as u64 + 1);
+        assert!(throttle.admit(100, false));
+        // Admission rearms the cooldown.
+        assert!(!throttle.admit(100, false));
+    }
+
+    #[test]
+    fn explicit_scan_rearms_the_cooldown() {
+        let throttle = RescanThrottle::default();
+        throttle.note_explicit_scan();
+        assert!(!throttle.admit(100, false));
+    }
+}
@@ -8,6 +8,8 @@ use crate::file_picker::FilePicker;
 use crate::frecency::FrecencyTracker;
 use crate::git::GitStatusCache;
 use crate::query_tracker::QueryTracker;
+use crate::rescan_stats::{RescanCounters, RescanReason, RescanStats};
+use crate::rescan_throttle::RescanThrottle;
 use crate::scan::ScanJob;
 use crate::watch::{WatchEvent, WatchId, WatchOptions, WatchRegistry};
 use git2::Repository;
@@ -77,6 +79,8 @@ pub struct SharedPickerInner {
    /// Watch subscriptions live outside the picker lock so delivery and
    /// (un)subscribing never contend with searches.
    watchers: Arc<WatchRegistry>,
+    rescans: RescanCounters,
+    rescan_throttle: RescanThrottle,
 }

 impl Default for SharedPickerInner {
@@ -84,6 +88,8 @@ impl Default for SharedPickerInner {
        Self {
            picker: parking_lot::RwLock::new(None),
            watchers: Arc::new(WatchRegistry::default()),
+            rescans: RescanCounters::default(),
+            rescan_throttle: RescanThrottle::default(),
        }
    }
 }
@@ -199,6 +205,39 @@ impl SharedFilePicker {
    /// Performs a safe async rescan. Guarantees only single active rescan per picker.
    /// If many rescans requested the last one guaranteed to be finished.
    pub fn trigger_full_rescan_async(&self, shared_frecency: &SharedFrecency) -> Result<(), Error> {
+        self.trigger_full_rescan_with_reason(shared_frecency, RescanReason::Explicit)
+            .map(|_| ())
+    }
+
+    /// Returns admitted and throttled rescan requests by reason.
+    /// Counters start at picker creation or the last reset.
+    pub fn rescan_stats(&self) -> RescanStats {
+        self.0.rescans.snapshot()
+    }
+
+    pub fn reset_rescan_stats(&self) {
+        self.0.rescans.reset();
+    }
+
+    /// Returns `Ok(true)` when a rescan was started (or queued behind an
+    /// active scan) and `Ok(false)` when the request was throttled — the
+    /// caller must then fall back to incremental event processing.
+    pub(crate) fn trigger_full_rescan_with_reason(
+        &self,
+        shared_frecency: &SharedFrecency,
+        reason: RescanReason,
+    ) -> Result<bool, Error> {
+        // for giant folders we have no other choice other than throttling rescans
+        // if user is running application in millions of files with a ton of rescan events
+        // we drop / throttle some of requests to avoid constant burst of IO
+        if reason == RescanReason::Explicit {
+            self.0.rescan_throttle.note_explicit_scan();
+        } else if !self.check_rescan_throttle(reason) {
+            return Ok(false);
+        }
+
+        self.0.rescans.record(reason);
+
        match ScanJob::new_rescan(self, shared_frecency)? {
            Some(job) => {
                job.spawn();
@@ -219,7 +258,27 @@ impl SharedFilePicker {
                }
            }
        }
-        Ok(())
+        Ok(true)
+    }
+
+    fn check_rescan_throttle(&self, reason: RescanReason) -> bool {
+        let (live_files, has_git) = self
+            .read()
+            .ok()
+            .and_then(|guard| {
+                guard
+                    .as_ref()
+                    .map(|picker| (picker.live_file_count(), picker.has_git_repo()))
+            })
+            .unwrap_or((0, false));
+
+        if self.0.rescan_throttle.admit(live_files, has_git) {
+            return true;
+        }
+
+        self.0.rescans.record_throttled(reason);
+        tracing::debug!(%reason, live_files, "Rescan throttled, skipping");
+        false
    }

    /// Subscribe to filesystem changes matching `pattern`.
@@ -432,20 +491,31 @@ impl<T: LmdbStore> SharedDb<T> {

    /// Drop the in-memory tracker and delete the on-disk database directory.
    ///
-    /// Acquires the write lock, ensuring all readers (including any active mmap
-    /// access) are finished before the LMDB environment is closed and the files
-    /// are removed.
-    ///
    /// Returns `Ok(Some(path))` with the deleted path, or `Ok(None)` if no tracker was initialized.
    pub fn destroy(&self) -> Result<Option<PathBuf>, Error> {
        let mut guard = self.write()?;
        let Some(tracker) = guard.take() else {
            return Ok(None);
        };
+
+        let closing_event = match tracker.shared_env().destroy() {
+            Ok(closing) => closing,
+            Err(e) => {
+                *guard = Some(tracker);
+                return Err(e);
+            }
+        };
+
        let db_path = tracker.env().path().to_path_buf();
        // Drop closes the LMDB env and unmaps the files
        drop(tracker);
        drop(guard);
+
+        // Deleting before mdb_env_close finishes would race the unmap.
+        if let Some(event) = closing_event {
+            event.wait_timeout(Duration::from_secs(5));
+        }
+
        std::fs::remove_dir_all(&db_path).map_err(|source| Error::RemoveDbDir {
            path: db_path.clone(),
            source,
@@ -415,7 +415,10 @@ mod tests {

    #[test]
    fn test_chunked_string_full_path() {
-        let (store, strings, _files) = build_test_store(&["src/components/Button.tsx"]);
+        let (store, strings, _files) = build_test_store(&[
+            "src/components/Button.tsx",
+            "src/components/Button.test.tsx",
+        ]);
        let arena = store.as_arena_ptr();
        let cs = &strings[0];

@@ -423,6 +426,15 @@ mod tests {
        assert_eq!(cs.read_to_buf(arena, &mut buf), "src/components/Button.tsx");
        assert_eq!(cs.byte_len, 25);
        assert_eq!(cs.filename_offset, 15);
+
+        let cs = &strings[1];
+        let mut buf = [0u8; 512];
+        assert_eq!(
+            cs.read_to_buf(arena, &mut buf),
+            "src/components/Button.test.tsx"
+        );
+        assert_eq!(cs.byte_len, 30);
+        assert_eq!(cs.filename_offset, 15);
    }

    #[test]
@@ -2,6 +2,7 @@ use crate::constants::MAX_OVERFLOW_FILES;
 use crate::error::Error;
 use crate::file_picker::FFFMode;
 use crate::git_status_worker::GitStatusWorker;
+use crate::rescan_stats::RescanReason;
 use crate::shared::{SharedFilePicker, SharedFrecency};
 use crate::sort_buffer::sort_with_buffer;
 use crate::watch::{RawWatchEvent, WatchEventKind};
@@ -324,7 +325,7 @@ impl Drop for BackgroundWatcher {
 }

 #[tracing::instrument(name = "fs_events", skip(events, shared_picker, shared_frecency, git_status_worker), level = Level::DEBUG)]
-fn handle_debounced_events(
+pub(crate) fn handle_debounced_events(
    mode: FFFMode,
    events: Vec<DebouncedEvent>,
    base_path: &Path,
@@ -342,8 +343,8 @@ fn handle_debounced_events(
        .ok()
        .and_then(|g| g.as_ref().and_then(|p| p.ignore_rules()));
    let filter = IgnoreFilter::new(base_path, walker_rules, repo.as_ref());
-    let mut need_full_rescan = false;
    let mut need_full_git_rescan = false;
+    let mut batch_overflow_attempted = false;
    let mut paths_to_remove = Vec::new();
    let mut dirs_to_remove: Vec<PathBuf> = Vec::new();
    let mut paths_to_add_or_modify = Vec::new();
@@ -353,6 +354,21 @@ fn handle_debounced_events(
    let watch_registry = shared_picker.watch_registry();
    let need_events_propagation = watch_registry.is_active();

+    let try_trigger_full_rescan = |reason: RescanReason| -> bool {
+        match shared_picker.trigger_full_rescan_with_reason(shared_frecency, reason) {
+            Ok(true) => {
+                warn!(%reason, "Triggering full rescan");
+                watch_registry.dispatch_rescan(base_path);
+                true
+            }
+            Ok(false) => false,
+            Err(e) => {
+                error!(%reason, "Failed to trigger full rescan: {:?}", e);
+                false
+            }
+        }
+    };
+
    for debounced_event in &events {
        // It is very important to not react to the access errors because we inevitably
        // gonna trigger the sync by our own preview or other unnecessary noise
@@ -370,22 +386,19 @@ fn handle_debounced_events(
        // When macOS FSEvents (or other backends) overflow their event buffer, the kernel
        // drops individual events and emits a rescan flag telling us to re-scan the subtree
        if debounced_event.event.need_rescan() {
-            if debounced_event.event.paths.len() < 16 // this should be usually one event
+            let small_and_known = debounced_event.event.paths.len() < 16 // this should be usually one event
                && debounced_event
                    .paths
                    .iter()
                    // but we are smart enough and not falling into the paths
-                    .all(|p| !p.is_dir() && !filter.is_ignored(p))
-            {
-                break;
+                    .all(|p| !p.is_dir() && !filter.is_ignored(p));
+
+            if !small_and_known && try_trigger_full_rescan(RescanReason::KernelEventLoss) {
+                return Vec::new();
            }

-            warn!(
-                "Received rescan event for paths {:?}, triggering full rescan",
-                debounced_event.event.paths
-            );
-            need_full_rescan = true;
-            break;
+            // Small batches and throttled rescans fall through: the listed
+            // paths are still applied incrementally below.
        }

        tracing::debug!(event = ?debounced_event.event, "Processing FS event");
@@ -394,13 +407,24 @@ fn handle_debounced_events(
                path.file_name().and_then(|f| f.to_str()),
                Some(".ignore") | Some(".gitignore")
            ) {
+                if path
+                    .parent()
+                    .is_some_and(|parent| filter.is_ignored(parent))
+                {
+                    continue;
+                }
+
                info!(
                    "Detected change in ignore definition file: {}",
                    path.display()
                );

-                need_full_rescan = true;
-                break;
+                if try_trigger_full_rescan(RescanReason::IgnoreFileChanged) {
+                    return Vec::new();
+                }
+
+                // Throttled: fall through so the ignore file itself stays
+                // indexed; the stale rules heal on the next admitted rescan.
            }

            if is_dotgit_change_affecting_status(path, &repo) {
@@ -462,29 +486,18 @@ fn handle_debounced_events(
        }

        affected_paths_count += debounced_event.event.paths.len();
-        if affected_paths_count > MAX_OVERFLOW_FILES {
+        if !batch_overflow_attempted && affected_paths_count > MAX_OVERFLOW_FILES * 4 {
+            batch_overflow_attempted = true;
            warn!(
                ?affected_paths_count,
-                max = MAX_OVERFLOW_FILES,
+                max = MAX_OVERFLOW_FILES * 4,
                "Too many affected paths in a single batch, triggering full rescan",
            );

-            need_full_rescan = true;
-            break;
+            if try_trigger_full_rescan(RescanReason::EventBatchOverflow) {
+                return Vec::new();
+            }
        }
-
-        if need_full_rescan {
-            break;
-        }
-    }
-
-    if need_full_rescan {
-        info!(?affected_paths_count, "Triggering full rescan");
-        watch_registry.dispatch_rescan(base_path);
-        if let Err(e) = shared_picker.trigger_full_rescan_async(shared_frecency) {
-            error!("Failed to trigger full rescan: {:?}", e);
-        }
-        return Vec::new();
    }

    // It's important to get the allocated sort
@@ -511,7 +524,7 @@ fn handle_debounced_events(
    }

    let mut files_to_update_git_status = Vec::new();
-    let mut need_full_rescan = false;
+    let mut index_update_rejected = false;
    let mut overflow_count = 0;
    let mut removed_from_dirs = Vec::new();
    let mut watch_events = ahash::AHashMap::new();
@@ -572,6 +585,13 @@ fn handle_debounced_events(

        files_to_update_git_status.reserve(paths_to_add_or_modify.len());
        for path in &paths_to_add_or_modify {
+            if picker.get_overflow_files().len() >= MAX_OVERFLOW_FILES
+                && picker.get_file_by_path(path).is_none()
+            {
+                index_update_rejected = true;
+                break;
+            }
+
            let existed = need_events_propagation && picker.get_file_by_path(path).is_some();

            if picker.handle_create_or_modify(path).is_some() {
@@ -586,7 +606,7 @@ fn handle_debounced_events(
                    watch_events.insert(path.to_path_buf(), kind);
                }
            } else {
-                need_full_rescan = true;
+                index_update_rejected = true;
            }
        }

@@ -598,13 +618,22 @@ fn handle_debounced_events(
        overflow_count, "File index changes applied",
    );

-    if need_full_rescan || overflow_count > MAX_OVERFLOW_FILES {
-        info!("Watcher faced limit of index overflow. Triggering rescan");
-        watch_registry.dispatch_rescan(base_path);
-        if let Err(e) = shared_picker.trigger_full_rescan_async(shared_frecency) {
-            error!("Failed to trigger full rescan: {:?}", e);
-        }
-    } else if need_events_propagation {
+    let rescan_started = if index_update_rejected || overflow_count > MAX_OVERFLOW_FILES {
+        let reason = if index_update_rejected {
+            RescanReason::IndexUpdateRejected
+        } else {
+            RescanReason::OverflowCapacity
+        };
+
+        info!(%reason, "Watcher faced limit of index overflow. Triggering rescan");
+        try_trigger_full_rescan(reason)
+    } else {
+        false
+    };
+
+    // When the rescan is throttled the incrementally applied changes are
+    // still the freshest state we have — propagate them to subscribers.
+    if !rescan_started && need_events_propagation {
        watch_registry.dispatch(
            base_path,
            watch_events
@@ -664,7 +693,7 @@ fn handle_debounced_events(

    // do not try to update the paths if we anyway going to rescan everything from scratch
    // no repo => no consumer thread, so don't accumulate paths nobody will drain
-    if !need_full_rescan && repo.is_some() {
+    if !index_update_rejected && repo.is_some() {
        if need_full_git_rescan {
            // A full git rescan re-reads every tracked path (including ones that just
            // went clean after a commit), so it already subsumes the per-path update.
@@ -3,3 +3,7 @@ pub use background_watcher::*;

 mod watch;
 pub use watch::*;
+
+// The harness reads rescan counters, which release builds compile out.
+#[cfg(all(test, rescan_stats))]
+mod rescan_tests;
@@ -0,0 +1,621 @@
+use std::path::PathBuf;
+use std::process::Command;
+use std::sync::Arc;
+use std::time::{Duration, Instant};
+
+use notify::Event;
+use notify::EventKind;
+use notify::event::{
+    AccessKind, AccessMode, CreateKind, DataChange, Flag, ModifyKind, RemoveKind, RenameMode,
+};
+use notify_debouncer_full::DebouncedEvent;
+use tempfile::TempDir;
+
+use super::handle_debounced_events;
+use crate::constants::MAX_OVERFLOW_FILES;
+use crate::file_picker::{FFFMode, FilePicker, FilePickerOptions};
+use crate::git_status_worker::GitStatusWorker;
+use crate::rescan_stats::{RescanReason, RescanStats};
+use crate::shared::{SharedFilePicker, SharedFrecency};
+
+#[test]
+fn saving_an_indexed_file_stays_incremental() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    f.write("src/main.rs", "fn main() { println!(); }");
+    let delta = f.feed([modify(f.path("src/main.rs"))]);
+
+    f.assert_no_rescan(&delta, "saving a tracked file");
+}
+
+#[test]
+fn editor_atomic_save_stays_incremental() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    // write-to-temp + rename-over-target, the way vim/VSCode/IntelliJ save.
+    f.write("src/main.rs", "fn main() { println!(); }");
+    let target = f.path("src/main.rs");
+    let temp = f.path("src/.main.rs.swp");
+    let delta = f.feed([
+        DebouncedEvent::new(
+            Event::new(EventKind::Create(CreateKind::File)).add_path(temp.clone()),
+            Instant::now(),
+        ),
+        DebouncedEvent::new(
+            Event::new(EventKind::Modify(ModifyKind::Name(RenameMode::From)))
+                .add_path(temp.clone()),
+            Instant::now(),
+        ),
+        DebouncedEvent::new(
+            Event::new(EventKind::Modify(ModifyKind::Name(RenameMode::To)))
+                .add_path(target.clone()),
+            Instant::now(),
+        ),
+        DebouncedEvent::new(
+            Event::new(EventKind::Remove(RemoveKind::File)).add_path(temp),
+            Instant::now(),
+        ),
+    ]);
+
+    f.assert_no_rescan(&delta, "an atomic editor save");
+    assert!(f.is_indexed("src/main.rs"), "target must stay indexed");
+}
+
+#[test]
+fn creating_and_deleting_files_stays_incremental() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    f.write("src/added.rs", "pub fn added() {}");
+    let created = f.feed([create(f.path("src/added.rs"))]);
+    f.assert_no_rescan(&created, "creating a file");
+    assert!(f.is_indexed("src/added.rs"));
+
+    f.remove("src/added.rs");
+    let removed = f.feed([remove_file(f.path("src/added.rs"))]);
+    f.assert_no_rescan(&removed, "deleting a file");
+    assert!(!f.is_indexed("src/added.rs"));
+}
+
+#[test]
+fn deleting_a_directory_stays_incremental() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.write("src/nested/a.rs", "");
+    f.write("src/nested/b.rs", "");
+    f.index();
+
+    std::fs::remove_dir_all(f.path("src/nested")).unwrap();
+    let delta = f.feed([DebouncedEvent::new(
+        Event::new(EventKind::Remove(RemoveKind::Folder)).add_path(f.path("src/nested")),
+        Instant::now(),
+    )]);
+
+    f.assert_no_rescan(&delta, "deleting a directory");
+    assert!(!f.is_indexed("src/nested/a.rs"));
+    assert!(f.is_indexed("src/main.rs"));
+}
+
+#[test]
+fn read_only_access_events_are_ignored() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    // fff's own preview + grep reads generate these; reacting to them would
+    // make the picker rescan whenever the user scrolls the result list.
+    let path = f.path("src/main.rs");
+    let delta = f.feed([
+        DebouncedEvent::new(
+            Event::new(EventKind::Access(AccessKind::Read)).add_path(path.clone()),
+            Instant::now(),
+        ),
+        DebouncedEvent::new(
+            Event::new(EventKind::Access(AccessKind::Open(AccessMode::Read)))
+                .add_path(path.clone()),
+            Instant::now(),
+        ),
+        DebouncedEvent::new(
+            Event::new(EventKind::Access(AccessKind::Close(AccessMode::Read))).add_path(path),
+            Instant::now(),
+        ),
+    ]);
+
+    f.assert_no_rescan(&delta, "read-only access events");
+}
+
+#[test]
+fn recreating_the_same_paths_does_not_consume_overflow_capacity() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    // Recreated paths must reuse their overflow slots.
+    for _ in 0..8 {
+        for i in 0..200 {
+            let rel = format!("gen/out{i}.rs");
+            f.write(&rel, "generated");
+            f.feed([create(f.path(&rel))]);
+        }
+        for i in 0..200 {
+            let rel = format!("gen/out{i}.rs");
+            f.remove(&rel);
+            f.feed([remove_file(f.path(&rel))]);
+        }
+    }
+
+    let delta = f.all_rescans();
+    f.assert_no_rescan(&delta, "1600 create/delete cycles over 200 stable paths");
+    assert!(
+        f.overflow_len() <= 200,
+        "each path must claim one overflow slot at most, got {}",
+        f.overflow_len()
+    );
+}
+
+#[test]
+fn writes_inside_a_gitignored_directory_stay_incremental() {
+    let f = Fixture::with_git();
+    f.write(".gitignore", "target/\nnode_modules/\n");
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let mut events = Vec::new();
+    for i in 0..64 {
+        let rel = format!("target/debug/artifact{i}.o");
+        f.write(&rel, "binary");
+        events.push(create(f.path(&rel)));
+    }
+
+    let delta = f.feed(events);
+    f.assert_no_rescan(&delta, "build output written into an ignored directory");
+}
+
+#[test]
+fn ignored_event_batch_above_index_capacity_stays_incremental() {
+    let f = Fixture::with_git();
+    f.write(".gitignore", "node_modules/\n");
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let events = (0..MAX_OVERFLOW_FILES + 1)
+        .map(|i| {
+            let rel = format!("node_modules/pkg/file{i}.js");
+            f.write(&rel, "");
+            create(f.path(&rel))
+        })
+        .collect::<Vec<_>>();
+
+    let delta = f.feed(events);
+    f.assert_no_rescan(&delta, "ignored events above the index capacity");
+    assert_eq!(f.overflow_len(), 0);
+}
+
+#[test]
+fn repeated_edits_above_index_capacity_stay_incremental() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let path = f.path("src/main.rs");
+    let events = (0..MAX_OVERFLOW_FILES + 1)
+        .map(|_| modify(path.clone()))
+        .collect::<Vec<_>>();
+
+    let delta = f.feed(events);
+    f.assert_no_rescan(&delta, "repeated edits above the index capacity");
+    assert_eq!(f.overflow_len(), 0);
+}
+
+#[test]
+fn ignore_file_inside_an_ignored_directory_stays_incremental() {
+    let f = Fixture::with_git();
+    f.write(".gitignore", "node_modules/\n");
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let ignore_files =
+        ["left-pad", "lodash", "typescript"].map(|pkg| format!("node_modules/{pkg}/.gitignore"));
+
+    for rel in &ignore_files {
+        f.write(rel, "dist\n");
+    }
+    let delta = f.feed(ignore_files.iter().map(|rel| create(f.path(rel))));
+    f.assert_no_rescan(&delta, "creating ignored .gitignore files");
+
+    for rel in &ignore_files {
+        f.write(rel, "build\n");
+    }
+    let delta = f.feed(ignore_files.iter().map(|rel| modify(f.path(rel))));
+    f.assert_no_rescan(&delta, "modifying ignored .gitignore files");
+
+    for rel in &ignore_files {
+        f.remove(rel);
+    }
+    let delta = f.feed(ignore_files.iter().map(|rel| remove_file(f.path(rel))));
+    f.assert_no_rescan(&delta, "removing ignored .gitignore files");
+}
+
+#[test]
+fn ignore_file_inside_an_indexed_directory_triggers_a_rescan() {
+    let f = Fixture::with_git();
+    f.write("src/.gitignore", ".gitignore\ngenerated/\n");
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    f.write("src/.gitignore", ".gitignore\ngenerated/\nbuild/\n");
+    let delta = f.feed([modify(f.path("src/.gitignore"))]);
+
+    assert_eq!(delta.count(RescanReason::IgnoreFileChanged), 1);
+}
+
+#[test]
+fn git_internal_churn_stays_incremental() {
+    let f = Fixture::with_git();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let git_dir = f.path(".git");
+    let delta = f.feed([
+        create(git_dir.join("index.lock")),
+        modify(git_dir.join("index")),
+        remove_file(git_dir.join("index.lock")),
+        modify(git_dir.join("HEAD")),
+        modify(git_dir.join("logs/HEAD")),
+        modify(git_dir.join("COMMIT_EDITMSG")),
+        modify(git_dir.join("refs/heads/main")),
+    ]);
+
+    f.assert_no_rescan(&delta, "git writing its own metadata");
+}
+
+#[test]
+fn changing_the_root_ignore_file_triggers_a_rescan() {
+    let f = Fixture::with_git();
+    f.write(".gitignore", "target/\n");
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    f.write(".gitignore", "target/\nsrc/\n");
+    let delta = f.feed([modify(f.path(".gitignore"))]);
+
+    assert_eq!(
+        delta.count(RescanReason::IgnoreFileChanged),
+        1,
+        "the indexed set depends on the root ignore rules, got {delta}"
+    );
+}
+
+#[test]
+fn kernel_event_loss_on_a_directory_triggers_a_rescan() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let delta = f.feed([DebouncedEvent::new(
+        Event::new(EventKind::Modify(ModifyKind::Any))
+            .add_path(f.path("src"))
+            .set_flag(Flag::Rescan),
+        Instant::now(),
+    )]);
+
+    assert_eq!(
+        delta.count(RescanReason::KernelEventLoss),
+        1,
+        "a dropped-events flag over a directory means unknown subtree state, got {delta}"
+    );
+}
+
+#[test]
+fn new_files_above_index_capacity_trigger_a_rescan() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let events = (0..MAX_OVERFLOW_FILES + 1)
+        .map(|i| {
+            let rel = format!("src/bulk{i}.rs");
+            f.write(&rel, "");
+            create(f.path(&rel))
+        })
+        .collect::<Vec<_>>();
+
+    let delta = f.feed(events);
+    assert_eq!(
+        delta.count(RescanReason::IndexUpdateRejected),
+        1,
+        "new files above the overflow region cannot be applied incrementally, got {delta}"
+    );
+}
+
+#[test]
+fn batch_at_the_overflow_boundary_stays_incremental() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let events = (0..MAX_OVERFLOW_FILES)
+        .map(|i| {
+            let rel = format!("src/bulk{i}.rs");
+            f.write(&rel, "");
+            create(f.path(&rel))
+        })
+        .collect::<Vec<_>>();
+
+    let delta = f.feed(events);
+    f.assert_no_rescan(&delta, "a batch exactly at the overflow limit");
+}
+
+#[test]
+fn event_batch_at_four_times_index_capacity_stays_incremental() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let path = f.path("src/main.rs");
+    let events = (0..MAX_OVERFLOW_FILES * 4)
+        .map(|_| modify(path.clone()))
+        .collect::<Vec<_>>();
+
+    let delta = f.feed(events);
+    f.assert_no_rescan(&delta, "an event batch exactly at the event limit");
+}
+
+#[test]
+fn event_batch_above_four_times_index_capacity_triggers_a_rescan() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    let path = f.path("src/main.rs");
+    let events = (0..MAX_OVERFLOW_FILES * 4 + 1)
+        .map(|_| modify(path.clone()))
+        .collect::<Vec<_>>();
+
+    let delta = f.feed(events);
+    assert_eq!(
+        delta.count(RescanReason::EventBatchOverflow),
+        1,
+        "an event batch above four times the index capacity must rescan, got {delta}"
+    );
+}
+
+#[test]
+fn repeated_triggers_inside_the_cooldown_collapse_to_one_rescan() {
+    let f = Fixture::with_git();
+    f.write(".gitignore", "target/\n");
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    // Repeated batches during the cooldown must share one walk.
+    for round in 0..50 {
+        f.write(".gitignore", &format!("target/\n# round {round}\n"));
+        f.feed([modify(f.path(".gitignore"))]);
+    }
+
+    let stats = f.all_rescans();
+    assert_eq!(
+        stats.total, 1,
+        "50 triggers inside the cooldown must collapse to a single walk, got {stats}"
+    );
+    assert_eq!(
+        stats.throttled, 49,
+        "every suppressed request must be accounted for, got {stats}"
+    );
+}
+
+#[test]
+fn an_explicit_request_is_never_throttled() {
+    let f = Fixture::with_git();
+    f.write(".gitignore", "target/\n");
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    // Burn the cooldown with a watcher trigger, then confirm a user-initiated
+    // refresh still goes through.
+    f.write(".gitignore", "target/\nsrc/\n");
+    f.feed([modify(f.path(".gitignore"))]);
+
+    for _ in 0..3 {
+        f.picker.trigger_full_rescan_async(&f.frecency).unwrap();
+    }
+
+    let stats = f.all_rescans();
+    assert_eq!(
+        stats.count(RescanReason::Explicit),
+        3,
+        "explicit refreshes must bypass the throttle, got {stats}"
+    );
+    assert_eq!(stats.count_throttled(RescanReason::Explicit), 0);
+}
+
+#[test]
+fn events_after_a_suppressed_kernel_rescan_are_still_applied() {
+    let f = Fixture::new();
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    f.write("src/added.rs", "pub fn added() {}");
+    let delta = f.feed([
+        DebouncedEvent::new(
+            Event::new(EventKind::Modify(ModifyKind::Data(DataChange::Content)))
+                .add_path(f.path("src/main.rs"))
+                .set_flag(Flag::Rescan),
+            Instant::now(),
+        ),
+        create(f.path("src/added.rs")),
+    ]);
+
+    f.assert_no_rescan(&delta, "a dropped-events flag over a single tracked file");
+    assert!(
+        f.is_indexed("src/added.rs"),
+        "suppressing the rescan must not drop the rest of the batch"
+    );
+}
+
+#[test]
+fn a_throttled_ignore_file_event_is_still_applied_incrementally() {
+    let f = Fixture::with_git();
+    f.write(".gitignore", "target/\n");
+    f.write("src/main.rs", "fn main() {}");
+    f.index();
+
+    // Burn the cooldown: deleting .gitignore admits a full rescan.
+    f.remove(".gitignore");
+    let delta = f.feed([remove_file(f.path(".gitignore"))]);
+    assert_eq!(delta.count(RescanReason::IgnoreFileChanged), 1);
+    f.picker.wait_for_indexing_complete(Duration::from_secs(10));
+
+    // Recreating it inside the cooldown throttles the rescan, but the file
+    // itself must re-enter the index via the incremental fallback.
+    f.write(".gitignore", "target/\n__ignored_x/\n");
+    let delta = f.feed([create(f.path(".gitignore"))]);
+    assert_eq!(delta.total, 0, "the rescan must be throttled, got {delta}");
+    assert_eq!(delta.count_throttled(RescanReason::IgnoreFileChanged), 1);
+    assert!(
+        f.is_indexed(".gitignore"),
+        "a throttled ignore-file event must still index the file itself"
+    );
+}
+
+struct Fixture {
+    base: PathBuf,
+    picker: SharedFilePicker,
+    frecency: SharedFrecency,
+    git_workdir: Option<PathBuf>,
+    git_worker: Arc<GitStatusWorker>,
+    // Dropped last so background work started by a triggered rescan still
+    // sees the tree it was asked to walk.
+    _tmp: TempDir,
+}
+
+impl Fixture {
+    fn new() -> Self {
+        Self::build(false)
+    }
+
+    fn with_git() -> Self {
+        Self::build(true)
+    }
+
+    fn build(git: bool) -> Self {
+        let tmp = tempfile::tempdir().unwrap();
+        let base = crate::path_utils::canonicalize(tmp.path()).unwrap();
+        let git_workdir = git.then(|| {
+            let status = Command::new("git")
+                .args(["init", "-b", "main"])
+                .current_dir(&base)
+                .output()
+                .expect("git init");
+            assert!(status.status.success(), "git init failed");
+            base.clone()
+        });
+
+        Self {
+            base,
+            picker: SharedFilePicker::default(),
+            frecency: SharedFrecency::noop(),
+            git_workdir,
+            git_worker: GitStatusWorker::new(),
+            _tmp: tmp,
+        }
+    }
+
+    fn index(&self) {
+        let mut picker = FilePicker::new(FilePickerOptions {
+            base_path: self.base.to_string_lossy().into_owned(),
+            watch: false,
+            ..Default::default()
+        })
+        .unwrap();
+        picker.collect_files().unwrap();
+        self.picker.rebase_watches(&self.base);
+        *self.picker.write().unwrap() = Some(picker);
+    }
+
+    fn feed(&self, events: impl IntoIterator<Item = DebouncedEvent>) -> RescanStats {
+        let before = self.picker.rescan_stats();
+        handle_debounced_events(
+            FFFMode::Neovim,
+            events.into_iter().collect(),
+            &self.base,
+            &self.git_workdir,
+            &self.picker,
+            &self.frecency,
+            &self.git_worker,
+        );
+
+        self.picker.rescan_stats().since(&before)
+    }
+
+    fn assert_no_rescan(&self, delta: &RescanStats, what: &str) {
+        assert_eq!(delta.total, 0, "{what} must not trigger a rescan: {delta}");
+    }
+
+    fn path(&self, rel: &str) -> PathBuf {
+        self.base.join(rel)
+    }
+
+    fn write(&self, rel: &str, contents: &str) {
+        let path = self.path(rel);
+        std::fs::create_dir_all(path.parent().unwrap()).unwrap();
+        std::fs::write(path, contents).unwrap();
+    }
+
+    fn remove(&self, rel: &str) {
+        std::fs::remove_file(self.path(rel)).unwrap();
+    }
+
+    fn is_indexed(&self, rel: &str) -> bool {
+        let guard = self.picker.read().unwrap();
+        guard
+            .as_ref()
+            .and_then(|p| p.get_file_by_path(self.path(rel)))
+            .is_some_and(|file| !file.is_deleted())
+    }
+
+    fn all_rescans(&self) -> RescanStats {
+        self.picker.rescan_stats()
+    }
+
+    fn overflow_len(&self) -> usize {
+        let guard = self.picker.read().unwrap();
+        guard
+            .as_ref()
+            .map(|p| p.get_overflow_files().len())
+            .unwrap_or(0)
+    }
+}
+
+impl Drop for Fixture {
+    fn drop(&mut self) {
+        // A test that intentionally triggers a rescan leaves a walk running on
+        // the background pool; let it finish before the tree disappears.
+        self.picker
+            .wait_for_indexing_complete(Duration::from_secs(10));
+    }
+}
+
+fn event(kind: EventKind, path: PathBuf) -> DebouncedEvent {
+    DebouncedEvent::new(Event::new(kind).add_path(path), Instant::now())
+}
+
+fn create(path: PathBuf) -> DebouncedEvent {
+    event(EventKind::Create(CreateKind::File), path)
+}
+
+fn modify(path: PathBuf) -> DebouncedEvent {
+    event(
+        EventKind::Modify(ModifyKind::Data(DataChange::Content)),
+        path,
+    )
+}
+
+fn remove_file(path: PathBuf) -> DebouncedEvent {
+    event(EventKind::Remove(RemoveKind::File), path)
+}
@@ -0,0 +1,123 @@
+//! One process must be able to hold many trackers over the same LMDB path
+//! (issues #700/#760): they share a single pooled env instead of failing
+//! with `EnvAlreadyOpened`.
+
+use std::path::{Path, PathBuf};
+
+use fff_search::frecency::FrecencyTracker;
+use fff_search::query_tracker::QueryTracker;
+use fff_search::shared::SharedFrecency;
+
+fn unique_dir(name: &str) -> PathBuf {
+    let dir = std::env::temp_dir().join(format!("fff-env-pool-{name}-{}", std::process::id()));
+    let _ = std::fs::remove_dir_all(&dir);
+    dir
+}
+
+#[test]
+fn same_path_trackers_share_one_env() {
+    let dir = unique_dir("share");
+    let file = Path::new("/virtual/env-pool/shared.rs");
+
+    let a = FrecencyTracker::open(&dir).expect("first open");
+    let b = FrecencyTracker::open(&dir).expect("second open in the same process (#700/#760)");
+
+    a.track_access(file).expect("write via a");
+    assert_eq!(b.access_count(file).expect("read via b"), 1);
+
+    drop(a);
+    b.track_access(file)
+        .expect("b must stay usable after a drops");
+    assert_eq!(b.access_count(file).unwrap(), 2);
+    drop(b);
+
+    let c = FrecencyTracker::open(&dir).expect("reopen after all handles dropped");
+    assert_eq!(
+        c.access_count(file).unwrap(),
+        2,
+        "data persisted across reopen"
+    );
+
+    drop(c);
+    let _ = std::fs::remove_dir_all(&dir);
+}
+
+#[test]
+fn concurrent_open_and_drop_never_collide() {
+    let dir = unique_dir("hammer");
+    let file = Path::new("/virtual/env-pool/hammer.rs");
+
+    let mut handles = Vec::new();
+    for t in 0..8 {
+        let dir = dir.clone();
+        handles.push(std::thread::spawn(move || {
+            for i in 0..100 {
+                let tracker = FrecencyTracker::open(&dir)
+                    .unwrap_or_else(|e| panic!("thread {t} iteration {i}: {e}"));
+                if i % 20 == 0 {
+                    tracker.track_access(file).expect("track access");
+                }
+            }
+        }));
+    }
+    for handle in handles {
+        handle.join().expect("no thread may panic");
+    }
+
+    let _ = std::fs::remove_dir_all(&dir);
+}
+
+#[test]
+fn different_store_on_same_path_is_rejected_with_clear_error() {
+    let dir = unique_dir("mismatch");
+
+    let _frecency = FrecencyTracker::open(&dir).expect("frecency open");
+    let err = QueryTracker::open(&dir).expect_err("env options differ, must be rejected");
+
+    let msg = err.to_string();
+    assert!(
+        msg.contains("frecency") && msg.contains("query"),
+        "error must name both stores so the user can fix their config, got: {msg}"
+    );
+
+    let _ = std::fs::remove_dir_all(&dir);
+}
+
+#[test]
+fn destroy_refuses_while_shared_then_succeeds_when_sole() {
+    let dir = unique_dir("destroy");
+    let file = Path::new("/virtual/env-pool/destroy.rs");
+
+    let shared = SharedFrecency::default();
+    shared
+        .init(FrecencyTracker::open(&dir).expect("init open"))
+        .expect("init");
+    let other = FrecencyTracker::open(&dir).expect("second handle over the same db");
+
+    shared
+        .destroy()
+        .expect_err("destroy must refuse while another tracker uses the env");
+
+    // Refusal must keep both the files and the shared handle intact.
+    assert!(
+        dir.join("data.mdb").exists(),
+        "db files survive a refused destroy"
+    );
+    shared
+        .read()
+        .expect("read lock")
+        .as_ref()
+        .expect("tracker restored after refused destroy")
+        .track_access(file)
+        .expect("shared handle still works");
+
+    drop(other);
+    let removed = shared
+        .destroy()
+        .expect("sole-owner destroy succeeds")
+        .expect("a path was removed");
+    assert!(
+        !removed.exists(),
+        "db dir deleted once nobody shares the env"
+    );
+}
@@ -222,6 +222,17 @@ fn multi_grep_with_file_path_suffix_constraint() {
    }
 }

+#[test]
+fn multi_grep_with_missing_file_path_constraint_returns_no_matches() {
+    let tmp = TempDir::new().unwrap();
+    let picker = create_picker(tmp.path(), &[("other.lua", "handleRequest\n")]);
+
+    let constraints = [Constraint::FilePath("missing.lua")];
+    let result = picker.multi_grep(&["handleRequest"], &constraints, &plain_opts());
+
+    assert!(result.matches.is_empty());
+}
+
 /// Glob constraints must match native Windows paths — the picker normalises
 /// separators when handing paths to the glob matcher.
 #[test]
@@ -0,0 +1,343 @@
+#![cfg(rescan_stats)]
+
+use std::path::{Path, PathBuf};
+use std::process::Command;
+use std::time::Duration;
+
+use fff_search::file_picker::{FFFMode, FilePicker};
+use fff_search::{FilePickerOptions, RescanStats, SharedFilePicker, SharedFrecency};
+use tempfile::TempDir;
+
+const SETTLE: Duration = Duration::from_millis(600);
+
+#[test]
+fn saving_source_files_does_not_rescan() {
+    let repo = WatchedRepo::new(|base| {
+        write(base, ".gitignore", "target/\n");
+        for i in 0..20 {
+            write(base, &format!("src/mod{i}.rs"), "pub fn f() {}");
+        }
+    });
+
+    for round in 0..10 {
+        for i in 0..20 {
+            repo.write(
+                &format!("src/mod{i}.rs"),
+                &format!("pub fn f() {{ let _ = {round}; }}"),
+            );
+        }
+        repo.settle();
+    }
+
+    repo.assert_quiet("200 file saves");
+}
+
+#[test]
+fn build_output_in_ignored_directories_does_not_rescan() {
+    let repo = WatchedRepo::new(|base| {
+        write(base, ".gitignore", "target/\nnode_modules/\ndist/\n");
+        write(base, "src/main.rs", "fn main() {}");
+    });
+
+    for round in 0..4 {
+        for i in 0..150 {
+            repo.write(&format!("target/debug/deps/unit-{round}-{i}.o"), "binary");
+            repo.write(&format!("dist/chunk-{round}-{i}.js"), "bundled");
+        }
+        repo.settle();
+    }
+
+    repo.assert_quiet("1200 build artifacts written into ignored directories");
+}
+
+#[test]
+fn adding_source_files_and_directories_does_not_rescan() {
+    let repo = WatchedRepo::new(|base| {
+        write(base, ".gitignore", "target/\n");
+        write(base, "src/main.rs", "fn main() {}");
+    });
+
+    for i in 0..40 {
+        repo.write(&format!("src/feature{i}/mod.rs"), "pub mod inner;");
+        repo.write(&format!("src/feature{i}/inner.rs"), "pub fn go() {}");
+    }
+    repo.settle();
+
+    assert!(
+        repo.wait_indexed("src/feature39/inner.rs"),
+        "watcher must index files in newly created directories"
+    );
+    repo.assert_quiet("40 new directories with 80 files");
+}
+
+#[test]
+fn recreating_generated_files_does_not_rescan() {
+    let repo = WatchedRepo::new(|base| {
+        write(base, ".gitignore", "target/\n");
+        write(base, "src/main.rs", "fn main() {}");
+    });
+
+    // Recreated paths must reuse their overflow slots.
+    for round in 0..12 {
+        for i in 0..40 {
+            repo.write(&format!("src/generated/api{i}.rs"), "pub struct A;");
+        }
+        repo.settle();
+        for i in 0..40 {
+            repo.remove(&format!("src/generated/api{i}.rs"));
+        }
+        repo.settle();
+        assert!(
+            repo.overflow_len() <= 64,
+            "round {round}: regenerating the same paths grew the overflow region to {}",
+            repo.overflow_len()
+        );
+    }
+
+    repo.assert_quiet("12 codegen cycles over 40 stable paths");
+}
+
+#[test]
+fn git_workflow_does_not_rescan() {
+    let repo = WatchedRepo::new(|base| {
+        write(base, ".gitignore", "target/\n");
+        write(base, "src/main.rs", "fn main() {}");
+        write(base, "src/lib.rs", "pub mod thing;");
+        git(base, &["init", "-b", "main"]);
+        git(base, &["add", "-A"]);
+        git(base, &["commit", "-m", "initial"]);
+    });
+
+    repo.write("src/main.rs", "fn main() { println!(\"hi\"); }");
+    repo.settle();
+    repo.git(&["add", "-A"]);
+    repo.settle();
+    repo.git(&["commit", "-m", "second"]);
+    repo.settle();
+    repo.git(&["checkout", "-b", "feature"]);
+    repo.settle();
+    repo.write("src/feature.rs", "pub fn feature() {}");
+    repo.git(&["add", "-A"]);
+    repo.git(&["commit", "-m", "feature"]);
+    repo.settle();
+    repo.git(&["checkout", "main"]);
+    repo.settle();
+    repo.git(&["merge", "feature"]);
+    repo.settle();
+
+    repo.assert_quiet("a commit / branch / merge cycle");
+}
+
+#[test]
+fn reading_files_does_not_rescan() {
+    let repo = WatchedRepo::new(|base| {
+        write(base, ".gitignore", "target/\n");
+        for i in 0..50 {
+            write(base, &format!("src/mod{i}.rs"), "pub fn f() {}");
+        }
+    });
+
+    // Preview rendering and grep open every file in the result list. Reacting
+    // to those reads would make the picker rescan while the user scrolls.
+    for _ in 0..5 {
+        for i in 0..50 {
+            let _ = std::fs::read(repo.path(&format!("src/mod{i}.rs"))).unwrap();
+        }
+    }
+    repo.settle();
+
+    repo.assert_quiet("reading every indexed file");
+}
+
+#[test]
+fn npm_install_style_churn_does_not_rescan() {
+    let repo = WatchedRepo::new(|base| {
+        write(base, ".gitignore", "node_modules/\n");
+        write(base, "src/index.ts", "export const a = 1;");
+    });
+
+    for pkg in 0..100 {
+        repo.write(&format!("node_modules/pkg{pkg}/package.json"), "{}");
+        repo.write(
+            &format!("node_modules/pkg{pkg}/index.js"),
+            "module.exports={}",
+        );
+        repo.write(&format!("node_modules/pkg{pkg}/.gitignore"), "dist\n");
+    }
+    repo.settle();
+    repo.settle();
+
+    repo.assert_quiet("an npm install into an ignored node_modules");
+}
+
+#[test]
+fn a_churning_root_is_capped_at_one_rescan_per_cooldown() {
+    let repo = WatchedRepo::new(|base| {
+        write(base, "src/main.rs", "fn main() {}");
+    });
+
+    // Root ignore changes force watcher rescan requests.
+    for round in 0..25 {
+        repo.write(".gitignore", &format!("target/\n# round {round}\n"));
+        std::thread::sleep(Duration::from_millis(120));
+    }
+    repo.settle();
+
+    let stats = repo.rescans();
+    assert!(
+        stats.total <= 1,
+        "a churning root must not exceed one walk per cooldown, got {stats}"
+    );
+    assert!(
+        stats.throttled > 0,
+        "the suppressed triggers must be recorded, got {stats}"
+    );
+}
+
+struct WatchedRepo {
+    base: PathBuf,
+    picker: SharedFilePicker,
+    _frecency: SharedFrecency,
+    _tmp: TempDir,
+}
+
+impl WatchedRepo {
+    fn new(setup: impl FnOnce(&Path)) -> Self {
+        let tmp = tempfile::tempdir().unwrap();
+        let base = fff_search::path_utils::canonicalize(tmp.path()).unwrap();
+        setup(&base);
+
+        let picker = SharedFilePicker::default();
+        let frecency = SharedFrecency::noop();
+        FilePicker::new_with_shared_state(
+            picker.clone(),
+            frecency.clone(),
+            FilePickerOptions {
+                base_path: base.to_string_lossy().into_owned(),
+                enable_mmap_cache: false,
+                mode: FFFMode::Neovim,
+                watch: true,
+                ..Default::default()
+            },
+        )
+        .expect("failed to create file picker");
+
+        assert!(
+            picker.wait_for_scan(Duration::from_secs(60)),
+            "timed out waiting for the initial scan"
+        );
+        assert!(
+            picker.wait_for_watcher(Duration::from_secs(60)),
+            "timed out waiting for the watcher"
+        );
+
+        let repo = Self {
+            base,
+            picker,
+            _frecency: frecency,
+            _tmp: tmp,
+        };
+        repo.settle();
+        repo.picker.reset_rescan_stats();
+        repo
+    }
+
+    fn settle(&self) {
+        std::thread::sleep(SETTLE);
+        assert!(
+            self.picker
+                .wait_for_indexing_complete(Duration::from_secs(60)),
+            "timed out waiting for background indexing to finish"
+        );
+    }
+
+    fn assert_quiet(&self, workload: &str) {
+        let stats = self.rescans();
+        assert_eq!(
+            stats.watcher_triggered(),
+            0,
+            "{workload} must be absorbed incrementally, but the watcher fell back to {stats}"
+        );
+    }
+
+    fn rescans(&self) -> RescanStats {
+        self.picker.rescan_stats()
+    }
+
+    fn path(&self, rel: &str) -> PathBuf {
+        self.base.join(rel)
+    }
+
+    fn write(&self, rel: &str, contents: &str) {
+        write(&self.base, rel, contents);
+    }
+
+    fn remove(&self, rel: &str) {
+        std::fs::remove_file(self.path(rel)).unwrap();
+    }
+
+    fn git(&self, args: &[&str]) {
+        git(&self.base, args);
+    }
+
+    fn wait_indexed(&self, rel: &str) -> bool {
+        let deadline = std::time::Instant::now() + Duration::from_secs(30);
+        while std::time::Instant::now() < deadline {
+            if self.is_indexed(rel) {
+                return true;
+            }
+            std::thread::sleep(Duration::from_millis(50));
+        }
+        false
+    }
+
+    fn is_indexed(&self, rel: &str) -> bool {
+        let guard = self.picker.read().unwrap();
+        guard
+            .as_ref()
+            .and_then(|p| p.get_file_by_path(self.path(rel)))
+            .is_some_and(|file| !file.is_deleted())
+    }
+
+    fn overflow_len(&self) -> usize {
+        let guard = self.picker.read().unwrap();
+        guard
+            .as_ref()
+            .map(|p| p.get_overflow_files().len())
+            .unwrap_or(0)
+    }
+}
+
+impl Drop for WatchedRepo {
+    fn drop(&mut self) {
+        // Stop the watcher before the tree disappears, otherwise a late batch
+        // races the tempdir removal.
+        if let Ok(mut guard) = self.picker.write() {
+            guard.take();
+        }
+    }
+}
+
+fn write(base: &Path, rel: &str, contents: &str) {
+    let path = base.join(rel);
+    std::fs::create_dir_all(path.parent().unwrap()).unwrap();
+    std::fs::write(path, contents).unwrap();
+}
+
+fn git(dir: &Path, args: &[&str]) {
+    let out = Command::new("git")
+        .args(args)
+        .current_dir(dir)
+        .env("GIT_AUTHOR_NAME", "test")
+        .env("GIT_AUTHOR_EMAIL", "test@test.com")
+        .env("GIT_COMMITTER_NAME", "test")
+        .env("GIT_COMMITTER_EMAIL", "test@test.com")
+        .output()
+        .unwrap_or_else(|e| panic!("git {args:?} failed to spawn: {e}"));
+
+    assert!(
+        out.status.success(),
+        "git {args:?} failed: {}",
+        String::from_utf8_lossy(&out.stderr)
+    );
+}
@@ -3,7 +3,7 @@ name = "fff-grep"
 description = "File grepping logic for fff"
 license = "MIT"
 authors = ["Dmitriy Kovalenko <dmtr.kovalenko@outlok.com>"]
-version = "0.10.1"
+version = "0.10.4"
 edition = "2024"

 [lints]
@@ -1,6 +1,6 @@
 [package]
 name = "fff-mcp"
-version = "0.10.1"
+version = "0.10.4"
 edition = "2024"
 description = "MCP server for FFF file finder - drop-in replacement for AI code assistant search tools"
 license = "MIT"
@@ -19,8 +19,8 @@ ripgrep = ["fff/ripgrep", "fff-query-parser/ripgrep"]
 zlob = ["fff/zlob", "fff-query-parser/zlob"]

 [dependencies]
-fff = { package = "fff-search", path = "../fff-core", default-features = false , version = "0.10.1", features = ["definitions"] }
-fff-query-parser = { path = "../fff-query-parser", default-features = false , version = "0.10.1" }
+fff = { package = "fff-search", path = "../fff-core", default-features = false , version = "0.10.4", features = ["definitions"] }
+fff-query-parser = { path = "../fff-query-parser", default-features = false , version = "0.10.4" }
 mimalloc = { workspace = true }
 rmcp = { version = "1.7.0", features = ["server", "transport-io"] }
 schemars = "1.0"
@@ -30,3 +30,13 @@ tokio = { version = "1", features = ["full"] }
 tracing = { workspace = true }
 git2 = { workspace = true }
 clap = { version = "4", features = ["derive", "env"] }
+
+[target.'cfg(windows)'.dependencies]
+windows-sys = { version = "0.60", features = [
+    "Win32_Foundation",
+    "Win32_System_Threading",
+    "Win32_System_Diagnostics_ToolHelp",
+] }
+
+[dev-dependencies]
+tempfile = "3.8"
@@ -1,9 +1,12 @@
 mod cursor;
 mod healthcheck;
 mod output;
+mod parent;
 mod server;
 mod update_check;

+use std::time::{Duration, SystemTime};
+
 use clap::Parser;
 use fff::file_picker::FilePicker;
 use fff::frecency::FrecencyTracker;
@@ -92,7 +95,7 @@ pub const MCP_INSTRUCTIONS: &str = concat!(
    "  !generated/ - exclude generated code",
 );

-/// FFF MCP Server -- a high performance & accuracy file finder for AI code assistants.
+/// FFF MCP Server - a high performance & accuracy file finder for AI code assistants.
 #[derive(Parser)]
 #[command(name = "fff-mcp", version = concat!(env!("CARGO_PKG_VERSION"), " (", env!("FFF_GIT_HASH"), ")"))]
 pub(crate) struct Args {
@@ -131,6 +134,7 @@ pub(crate) struct Args {

    /// Disable the content index built after the initial scan.
    /// This makes grep calls slower but consumes less RAM (recommended to not turn off)
+    #[arg(long = "no-content-indexing")]
    no_content_indexing: bool,

    /// Explicitly enable content indexing even when `--no-warmup` is set.
@@ -182,11 +186,12 @@ pub(crate) struct Args {
    #[arg(long = "healthcheck")]
    pub(crate) healthcheck: bool,

-    /// Exit after this many seconds of inactivity. 0 = never exit.
+    /// Timeout of inactivity after which fff mcp will be exited. Even if the parent process
+    /// is alive we don't want to occupy resources on index and file watches if fff is unused
    #[arg(
        long = "idle-timeout-secs",
        env = "FFF_MCP_IDLE_TIMEOUT_SECS",
-        default_value_t = 900
+        default_value_t = 60 * 60
    )]
    idle_timeout_secs: u64,
 }
@@ -342,9 +347,20 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
        }
    };

-    if idle_timeout_secs > 0 {
+    let parent_watcher = parent::ParentWatcher::new();
+    match &parent_watcher {
+        Some(watcher) => tracing::info!(
+            "Watching parent process (pid {}); will exit when it dies",
+            watcher.parent_pid()
+        ),
+        None => tracing::warn!(
+            "Parent process liveness detection unavailable; idle timeout will exit unconditionally"
+        ),
+    }
+
+    if idle_timeout_secs > 0 || parent_watcher.is_some() {
        last_activity.store(
-            std::time::SystemTime::now()
+            SystemTime::now()
                .duration_since(std::time::UNIX_EPOCH)
                .map(|d| d.as_secs())
                .unwrap_or(0),
@@ -353,9 +369,27 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {

        let last_activity_for_watchdog = last_activity.clone();
        tokio::spawn(async move {
-            let tick = std::time::Duration::from_secs(60);
+            let tick = watchdog_interval();
            loop {
                tokio::time::sleep(tick).await;
+
+                if let Some(ref watcher) = parent_watcher {
+                    if !watcher.parent_alive() {
+                        tracing::info!(
+                            "Parent process (pid {}) exited, shutting down",
+                            watcher.parent_pid()
+                        );
+                        flush_logs_and_exit().await;
+                    }
+                    // Parent is alive: it owns our lifecycle, never exit on idle
+                    // Clients like Codex do not restart MCP servers @see #703
+                    continue;
+                }
+
+                if idle_timeout_secs == 0 {
+                    continue;
+                }
+
                let now = std::time::SystemTime::now()
                    .duration_since(std::time::UNIX_EPOCH)
                    .map(|d| d.as_secs())
@@ -363,12 +397,8 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {

                let last = last_activity_for_watchdog.load(std::sync::atomic::Ordering::Relaxed);
                if now.saturating_sub(last) >= idle_timeout_secs {
-                    tracing::info!(
-                        "Exiting after {}s of inactivity (idle_timeout_secs={})",
-                        now.saturating_sub(last),
-                        idle_timeout_secs
-                    );
-                    std::process::exit(0);
+                    tracing::info!(?idle_timeout_secs, "Exiting due to inactivity",);
+                    flush_logs_and_exit().await;
                }
            }
        });
@@ -395,3 +425,20 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {

    Ok(())
 }
+
+// Tracing appender is non blocking, to get full log give it some time before hard exit
+async fn flush_logs_and_exit() -> ! {
+    tokio::time::sleep(std::time::Duration::from_millis(250)).await;
+    std::process::exit(0);
+}
+
+fn watchdog_interval() -> Duration {
+    if cfg!(debug_assertions)
+        && let Some(milliseconds) = std::env::var("FFF_MCP_TEST_WATCHDOG_INTERVAL_MS")
+            .ok()
+            .and_then(|value| value.parse().ok())
+    {
+        return Duration::from_millis(milliseconds);
+    }
+    Duration::from_secs(60)
+}
@@ -0,0 +1,99 @@
+#[cfg(unix)]
+mod imp {
+    pub struct ParentWatcher {
+        ppid: u32,
+    }
+
+    impl ParentWatcher {
+        pub fn new() -> Option<Self> {
+            let ppid = std::os::unix::process::parent_id();
+            // ppid <= 1 means we were spawned by init and can't detect death
+            (ppid > 1).then_some(Self { ppid })
+        }
+
+        pub fn parent_pid(&self) -> u32 {
+            self.ppid
+        }
+
+        // When the parent dies the kernel reparents us, so getppid() changes.
+        // Race-free and immune to PID reuse, unlike kill(ppid, 0).
+        pub fn parent_alive(&self) -> bool {
+            std::os::unix::process::parent_id() == self.ppid
+        }
+    }
+}
+
+#[cfg(windows)]
+mod imp {
+    use windows_sys::Win32::Foundation::{CloseHandle, HANDLE, INVALID_HANDLE_VALUE, WAIT_TIMEOUT};
+    use windows_sys::Win32::System::Diagnostics::ToolHelp::{
+        CreateToolhelp32Snapshot, PROCESSENTRY32, Process32First, Process32Next, TH32CS_SNAPPROCESS,
+    };
+    use windows_sys::Win32::System::Threading::{
+        GetCurrentProcessId, OpenProcess, PROCESS_SYNCHRONIZE, WaitForSingleObject,
+    };
+
+    pub struct ParentWatcher {
+        handle: HANDLE,
+        ppid: u32,
+    }
+
+    // HANDLE is a raw pointer; it is only ever used via WaitForSingleObject
+    // which is thread-safe, so moving/sharing the watcher across threads is fine.
+    unsafe impl Send for ParentWatcher {}
+    unsafe impl Sync for ParentWatcher {}
+
+    impl ParentWatcher {
+        pub fn new() -> Option<Self> {
+            let ppid = parent_pid_of_current()?;
+            let handle = unsafe { OpenProcess(PROCESS_SYNCHRONIZE, 0, ppid) };
+            if handle.is_null() {
+                return None;
+            }
+            // Holding the handle pins the PID, preventing reuse for the process lifetime
+            Some(Self { handle, ppid })
+        }
+
+        pub fn parent_pid(&self) -> u32 {
+            self.ppid
+        }
+
+        pub fn parent_alive(&self) -> bool {
+            unsafe { WaitForSingleObject(self.handle, 0) == WAIT_TIMEOUT }
+        }
+    }
+
+    impl Drop for ParentWatcher {
+        fn drop(&mut self) {
+            unsafe { CloseHandle(self.handle) };
+        }
+    }
+
+    fn parent_pid_of_current() -> Option<u32> {
+        unsafe {
+            let snapshot = CreateToolhelp32Snapshot(TH32CS_SNAPPROCESS, 0);
+            if snapshot == INVALID_HANDLE_VALUE {
+                return None;
+            }
+            let mut entry: PROCESSENTRY32 = std::mem::zeroed();
+            entry.dwSize = std::mem::size_of::<PROCESSENTRY32>() as u32;
+            let current = GetCurrentProcessId();
+            let mut found = None;
+            if Process32First(snapshot, &mut entry) != 0 {
+                loop {
+                    if entry.th32ProcessID == current {
+                        found = Some(entry.th32ParentProcessID);
+                        break;
+                    }
+                    if Process32Next(snapshot, &mut entry) == 0 {
+                        break;
+                    }
+                }
+            }
+            CloseHandle(snapshot);
+            found
+        }
+    }
+}
+
+pub use imp::ParentWatcher;
@@ -7,7 +7,6 @@ use fff_query_parser::AiGrepConfig;
 use rmcp::handler::server::wrapper::Parameters;
 use rmcp::model::*;
 use rmcp::{ServerHandler, schemars, tool, tool_handler, tool_router};
-use std::borrow::Cow;
 use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
 use std::sync::{Arc, Mutex};
 use std::time::{SystemTime, UNIX_EPOCH};
@@ -421,7 +420,12 @@ impl FffServer {
    /// IMPORTANT: Keep queries SHORT — prefer 1-2 terms max.
    #[tool(
        name = "find_files",
-        description = "Fuzzy file search by name. Searches FILE NAMES, not file contents. Use it when you need to find a file, not a definition. Use grep instead for searching code content (definitions, usage patterns). Supports fuzzy matching, path prefixes ('src/'), and glob constraints ('name **/src/*.{ts,tsx} !test/'). IMPORTANT: Keep queries SHORT — prefer 1-2 terms max. Multiple words are a waterfall (each narrows results), NOT OR. If unsure, start broad with 1 term and refine."
+        description = "Fuzzy file search by name. Searches FILE NAMES, not file contents. Use it when you need to find a file, not a definition. Use grep instead for searching code content (definitions, usage patterns). Supports fuzzy matching, path prefixes ('src/'), and glob constraints ('name **/src/*.{ts,tsx} !test/'). IMPORTANT: Keep queries SHORT — prefer 1-2 terms max. Multiple words are a waterfall (each narrows results), NOT OR. If unsure, start broad with 1 term and refine.",
+        annotations(
+            read_only_hint = true,
+            destructive_hint = false,
+            open_world_hint = false
+        )
    )]
    fn find_files(
        &self,
@@ -536,7 +540,12 @@ impl FffServer {
    /// Prefer plain text over regex. Filter files with constraints.
    #[tool(
        name = "grep",
-        description = "Search file contents. Search for bare identifiers (e.g. 'InProgressQuote', 'ActorAuth'), NOT code syntax or regex. Filter files with constraints (e.g. '*.rs query', 'src/ query'). Use filename, directory (ending with /) or glob expressions to prefilter. See server instructions for constraint syntax and core rules."
+        description = "Search file contents. Search for bare identifiers (e.g. 'InProgressQuote', 'ActorAuth'), NOT code syntax or regex. Filter files with constraints (e.g. '*.rs query', 'src/ query'). Use filename, directory (ending with /) or glob expressions to prefilter. See server instructions for constraint syntax and core rules.",
+        annotations(
+            read_only_hint = true,
+            destructive_hint = false,
+            open_world_hint = false
+        )
    )]
    fn grep(
        &self,
@@ -573,7 +582,12 @@ impl FffServer {
    /// Patterns are literal text — NEVER escape special characters.
    #[tool(
        name = "multi_grep",
-        description = "Search file contents for lines matching ANY of multiple patterns (OR logic). IMPORTANT: This returns files where ANY query matches, NOT all patterns. Patterns are literal text — NEVER escape special characters (no \\( \\) \\. etc). Faster than regex alternation for literal text. See server instructions for constraint syntax."
+        description = "Search file contents for lines matching ANY of multiple patterns (OR logic). IMPORTANT: This returns files where ANY query matches, NOT all patterns. Patterns are literal text — NEVER escape special characters (no \\( \\) \\. etc). Faster than regex alternation for literal text. See server instructions for constraint syntax.",
+        annotations(
+            read_only_hint = true,
+            destructive_hint = false,
+            open_world_hint = false
+        )
    )]
    fn multi_grep(
        &self,
@@ -613,61 +627,12 @@ impl FffServer {
            .ok_or_else(|| ErrorData::internal_error("File picker not initialized", None))?;
        let patterns_refs: Vec<&str> = params.patterns.iter().map(|s| s.as_str()).collect();

-        let parser = fff_query_parser::QueryParser::new(fff_query_parser::AiGrepConfig);
-        let parsed_constraints = parser.parse(constraint_query);
-        let constraints = parsed_constraints.constraints.as_slice();
+        let parser = QueryParser::new(AiGrepConfig);
+        let constraints = parser.parse_constraints(constraint_query);

-        let result = picker.multi_grep(&patterns_refs, constraints, &options);
+        let result = picker.multi_grep(&patterns_refs, &constraints, &options);
        let file_refs: Vec<&FileItem> = result.files.to_vec();

-        if result.matches.is_empty() && file_offset == 0 {
-            // Fallback: try individual patterns with plain grep
-            let (fallback_options, _) =
-                make_grep_options(output_mode, GrepMode::PlainText, 0, context);
-
-            let fallback_options = GrepSearchOptions {
-                time_budget_ms: 3000,
-                before_context: 0,
-                ..fallback_options
-            };
-
-            for pat in &params.patterns {
-                let full_query: Cow<str> = if !constraint_query.is_empty() {
-                    Cow::Owned(format!("{} {}", constraint_query, pat))
-                } else {
-                    Cow::Borrowed(pat)
-                };
-
-                let parsed = parser.parse(&full_query);
-                let fb_result = picker.grep(&parsed, &fallback_options);
-
-                if !fb_result.matches.is_empty() {
-                    let fb_file_refs: Vec<&FileItem> = fb_result.files.to_vec();
-                    let mut cs = self.lock_cursors()?;
-                    let text = &GrepFormatter {
-                        matches: &fb_result.matches,
-                        files: &fb_file_refs,
-                        total_matched: fb_result.matches.len(),
-                        next_file_offset: fb_result.next_file_offset,
-                        output_mode,
-                        max_results,
-                        show_context: false,
-                        auto_expand_defs: auto_expand,
-                        picker,
-                    }
-                    .format(&mut cs);
-                    return Ok(CallToolResult::success(vec![Content::text(format!(
-                        "0 multi-pattern matches. Plain grep fallback for \"{}\":\n{}",
-                        pat, text
-                    ))]));
-                }
-            }
-
-            return Ok(CallToolResult::success(vec![Content::text(
-                "0 matches.".to_string(),
-            )]));
-        }
-
        if result.matches.is_empty() {
            return Ok(CallToolResult::success(vec![Content::text(
                "0 matches.".to_string(),
@@ -1,6 +1,6 @@
 use std::sync::OnceLock;

-const REPO: &str = "dmtrKovalenko/fff.nvim";
+const REPO: &str = "dmtrKovalenko/fff";
 const CURRENT_VERSION: &str = env!("CARGO_PKG_VERSION");

 static UPDATE_NOTICE: OnceLock<String> = OnceLock::new();
@@ -0,0 +1,181 @@
+use std::io::{BufRead, BufReader, Write};
+use std::process::{Child, ChildStdin, Command, Stdio};
+use std::sync::mpsc;
+use std::time::{Duration, Instant};
+
+const BIN: &str = env!("CARGO_BIN_EXE_fff-mcp");
+
+#[test]
+fn stays_alive_while_parent_alive_despite_idle_timeout() {
+    let dir = tempfile::tempdir().unwrap();
+    std::fs::write(dir.path().join("a.txt"), "hello").unwrap();
+
+    let mut child = Command::new(BIN)
+        .arg(dir.path())
+        .args([
+            "--no-update-check",
+            "--no-warmup",
+            "--no-watch",
+            "--idle-timeout-secs",
+            "1",
+        ])
+        .arg("--log-file")
+        .arg(dir.path().join("test.log"))
+        .env("FFF_MCP_TEST_WATCHDOG_INTERVAL_MS", "100")
+        .stdin(Stdio::piped())
+        .stdout(Stdio::piped())
+        .stderr(Stdio::null())
+        .spawn()
+        .unwrap();
+
+    let mut stdin = child.stdin.take().unwrap();
+    let stdout_lines = spawn_line_reader(child.stdout.take().unwrap());
+    do_handshake(&mut stdin, &stdout_lines);
+
+    // Wait past the idle timeout and several watchdog ticks.
+    std::thread::sleep(Duration::from_secs(2));
+    assert!(
+        child.try_wait().unwrap().is_none(),
+        "fff-mcp exited on idle timeout even though its parent is alive"
+    );
+
+    // Closing stdin ends the transport; the server must still shut down cleanly.
+    drop(stdin);
+    wait_for_exit(&mut child, Duration::from_secs(15));
+}
+
+#[cfg(unix)]
+#[test]
+fn exits_when_parent_dies_even_without_idle_timeout() {
+    let dir = tempfile::tempdir().unwrap();
+    std::fs::write(dir.path().join("a.txt"), "hello").unwrap();
+    let log_path = dir.path().join("test.log");
+    let exit_signal = dir.path().join("exit-parent");
+
+    // Intermediary parent: sh backgrounds fff-mcp and waits until the handshake
+    // completes before dying and orphaning it.
+    let mut sh = Command::new("sh")
+        .arg("-c")
+        .arg(
+            // Preserve stdin before POSIX shells assign /dev/null to background jobs.
+            r#"exec 3<&0
+                "$1" "$2" --no-update-check --no-warmup --no-watch \
+                --idle-timeout-secs 0 --log-file "$3" <&3 &
+                while [ ! -e "$4" ]; do sleep 0.1; done"#,
+        )
+        .arg("sh")
+        .arg(BIN)
+        .arg(dir.path())
+        .arg(&log_path)
+        .arg(&exit_signal)
+        .env("FFF_MCP_TEST_WATCHDOG_INTERVAL_MS", "100")
+        .stdin(Stdio::piped())
+        .stdout(Stdio::piped())
+        .stderr(Stdio::null())
+        .spawn()
+        .unwrap();
+
+    let mut stdin = sh.stdin.take().unwrap();
+    let stdout_lines = spawn_line_reader(sh.stdout.take().unwrap());
+    do_handshake(&mut stdin, &stdout_lines);
+
+    std::fs::write(exit_signal, "").unwrap();
+    sh.wait().unwrap();
+
+    // We still hold the stdin write end, so the only exit path is the parent
+    // liveness check. EOF on stdout means fff-mcp closed it by exiting.
+    let deadline = Instant::now() + Duration::from_secs(5);
+    loop {
+        match stdout_lines.recv_timeout(deadline.saturating_duration_since(Instant::now())) {
+            Ok(_) => continue,
+            Err(mpsc::RecvTimeoutError::Disconnected) => break,
+            Err(mpsc::RecvTimeoutError::Timeout) => {
+                panic!("fff-mcp did not exit within 5s of its parent dying")
+            }
+        }
+    }
+    drop(stdin);
+
+    let logs = read_session_logs(dir.path());
+    assert!(
+        logs.contains("Parent process") && logs.contains("exited, shutting down"),
+        "expected parent-death exit reason in logs, got:\n{}",
+        logs
+    );
+}
+
+fn do_handshake(stdin: &mut ChildStdin, stdout_lines: &mpsc::Receiver<String>) {
+    let initialize = serde_json::json!({
+        "jsonrpc": "2.0",
+        "id": 1,
+        "method": "initialize",
+        "params": {
+            "protocolVersion": "2024-11-05",
+            "capabilities": {},
+            "clientInfo": { "name": "parent-liveness-test", "version": "0.0.0" }
+        }
+    });
+    writeln!(stdin, "{}", initialize).unwrap();
+    stdin.flush().unwrap();
+
+    let response = stdout_lines
+        .recv_timeout(Duration::from_secs(30))
+        .expect("no initialize response within 30s");
+    assert!(
+        response.contains("\"serverInfo\""),
+        "unexpected initialize response: {}",
+        response
+    );
+
+    writeln!(
+        stdin,
+        "{}",
+        serde_json::json!({ "jsonrpc": "2.0", "method": "notifications/initialized" })
+    )
+    .unwrap();
+    stdin.flush().unwrap();
+}
+
+fn spawn_line_reader(stdout: std::process::ChildStdout) -> mpsc::Receiver<String> {
+    let (tx, rx) = mpsc::channel();
+    std::thread::spawn(move || {
+        for line in BufReader::new(stdout).lines() {
+            match line {
+                Ok(line) => {
+                    if tx.send(line).is_err() {
+                        break;
+                    }
+                }
+                Err(_) => break,
+            }
+        }
+    });
+    rx
+}
+
+fn wait_for_exit(child: &mut Child, timeout: Duration) {
+    let deadline = Instant::now() + timeout;
+    while Instant::now() < deadline {
+        if child.try_wait().unwrap().is_some() {
+            return;
+        }
+        std::thread::sleep(Duration::from_millis(100));
+    }
+    child.kill().ok();
+    panic!(
+        "fff-mcp did not exit within {:?} after stdin closed",
+        timeout
+    );
+}
+
+#[cfg(unix)]
+fn read_session_logs(dir: &std::path::Path) -> String {
+    let mut combined = String::new();
+    for entry in std::fs::read_dir(dir).unwrap().flatten() {
+        let name = entry.file_name().to_string_lossy().to_string();
+        if name.starts_with("test") && name.ends_with(".log") {
+            combined.push_str(&std::fs::read_to_string(entry.path()).unwrap_or_default());
+        }
+    }
+    combined
+}
@@ -1,6 +1,6 @@
 [package]
 name = "fff-nvim"
-version = "0.10.1"
+version = "0.10.4"
 edition = "2024"

 [lints]
@@ -15,6 +15,8 @@ crate-type = ["cdylib", "rlib"]
 default = ["ripgrep"]
 ripgrep = ["fff/ripgrep", "fff-query-parser/ripgrep", "dep:ignore"]
 zlob = ["fff/zlob", "fff-query-parser/zlob", "dep:zlob"]
+# Keep full-rescan accounting in a release build; required by rescan_probe.
+rescan-stats = ["fff/rescan-stats"]

 [dependencies]
 # Workspace dependencies
@@ -22,10 +24,10 @@ ahash = { workspace = true }
 tracing = { workspace = true }

 # Local crates
-fff = { package = "fff-search", path = "../fff-core", version = "0.10.1", default-features = false, features = [
+fff = { package = "fff-search", path = "../fff-core", version = "0.10.4", default-features = false, features = [
 	"mimalloc-collect",
 ] }
-fff-query-parser = { path = "../fff-query-parser", version = "0.10.1", default-features = false }
+fff-query-parser = { path = "../fff-query-parser", version = "0.10.4", default-features = false }
 chrono = { version = "0.4", features = ["serde"] }
 ctrlc = "3.4.2"
 git2 = { workspace = true }
@@ -0,0 +1,158 @@
+use fff::file_picker::FilePicker;
+use fff::{
+    FFFMode, FilePickerOptions, RESCAN_STATS_ENABLED, RescanReason, RescanStats, SharedFilePicker,
+    SharedFrecency,
+};
+use std::sync::Arc;
+use std::sync::atomic::{AtomicBool, Ordering};
+use std::time::{Duration, Instant};
+
+const POLL: Duration = Duration::from_millis(250);
+
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    let (base_path, run_for) = parse_args()?;
+
+    if !RESCAN_STATS_ENABLED {
+        return Err(
+            "this build has rescan accounting compiled out; rebuild with \
+                    `--features rescan-stats` (or drop `--release`)"
+                .into(),
+        );
+    }
+
+    let picker = SharedFilePicker::default();
+    let frecency = SharedFrecency::noop();
+
+    println!("indexing {base_path} ...");
+    let started = Instant::now();
+    FilePicker::new_with_shared_state(
+        picker.clone(),
+        frecency.clone(),
+        FilePickerOptions {
+            base_path: base_path.clone(),
+            enable_mmap_cache: false,
+            mode: FFFMode::default(),
+            watch: true,
+            ..Default::default()
+        },
+    )?;
+
+    if !picker.wait_for_scan(Duration::from_secs(600)) {
+        return Err("timed out waiting for the initial scan".into());
+    }
+    if !picker.wait_for_watcher(Duration::from_secs(600)) {
+        return Err("timed out waiting for the watcher".into());
+    }
+
+    println!(
+        "indexed {} files in {:.2}s; watching for rescan requests.\n",
+        live_files(&picker),
+        started.elapsed().as_secs_f64()
+    );
+    picker.reset_rescan_stats();
+
+    let running = Arc::new(AtomicBool::new(true));
+    let stop = Arc::clone(&running);
+    ctrlc::set_handler(move || stop.store(false, Ordering::SeqCst))?;
+
+    let watching_since = Instant::now();
+    let mut last = RescanStats::default();
+
+    while running.load(Ordering::SeqCst) {
+        std::thread::sleep(POLL);
+
+        let stats = picker.rescan_stats();
+        let delta = stats.since(&last);
+        if delta.total > 0 || delta.throttled > 0 {
+            let now = watching_since.elapsed().as_secs_f64();
+            let files = live_files(&picker);
+            let overflow = overflow_files(&picker);
+
+            for reason in RescanReason::ALL {
+                for _ in 0..delta.count(reason) {
+                    println!(
+                        "[{now:>8.2}s] request   {reason:<21} files={files} overflow={overflow}"
+                    );
+                }
+                let suppressed = delta.count_throttled(reason);
+                if suppressed > 0 {
+                    println!("[{now:>8.2}s] throttled {reason:<21} x{suppressed}");
+                }
+            }
+            last = stats;
+        }
+
+        if run_for.is_some_and(|limit| watching_since.elapsed() >= limit) {
+            break;
+        }
+    }
+
+    let elapsed = watching_since.elapsed();
+    let stats = picker.rescan_stats();
+    println!("\n{:.1}s watched", elapsed.as_secs_f64());
+    println!("{stats}");
+    if stats.watcher_triggered() > 0 {
+        println!(
+            "{:.1} watcher rescan requests/minute",
+            stats.watcher_triggered() as f64 / elapsed.as_secs_f64().max(1.0) * 60.0
+        );
+    } else {
+        println!("no full rescans: every change was applied incrementally");
+    }
+    if stats.throttled > 0 {
+        println!(
+            "{} additional request(s) were throttled; {} total requests observed",
+            stats.throttled,
+            stats.total + stats.throttled
+        );
+    }
+
+    if let Ok(mut guard) = picker.write() {
+        guard.take();
+    }
+
+    Ok(())
+}
+
+fn parse_args() -> Result<(String, Option<Duration>), Box<dyn std::error::Error>> {
+    let mut base_path = None;
+    let mut run_for = None;
+    let mut args = std::env::args().skip(1);
+
+    while let Some(arg) = args.next() {
+        match arg.as_str() {
+            "--seconds" | "-s" => {
+                let value = args.next().ok_or("--seconds needs a value")?;
+                run_for = Some(Duration::from_secs(value.parse()?));
+            }
+            "--help" | "-h" => {
+                println!("usage: rescan_probe [path] [--seconds N]");
+                std::process::exit(0);
+            }
+            other => base_path = Some(other.to_string()),
+        }
+    }
+
+    let base_path = match base_path {
+        Some(path) => path,
+        None => std::env::current_dir()?.to_string_lossy().into_owned(),
+    };
+
+    Ok((base_path, run_for))
+}
+
+fn live_files(picker: &SharedFilePicker) -> usize {
+    picker
+        .read()
+        .ok()
+        .and_then(|g| g.as_ref().map(|p| p.live_file_count()))
+        .unwrap_or(0)
+}
+
+fn overflow_files(picker: &SharedFilePicker) -> usize {
+    picker
+        .read()
+        .ok()
+        .and_then(|g| g.as_ref().map(|p| p.get_overflow_files().len()))
+        .unwrap_or(0)
+}
@@ -1,6 +1,6 @@
 [package]
 name = "fff-python"
-version = "0.10.1"
+version = "0.10.4"
 edition = "2024"

 [lints]
@@ -17,7 +17,7 @@ ripgrep = ["fff/ripgrep", "fff-query-parser/ripgrep"]
 zlob = ["fff/zlob", "fff-query-parser/zlob"]

 [dependencies]
-fff = { package = "fff-search", path = "../fff-core", version = "0.10.1", default-features = false }
-fff-query-parser = { path = "../fff-query-parser", version = "0.10.1", default-features = false }
+fff = { package = "fff-search", path = "../fff-core", version = "0.10.4", default-features = false }
+fff-query-parser = { path = "../fff-query-parser", version = "0.10.4", default-features = false }
 git2 = { workspace = true }
 pyo3 = { version = "0.24.0", features = ["extension-module", "abi3-py310"] }
@@ -693,15 +693,11 @@ impl FileFinder {
            }
            let pattern_refs: Vec<&str> = patterns.iter().map(|s| s.as_str()).collect();

-            let parsed_constraints = constraints.as_ref().map(|c| {
-                if picker.mode().is_ai() {
-                    QueryParser::new(fff_query_parser::AiGrepConfig).parse(c)
-                } else {
-                    fff::grep::parse_grep_query(c)
-                }
-            });
+            let parsed_constraints = constraints
+                .as_ref()
+                .map(|c| QueryParser::new(fff_query_parser::AiGrepConfig).parse_constraints(c));
            let constraint_refs: &[fff::Constraint<'_>] = match &parsed_constraints {
-                Some(q) => &q.constraints,
+                Some(constraints) => constraints,
                None => &[],
            };
            let options = grep_options(
@@ -1,6 +1,6 @@
 [package]
 name = "fff-query-parser"
-version = "0.10.1"
+version = "0.10.4"
 edition = "2024"
 description = "Query parser for fff file finder - includes specific syntax for various constraints like globs, extensions, regex etc"
 license = "MIT"
@@ -49,6 +49,14 @@ impl<C: ParserConfig> QueryParser<C> {
        Self { config }
    }

+    /// Parse a field containing only constraints.
+    pub fn parse_constraints<'a>(&self, query: &'a str) -> ConstraintVec<'a> {
+        query
+            .split_whitespace()
+            .filter_map(|token| parse_token(token, &self.config))
+            .collect()
+    }
+
    pub fn parse<'a>(&self, query: &'a str) -> FFFQuery<'a> {
        let raw_query = query;
        let config: &C = &self.config;
@@ -498,7 +506,7 @@ fn parse_git_status(value: &str) -> Option<Constraint<'_>> {
 #[cfg(test)]
 mod tests {
    use super::*;
-    use crate::{FileSearchConfig, GrepConfig};
+    use crate::{AiGrepConfig, FileSearchConfig, GrepConfig};

    /// File-picker-like config with filename-constraint detection enabled,
    /// mirroring the Neovim layer's opt-in behavior.
@@ -1019,6 +1027,42 @@ mod tests {
        assert_eq!(result.grep_text(), "pattern");
    }

+    #[test]
+    fn test_standalone_constraints_preserve_directory() {
+        let result = QueryParser::new(AiGrepConfig).parse_constraints("scope-a/");
+        assert_eq!(result.as_slice(), &[Constraint::PathSegment("scope-a")]);
+    }
+
+    #[test]
+    fn test_plain_constraints_preserve_directory_only() {
+        let directory = QueryParser::new(GrepConfig).parse_constraints("scope-a/");
+        assert_eq!(directory.as_slice(), &[Constraint::PathSegment("scope-a")]);
+
+        let file = QueryParser::new(GrepConfig).parse_constraints("scope-a/one.txt");
+        assert!(file.is_empty());
+    }
+
+    #[test]
+    fn test_standalone_constraints_preserve_file() {
+        let result = QueryParser::new(AiGrepConfig).parse_constraints("scope-a/one.txt");
+        assert_eq!(
+            result.as_slice(),
+            &[Constraint::FilePath("scope-a/one.txt")]
+        );
+    }
+
+    #[test]
+    fn test_standalone_constraints_preserve_file_without_search_text() {
+        let result = QueryParser::new(AiGrepConfig).parse_constraints("scope-a/ scope-a/one.txt");
+        assert_eq!(
+            result.as_slice(),
+            &[
+                Constraint::PathSegment("scope-a"),
+                Constraint::FilePath("scope-a/one.txt")
+            ]
+        );
+    }
+
    #[test]
    fn test_ai_grep_filename_with_pathsegment_only_promotes_to_text() {
        // When the ONLY non-text constraints are path-scoping (PathSegment,
@@ -24,8 +24,9 @@ INSTALLATION ~
 LAZY.NVIM

 >lua
+    -- Package name changed from `fff.nvim` to `fff`. If you installed fff.nvim before, clean with `:Lazy clean`
    {
-      'dmtrKovalenko/fff.nvim',
+      'dmtrKovalenko/fff',
      build = function()
        -- downloads a prebuilt binary or falls back to cargo build
        require("fff.download").download_or_build_binary()
@@ -59,13 +60,14 @@ LAZY.NVIM
 VIM.PACK

 >lua
-    vim.pack.add({ 'https://github.com/dmtrKovalenko/fff.nvim' })
+    -- Package name changed from `fff.nvim` to `fff`. If you installed fff.nvim before, clean with `:packdel fff.nvim`
+    vim.pack.add({ 'https://github.com/dmtrKovalenko/fff' })
    
    vim.api.nvim_create_autocmd('PackChanged', {
      callback = function(ev)
        local name, kind = ev.data.spec.name, ev.data.kind
-        if name == 'fff.nvim' and (kind == 'install' or kind == 'update') then
-          if not ev.data.active then vim.cmd.packadd('fff.nvim') end
+        if name == 'fff' and (kind == 'install' or kind == 'update') then
+          if not ev.data.active then vim.cmd.packadd('fff') end
          require('fff.download').download_or_build_binary()
        end
      end,
@@ -1,5 +1,5 @@
-- Single file Neovim config for testing fff.nvim locally
-- Usage: nvim -u /Users/neogoose/dev/fff.nvim/init.lua
+-- Single file Neovim config for testing fff locally
+-- Usage: nvim -u /Users/neogoose/dev/fff/init.lua

 -- Set up lazy.nvim plugin manager
 local lazypath = vim.fn.stdpath('data') .. '/lazy/lazy.nvim'
@@ -17,8 +17,8 @@ vim.opt.rtp:prepend(lazypath)

 require('lazy').setup({
  {
-    dir = '~/dev/fff.nvim',
-    'https://github.com/dmtrKovalenko/fff.nvim',
+    dir = '~/dev/fff',
+    'https://github.com/dmtrKovalenko/fff',
    build = function()
      -- this will download prebuild binary or try to use existing rustup toolchain to build from source
      -- (if you are using lazy you can use gb for rebuilding a plugin if needed)
@@ -34,7 +34,7 @@ require('lazy').setup({
    },
    config = function()
      require('fff').setup({
-        -- Configure fff.nvim here
+        -- Configure fff here
        ui = {
          width = 0.8,
          height = 0.8,
@@ -59,4 +59,4 @@ vim.keymap.set('n', 'fg', function() require('fff').find_in_git_root() end, { de
 vim.keymap.set('n', 'fr', function() require('fff').scan_files() end, { desc = 'Rescan files' })
 vim.keymap.set('n', 'fs', function() require('fff').refresh_git_status() end, { desc = 'Refresh git status' })

-vim.notify('FFF.nvim local config loaded! Press ff', vim.log.levels.INFO)
+vim.notify('FFF local config loaded! Press ff', vim.log.levels.INFO)
@@ -2,11 +2,11 @@
  "nodes": {
    "crane": {
      "locked": {
-        "lastModified": 1773857772,
-        "narHash": "sha256-5xsK26KRHf0WytBtsBnQYC/lTWDhQuT57HJ7SzuqZcM=",
+        "lastModified": 1785782307,
+        "narHash": "sha256-MPaRdVkf6zZP5fCPxYCi8Dr4pZzgmXzg8T9nVEbp3Mw=",
        "owner": "ipetkov",
        "repo": "crane",
-        "rev": "b556d7bbae5ff86e378451511873dfd07e4504cd",
+        "rev": "2c71e194474d13de031d729b729c968ddbe3507f",
        "type": "github"
      },
      "original": {
@@ -51,11 +51,11 @@
    },
    "nixpkgs": {
      "locked": {
-        "lastModified": 1776329215,
-        "narHash": "sha256-a8BYi3mzoJ/AcJP8UldOx8emoPRLeWqALZWu4ZvjPXw=",
+        "lastModified": 1786593342,
+        "narHash": "sha256-smTKQXMLLStzc8zJevMCckbk3My7SvbbLmPYZUJJKW4=",
        "owner": "NixOS",
        "repo": "nixpkgs",
-        "rev": "b86751bc4085f48661017fa226dee99fab6c651b",
+        "rev": "6b5e5b7a6631f065bf6908986990b37d845f847f",
        "type": "github"
      },
      "original": {
@@ -81,11 +81,11 @@
        ]
      },
      "locked": {
-        "lastModified": 1773803479,
-        "narHash": "sha256-GD6i1F2vrSxbsmbS92+8+x3DbHOJ+yrS78Pm4xigW4M=",
+        "lastModified": 1786762605,
+        "narHash": "sha256-iQpYIhInh8gRx+cSnPtX+Yp2Gg9kr2h+MehETCqRgDo=",
        "owner": "oxalica",
        "repo": "rust-overlay",
-        "rev": "f17186f52e82ec5cf40920b58eac63b78692ac7c",
+        "rev": "ad8ebb59d84bcf3780c46f107e1d99eb4ca2fe7f",
        "type": "github"
      },
      "original": {
@@ -134,11 +134,11 @@
        "systems": "systems_2"
      },
      "locked": {
-        "lastModified": 1776789209,
-        "narHash": "sha256-G6B7Q4TXn7MZ1mB+f9rymjsYF5PLWoSvmbxijb/99bw=",
+        "lastModified": 1786753251,
+        "narHash": "sha256-7UGNbG2MwgYCdsknKBL6wMXQQeiTmCiz+JS9zvcmqv4=",
        "owner": "mitchellh",
        "repo": "zig-overlay",
-        "rev": "14fe971844e841297ddd2ce9783d6892b467af39",
+        "rev": "a07cc99cee149044f27d95500717161fca2afdf6",
        "type": "github"
      },
      "original": {
@@ -1,5 +1,5 @@
 {
-  description = "fff.nvim";
+  description = "fff";

  inputs = {
    nixpkgs.url = "github:NixOS/nixpkgs/nixpkgs-unstable";
@@ -4,7 +4,7 @@
    FFF MCP Server installer for Windows.
 .DESCRIPTION
    Pipe usage:
-        irm https://raw.githubusercontent.com/dmtrKovalenko/fff.nvim/main/install-mcp.ps1 | iex
+        irm https://raw.githubusercontent.com/dmtrKovalenko/fff/main/install-mcp.ps1 | iex
    Direct usage (supports params):
        iwr https://.../install-mcp.ps1 -OutFile install-mcp.ps1; .\install-mcp.ps1 -Version v0.1.2
    Env-var fallbacks (for the piped form):
@@ -29,7 +29,7 @@ $ErrorActionPreference = 'Stop'
 # Force TLS 1.2 — PS 5.1 on older Win10 may default to SSL3/TLS1.0 which GitHub rejects.
 [Net.ServicePointManager]::SecurityProtocol = [Net.ServicePointManager]::SecurityProtocol -bor [Net.SecurityProtocolType]::Tls12

-$Repo = 'dmtrKovalenko/fff.nvim'
+$Repo = 'dmtrKovalenko/fff'
 $BinaryName = 'fff-mcp'
 if (-not $InstallDir) { $InstallDir = Join-Path $env:LOCALAPPDATA 'fff-mcp\bin' }

@@ -2,20 +2,20 @@
 set -eo pipefail

 # FFF MCP Server installer
-# Usage: curl -fsSL https://raw.githubusercontent.com/dmtrKovalenko/fff.nvim/main/install-mcp.sh | bash
+# Usage: curl -fsSL https://raw.githubusercontent.com/dmtrKovalenko/fff/main/install-mcp.sh | bash

-REPO="dmtrKovalenko/fff.nvim"
+REPO="dmtrKovalenko/fff"
 BINARY_NAME="fff-mcp"
 INSTALL_DIR="${FFF_MCP_INSTALL_DIR:-$HOME/.local/bin}"

-PINNED_RELEASE_TAG="v0.10.1"
+PINNED_RELEASE_TAG="v0.10.4"

-SHA256_X86_64_UNKNOWN_LINUX_MUSL="c17637c333afbbbdea4b03cf3e1573240c4147ae121756e363aafa3c9d0efb58"
-SHA256_AARCH64_UNKNOWN_LINUX_MUSL="2a25019101ee9373327dabd7ac1d4800638688e95b67bfb3bec441bf8d938f28"
-SHA256_X86_64_APPLE_DARWIN="d3c8d70d47ab80afa2287e1b4546b23f494bb4604b80e5a706a0af714ee25674"
-SHA256_AARCH64_APPLE_DARWIN="eff6660a9c48e3e1972d557c1003e057e5ff99d6039f5f810671f2123093fdfc"
-SHA256_X86_64_PC_WINDOWS_MSVC="e341b78464095c349b0c6b0a32b146fd217b542d973917b89645a5aa511640d8"
-SHA256_AARCH64_PC_WINDOWS_MSVC="c2fc1e55395f314a9493340f11f9b75501a6cd2c1e8fea646d343c7eaf3cd556"
+SHA256_X86_64_UNKNOWN_LINUX_MUSL="68384121a9173b214190c594ea938cb5a1c52ca3069a3f3b1a0eecc28fd8ea82"
+SHA256_AARCH64_UNKNOWN_LINUX_MUSL="334ff6f28d5ee47b9652a6c590cf47b908541ff25cb1bd3c7c3303d5f5342c40"
+SHA256_X86_64_APPLE_DARWIN="821aedd5873292841f32d7584eb68fea085b248e5c970ff5d4f6dca790e9b18c"
+SHA256_AARCH64_APPLE_DARWIN="f55ce2945cd30fd252a4b1e7d8674ff7ce41b104cc009a8d3d52d08bfccca2c0"
+SHA256_X86_64_PC_WINDOWS_MSVC="2ff065dc7636c3d50d93a6ea55209760ed382bb6c096c01a3ba7357d6fa6b59e"
+SHA256_AARCH64_PC_WINDOWS_MSVC="42244ce562fec2ff13d97d819430e028f1f2e5a0d08418f9bd24f97d060741eb"

 expected_sha_for() {
    case "$1" in
@@ -3,6 +3,8 @@ if not fuzzy then error('Failed to load fff.fuzzy module. Ensure the Rust backen

 local M = {}

+local fs_scanning_refusal
+
 ---@class fff.core.State
 local state = {
  ---@type boolean
@@ -111,6 +113,13 @@ M.change_indexing_directory = function(new_path)

  local fff_rust = M.ensure_initialized()
  local config = require('fff.conf').get()
+
+  local refusal = fs_scanning_refusal(vim.tbl_extend('force', config, { base_path = expanded_path }))
+  if refusal then
+    vim.notify('FFF: ' .. refusal, vim.log.levels.WARN)
+    return false
+  end
+
  local ok, err = pcall(fff_rust.restart_index_in_path, expanded_path, {
    follow_symlinks = config.follow_symlinks,
    enable_fs_root_scanning = config.enable_fs_root_scanning,
@@ -126,51 +135,91 @@ M.change_indexing_directory = function(new_path)
  return true
 end

+--- Reset the file-picker flag so the next `ensure_initialized` recreates the
+--- Rust picker. Call after `cleanup_file_picker` drops it (`FFFClearCache`);
+--- otherwise the flag stays set and every later call operates on a dropped
+--- picker (see #772).
+M.mark_file_picker_uninitialized = function() state.file_picker_initialized = false end
+
 M.ensure_initialized = function()
-  if state.initialized then return fuzzy end
-  state.initialized = true
-
  local config = require('fff.conf').get()
-  if config.logging.enabled then
-    local log_success, log_error =
-      pcall(fuzzy.init_tracing, config.logging.log_file, config.logging.log_level, config.logging.retain_runs)
-    if log_success then
-      M.log_file_path = log_error
-    else
-      vim.notify('Failed to initialize logging: ' .. (tostring(log_error) or 'unknown error'), vim.log.levels.WARN)
-    end
-  end

-  local frecency_db_path = config.frecency.db_path or (vim.fn.stdpath('cache') .. '/fff_frecency')
-  local history_db_path = config.history.db_path or (vim.fn.stdpath('data') .. '/fff_history')
-
-  local ok, result = pcall(fuzzy.init_db, frecency_db_path, history_db_path, true)
-  if not ok then vim.notify('Failed to databases: ' .. tostring(result), vim.log.levels.WARN) end
-
-  ok, result = pcall(fuzzy.init_file_picker, config.base_path, {
-    follow_symlinks = config.follow_symlinks,
-    enable_fs_root_scanning = config.enable_fs_root_scanning,
-    enable_home_dir_scanning = config.enable_home_dir_scanning,
-    enable_filename_constraint = config.grep and config.grep.enable_filename_constraint,
-  })
-  if not ok then
-    vim.notify('Failed to initialize file picker: ' .. tostring(result), vim.log.levels.ERROR)
+  -- Refusal gates both one-time setup and (re)creating the picker so we never
+  -- index fs-root / home, even after a cache clear.
+  -- Some folks are complaining that neovim instance is closing if ffi returns error on startup (via lazy=false)
+  -- I can't repro so just precheck on lua side to prevent crashing neovim instance
+  local refusal = fs_scanning_refusal(config)
+  if refusal then
+    state.initialized = true
+    vim.notify('FFF: ' .. refusal, vim.log.levels.WARN)
    return fuzzy
  end

-  state.file_picker_initialized = true
-  setup_global_autocmds(config)
+  if not state.initialized then
+    state.initialized = true
+    if config.logging.enabled then
+      local log_success, log_error =
+        pcall(fuzzy.init_tracing, config.logging.log_file, config.logging.log_level, config.logging.retain_runs)
+      if log_success then
+        M.log_file_path = log_error
+      else
+        vim.notify('Failed to initialize logging: ' .. (tostring(log_error) or 'unknown error'), vim.log.levels.WARN)
+      end
+    end

-  local highlights = require('fff.highlights')
-  highlights.setup()
+    local frecency_db_path = config.frecency.db_path or (vim.fn.stdpath('cache') .. '/fff_frecency')
+    local history_db_path = config.history.db_path or (vim.fn.stdpath('data') .. '/fff_history')

-  vim.api.nvim_create_autocmd('ColorScheme', {
-    group = vim.api.nvim_create_augroup('fff_highlights', { clear = true }),
-    callback = function() highlights.setup() end,
-    desc = 'Re-apply FFF highlights on colorscheme change',
-  })
+    local ok, result = pcall(fuzzy.init_db, frecency_db_path, history_db_path, true)
+    if not ok then vim.notify('Failed to databases: ' .. tostring(result), vim.log.levels.WARN) end
+
+    setup_global_autocmds(config)
+
+    local highlights = require('fff.highlights')
+    highlights.setup()
+
+    vim.api.nvim_create_autocmd('ColorScheme', {
+      group = vim.api.nvim_create_augroup('fff_highlights', { clear = true }),
+      callback = function() highlights.setup() end,
+      desc = 'Re-apply FFF highlights on colorscheme change',
+    })
+  end
+
+  -- Recreated whenever the picker was torn down (e.g. `FFFClearCache files`).
+  -- Guarded separately from one-time setup so a cache clear rebuilds the
+  -- picker instead of leaving a dropped one behind (#772).
+  if not state.file_picker_initialized then
+    local ok, result = pcall(fuzzy.init_file_picker, config.base_path, {
+      follow_symlinks = config.follow_symlinks,
+      enable_fs_root_scanning = config.enable_fs_root_scanning,
+      enable_home_dir_scanning = config.enable_home_dir_scanning,
+      enable_filename_constraint = config.grep and config.grep.enable_filename_constraint,
+    })
+    if not ok then
+      vim.notify('Failed to initialize file picker: ' .. tostring(result), vim.log.levels.ERROR)
+      return fuzzy
+    end
+    state.file_picker_initialized = true
+  end

  return fuzzy
 end

+function fs_scanning_refusal(config)
+  local path = vim.fn.fnamemodify(vim.fn.expand(config.base_path), ':p'):gsub('/+$', '')
+
+  if not config.enable_fs_root_scanning and (path == '' or path:match('^%a:$')) then
+    return 'Refusing to index filesystem root. Set enable_fs_root_scanning = true to override.'
+  end
+
+  if not config.enable_home_dir_scanning then
+    local home = (vim.fn.expand('$HOME') or ''):gsub('/+$', '')
+    if home ~= '' and path == home then
+      return 'Refusing to index home directory. Set enable_home_dir_scanning = true to override.'
+    end
+  end
+
+  return nil
+end
+
 return M
@@ -3,7 +3,7 @@ local system = require('fff.utils.system')
 local fs_utils = require('fff.utils.fs')
 local fff_version = require('fff.utils.version')

-local GITHUB_REPO = 'dmtrKovalenko/fff.nvim'
+local GITHUB_REPO = 'dmtrKovalenko/fff'

 local function get_binary_dir(plugin_dir) return plugin_dir .. '/../target/release' end

@@ -101,7 +101,13 @@ function M.clear_cache(scope)

  if scope == 'all' or scope == 'files' then
    local ok, err = pcall(fuzzy.cleanup_file_picker)
-    if not ok then table.insert(errors, 'cleanup file picker: ' .. tostring(err)) end
+    if not ok then
+      table.insert(errors, 'cleanup file picker: ' .. tostring(err))
+    else
+      -- Rust picker is gone; clear the core flag so the next ensure_initialized
+      -- rebuilds it instead of operating on a dropped picker (#772).
+      require('fff.core').mark_file_picker_uninitialized()
+    end
  end

  if scope == 'all' or scope == 'frecency' then
@@ -174,8 +174,8 @@ If prebuilt binaries aren't available for your platform:

 ```bash
 # Clone the repository
-git clone https://github.com/dmtrKovalenko/fff.nvim
-cd fff.nvim
+git clone https://github.com/dmtrKovalenko/fff
+cd fff

 # Build the C library
 cargo build --release -p fff-c
@@ -4,19 +4,24 @@
  "private": false,
  "description": "High-performance fuzzy file finder for Bun - perfect for LLM agent tools",
  "type": "module",
-  "main": "src/index.ts",
-  "types": "src/index.ts",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
  "exports": {
    ".": {
-      "import": "./src/index.ts",
-      "types": "./src/index.ts"
-    }
+      "types": "./dist/index.d.ts",
+      "bun": "./src/index.ts",
+      "import": "./dist/index.js",
+      "default": "./dist/index.js"
+    },
+    "./package.json": "./package.json"
  },
  "files": [
+    "dist",
    "src",
    "examples"
  ],
  "scripts": {
+    "build": "bun build ./src/index.ts --format=esm --target=bun --sourcemap=external --outdir dist && tsc -p tsconfig.build.json",
    "test": "bun test test/",
    "typecheck": "tsc --noEmit",
    "demo": "bun ./examples/search.ts"
@@ -573,16 +573,10 @@ export interface FileFinderApi {
  glob(pattern: string, options?: GlobOptions): Result<SearchResult>;

  /** Fuzzy directory search. */
-  directorySearch(
-    query: string,
-    options?: DirSearchOptions,
-  ): Result<DirSearchResult>;
+  directorySearch(query: string, options?: DirSearchOptions): Result<DirSearchResult>;

  /** Fuzzy search over files and directories interleaved by score. */
-  mixedSearch(
-    query: string,
-    options?: SearchOptions,
-  ): Result<MixedSearchResult>;
+  mixedSearch(query: string, options?: SearchOptions): Result<MixedSearchResult>;

  /** Content search (live grep). */
  grep(query: string, options?: GrepOptions): Result<GrepResult>;
@@ -647,10 +641,7 @@ export interface FileFinderApi {
   * Events are debounced and submitted in batches per 100-ms window at most 128 events.
   * Gitignored and other ignored files are never triggering watcher.
   */
-  watch(
-    callback: WatchBatchCallback,
-    options?: WatchOptions,
-  ): Result<WatchUnsubscribe>;
+  watch(callback: WatchBatchCallback, options?: WatchOptions): Result<WatchUnsubscribe>;
  watch(
    pattern: string,
    callback: WatchBatchCallback,
@@ -0,0 +1,230 @@
+import { afterEach, describe, expect, mock, test } from "bun:test";
+import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { FileFinder } from "../src/index";
+
+// Real-native tests for #700/#760. Lives here, not in pi-fff/test: that suite
+// mocks @ff-labs/fff-bun process-globally and bun module mocks can't be undone.
+mock.module("@earendil-works/pi-tui", () => ({
+  Text: class Text {
+    text: string;
+    constructor(text: string) {
+      this.text = text;
+    }
+    setText(text: string) {
+      this.text = text;
+    }
+  },
+}));
+
+const schema = (type: string) => (options?: unknown) => ({ type, options });
+mock.module("@sinclair/typebox", () => ({
+  Type: {
+    Array: (items: unknown, options?: unknown) => ({ type: "array", items, options }),
+    Boolean: schema("boolean"),
+    Number: schema("number"),
+    Object: (properties: unknown, options?: unknown) => ({
+      type: "object",
+      properties,
+      options,
+    }),
+    Optional: (value: unknown) => ({ ...(value as object), optional: true }),
+    String: schema("string"),
+    Union: (items: unknown[], options?: unknown) => ({ type: "union", items, options }),
+  },
+}));
+
+const { default: fffExtension } = await import("../../pi-fff/src/index");
+
+// Inject this package as the extension's SDK through the cache hook sdk.ts
+// already uses for reloads: CI has no node_modules to resolve "@ff-labs/fff-bun"
+// from pi-fff, and the finder stays the real native one either way.
+(globalThis as Record<string, unknown>).__fffSdkPromiseGlobal = Promise.resolve({
+  FileFinder,
+});
+
+const cleanups: Array<() => void> = [];
+
+afterEach(() => {
+  while (cleanups.length) cleanups.pop()?.();
+});
+
+function makeWorkspace(name: string): string {
+  const dir = mkdtempSync(join(tmpdir(), `pi-fff-native-${name}-`));
+  cleanups.push(() => rmSync(dir, { recursive: true, force: true }));
+  writeFileSync(join(dir, "alpha.ts"), "export const alpha = 1;\n");
+  writeFileSync(join(dir, "beta.ts"), "export const beta = 2;\n");
+  mkdirSync(join(dir, "src"));
+  writeFileSync(join(dir, "src", "gamma.ts"), "export const gamma = 3;\n");
+  return dir;
+}
+
+function makeDbPaths() {
+  const root = mkdtempSync(join(tmpdir(), "pi-fff-native-dbs-"));
+  cleanups.push(() => rmSync(root, { recursive: true, force: true }));
+  return { frecencyDbPath: join(root, "frecency"), historyDbPath: join(root, "history") };
+}
+
+function createFinder(options: Parameters<typeof FileFinder.create>[0]) {
+  const result = FileFinder.create(options);
+  if (result.ok) {
+    const finder = result.value;
+    cleanups.push(() => {
+      if (!finder.isDestroyed) finder.destroy();
+    });
+  }
+  return result;
+}
+
+type SearchOk = { ok: true; value: { items: Array<{ fileName: string }> } };
+type SearchResult = SearchOk | { ok: false; error: string };
+
+function fileNames(
+  finder: { fileSearch: (q: string, o: { pageSize: number }) => SearchResult },
+  query: string,
+): string[] {
+  const search = finder.fileSearch(query, { pageSize: 10 });
+  expect(search.ok).toBe(true);
+  return search.ok ? search.value.items.map((i) => i.fileName) : [];
+}
+
+describe("fff-bun: many finders share one LMDB env per path (#700/#760)", () => {
+  test("a second finder on the same db paths works and searches", async () => {
+    const dbs = makeDbPaths();
+    const main = createFinder({ basePath: makeWorkspace("main"), ...dbs });
+    expect(main.ok).toBe(true);
+    if (!main.ok) return;
+    await main.value.waitForScan(15_000);
+    expect(fileNames(main.value, "alpha")).toContain("alpha.ts");
+
+    // The createAgentSession scenario: same process, same db paths.
+    const sub = createFinder({ basePath: makeWorkspace("subagent"), ...dbs });
+    expect(sub.ok).toBe(true);
+    if (!sub.ok) return;
+    await sub.value.waitForScan(15_000);
+    expect(fileNames(sub.value, "gamma")).toContain("gamma.ts");
+  }, 30_000);
+
+  test("destroying one finder keeps the shared env alive for the other", async () => {
+    const dbs = makeDbPaths();
+    const first = createFinder({ basePath: makeWorkspace("first"), ...dbs });
+    const second = createFinder({ basePath: makeWorkspace("second"), ...dbs });
+    expect(first.ok).toBe(true);
+    expect(second.ok).toBe(true);
+    if (!first.ok || !second.ok) return;
+
+    first.value.destroy();
+    await second.value.waitForScan(15_000);
+    expect(fileNames(second.value, "alpha")).toContain("alpha.ts");
+
+    // And once the survivor is gone too, the paths are reusable.
+    second.value.destroy();
+    const third = createFinder({ basePath: makeWorkspace("third"), ...dbs });
+    expect(third.ok).toBe(true);
+  }, 30_000);
+
+  test("a db-less aux finder coexists with the main finder (#700)", async () => {
+    const main = createFinder({ basePath: makeWorkspace("main"), ...makeDbPaths() });
+    expect(main.ok).toBe(true);
+
+    const aux = createFinder({ basePath: makeWorkspace("aux") });
+    expect(aux.ok).toBe(true);
+    if (!aux.ok) return;
+    await aux.value.waitForScan(15_000);
+    expect(fileNames(aux.value, "gamma")).toContain("gamma.ts");
+  }, 30_000);
+});
+
+type EventHandler = (...args: unknown[]) => unknown;
+type RegisteredTool = {
+  name: string;
+  execute: (
+    toolCallId: string,
+    params: unknown,
+    signal?: AbortSignal,
+  ) => Promise<unknown>;
+};
+
+function startSession(
+  cwd: string,
+  dbs: { frecencyDbPath: string; historyDbPath: string },
+) {
+  const events = new Map<string, EventHandler>();
+  const tools = new Map<string, RegisteredTool>();
+  const notifications: Array<{ message: string; level?: string }> = [];
+
+  const flags: Record<string, unknown> = {
+    "fff-frecency-db": dbs.frecencyDbPath,
+    "fff-history-db": dbs.historyDbPath,
+  };
+
+  const pi = {
+    getFlag: (name: string) => flags[name],
+    on: (event: string, handler: EventHandler) => events.set(event, handler),
+    registerCommand: () => undefined,
+    registerFlag: () => undefined,
+    registerTool: (tool: RegisteredTool) => tools.set(tool.name, tool),
+    appendEntry: () => undefined,
+  };
+
+  const ctx = {
+    cwd,
+    ui: {
+      notify: (message: string, level?: string) => notifications.push({ message, level }),
+      setStatus: () => undefined,
+    },
+  };
+
+  fffExtension(pi as never);
+  cleanups.push(() => {
+    void events.get("session_shutdown")?.({}, undefined);
+  });
+
+  return {
+    start: async () => events.get("session_start")?.({ reason: "startup" }, ctx),
+    shutdown: async () => events.get("session_shutdown")?.({}, undefined),
+    find: async (pattern: string, params?: Record<string, unknown>) =>
+      JSON.stringify(
+        await tools.get("fffind")?.execute("test-call", { pattern, ...params }),
+      ),
+    errors: () => notifications.filter((n) => n.level === "error").map((n) => n.message),
+  };
+}
+
+describe("pi-fff: in-process double activation works (#760)", () => {
+  test("two sessions in one process both search against the same dbs", async () => {
+    const dbs = makeDbPaths();
+    const first = startSession(makeWorkspace("session1"), dbs);
+    await first.start();
+    expect(first.errors()).toEqual([]);
+    expect(await first.find("alpha")).toContain("alpha.ts");
+
+    // What createAgentSession does: activate the extension again in-process.
+    const second = startSession(makeWorkspace("session2"), dbs);
+    await second.start();
+    expect(second.errors()).toEqual([]);
+    expect(await second.find("gamma")).toContain("gamma.ts");
+
+    // And the first session keeps working alongside it.
+    expect(await first.find("beta")).toContain("beta.ts");
+
+    await second.shutdown();
+    await first.shutdown();
+  }, 40_000);
+
+  test("aux finder over an external root shares the session dbs (#700)", async () => {
+    const dbs = makeDbPaths();
+    const session = startSession(makeWorkspace("aux-session"), dbs);
+    await session.start();
+    expect(session.errors()).toEqual([]);
+
+    // An absolute out-of-workspace path constraint routes to an aux finder,
+    // which now opens the same frecency/history LMDB paths as the main finder.
+    const external = makeWorkspace("aux-external");
+    expect(await session.find("gamma", { path: external })).toContain("gamma.ts");
+    expect(session.errors()).toEqual([]);
+
+    await session.shutdown();
+  }, 40_000);
+});
@@ -0,0 +1,10 @@
+{
+  "extends": "./tsconfig.json",
+  "compilerOptions": {
+    "rootDir": "./src",
+    "outDir": "./dist",
+    "emitDeclarationOnly": true,
+    "noEmit": false
+  },
+  "include": ["src/**/*"]
+}
@@ -152,8 +152,8 @@ If prebuilt binaries aren't available for your platform:

 ```bash
 # Clone the repository
-git clone https://github.com/dmtrKovalenko/fff.nvim
-cd fff.nvim
+git clone https://github.com/dmtrKovalenko/fff
+cd fff

 # Build the C library
 cargo build --release -p fff-c
@@ -4,19 +4,26 @@
  "private": false,
  "description": "High-performance fuzzy file finder for Node.js - perfect for LLM agent tools",
  "type": "module",
-  "main": "dist/src/index.js",
-  "types": "dist/src/index.d.ts",
+  "main": "dist/index.cjs",
+  "module": "dist/index.js",
+  "types": "dist/index.d.ts",
  "exports": {
    ".": {
-      "import": "./dist/src/index.js",
-      "types": "./dist/src/index.d.ts"
-    }
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js",
+      "require": "./dist/index.cjs",
+      "default": "./dist/index.js"
+    },
+    "./package.json": "./package.json"
  },
  "files": [
    "dist"
  ],
  "scripts": {
-    "build": "tsc",
+    "build": "npm run build:esm && npm run build:cjs && npm run build:types",
+    "build:esm": "bun build ./src/index.ts --format=esm --target=node --external ffi-rs --sourcemap=external --outdir dist",
+    "build:cjs": "bun build ./src/index.ts --format=cjs --target=node --external ffi-rs --sourcemap=external --outdir dist --entry-naming \"[name].cjs\"",
+    "build:types": "tsc -p tsconfig.build.json",
    "test": "node test/e2e.mjs && node test/watch.mjs",
    "typecheck": "tsc --noEmit"
  },
@@ -16,6 +16,9 @@ import { getLibFilename, getNpmPackageName } from "./platform.js";
 * Get the current file's directory
 */
 function getCurrentDir(): string {
+  // CJS build: import.meta.url is inlined at bundle time, __dirname is the truth
+  if (typeof __dirname !== "undefined") return __dirname;
+
  const url = import.meta.url;

  if (url.startsWith("file://")) {
@@ -30,7 +33,7 @@ function getCurrentDir(): string {
 function getPackageDir(): string {
  const currentDir = getCurrentDir();
  // In dev: src/ -> package root
-  // In dist: dist/src/ -> package root
+  // In dist: dist/ -> package root
  // We look for package.json to find the actual root
  let dir = currentDir;
  for (let i = 0; i < 5; i++) {
@@ -573,16 +573,10 @@ export interface FileFinderApi {
  glob(pattern: string, options?: GlobOptions): Result<SearchResult>;

  /** Fuzzy directory search. */
-  directorySearch(
-    query: string,
-    options?: DirSearchOptions,
-  ): Result<DirSearchResult>;
+  directorySearch(query: string, options?: DirSearchOptions): Result<DirSearchResult>;

  /** Fuzzy search over files and directories interleaved by score. */
-  mixedSearch(
-    query: string,
-    options?: SearchOptions,
-  ): Result<MixedSearchResult>;
+  mixedSearch(query: string, options?: SearchOptions): Result<MixedSearchResult>;

  /** Content search (live grep). */
  grep(query: string, options?: GrepOptions): Result<GrepResult>;
@@ -647,10 +641,7 @@ export interface FileFinderApi {
   * Events are debounced and submitted in batches per 100-ms window at most 128 events.
   * Gitignored and other ignored files are never triggering watcher.
   */
-  watch(
-    callback: WatchBatchCallback,
-    options?: WatchOptions,
-  ): Result<WatchUnsubscribe>;
+  watch(callback: WatchBatchCallback, options?: WatchOptions): Result<WatchUnsubscribe>;
  watch(
    pattern: string,
    callback: WatchBatchCallback,
@@ -246,11 +246,7 @@ function readResultEnvelope(
  paramsValue: unknown[],
 ): { rawPtr: JsExternal; struct: FffResultRaw } | Result<never> {
  loadLibrary();
-  const { rawPtr, struct: structData } = callRaw(
-    funcName,
-    paramsType,
-    paramsValue,
-  );
+  const { rawPtr, struct: structData } = callRaw(funcName, paramsType, paramsValue);

  if (structData.success === 0) {
    const errorStr = readCString(structData.error);
@@ -328,8 +324,7 @@ function callJsonResult<T>(
  if (isNullPointer(handlePtr)) return { ok: true, value: undefined as T };
  const jsonStr = readCString(handlePtr);
  freeString(handlePtr);
-  if (jsonStr === null || jsonStr === "")
-    return { ok: true, value: undefined as T };
+  if (jsonStr === null || jsonStr === "") return { ok: true, value: undefined as T };
  try {
    return { ok: true, value: snakeToCamel(JSON.parse(jsonStr)) as T };
  } catch {
@@ -849,16 +844,10 @@ function readGrepMatchFromRaw(raw: FffGrepMatchRaw): GrepMatch {
    match.fuzzyScore = raw.fuzzy_score;
  }
  if (raw.context_before_count > 0) {
-    match.contextBefore = readCStringArray(
-      raw.context_before,
-      raw.context_before_count,
-    );
+    match.contextBefore = readCStringArray(raw.context_before, raw.context_before_count);
  }
  if (raw.context_after_count > 0) {
-    match.contextAfter = readCStringArray(
-      raw.context_after,
-      raw.context_after_count,
-    );
+    match.contextAfter = readCStringArray(raw.context_after, raw.context_after_count);
  }
  if (raw.is_definition !== 0) {
    match.isDefinition = true;
@@ -927,8 +916,7 @@ function parseGrepResult(rawPtr: JsExternal): Result<GrepResult> {
    totalFilesSearched: gr.total_files_searched,
    totalFiles: gr.total_files,
    filteredFileCount: gr.filtered_file_count,
-    nextCursor:
-      gr.next_file_offset > 0 ? createGrepCursor(gr.next_file_offset) : null,
+    nextCursor: gr.next_file_offset > 0 ? createGrepCursor(gr.next_file_offset) : null,
  };
  if (regexFallbackError) {
    grepResult.regexFallbackError = regexFallbackError;
@@ -1280,14 +1268,7 @@ export function ffiGlob(
      DataType.U32, // page_index
      DataType.U32, // page_size
    ],
-    paramsValue: [
-      handle,
-      pattern,
-      currentFile,
-      maxThreads,
-      pageIndex,
-      pageSize,
-    ],
+    paramsValue: [handle, pattern, currentFile, maxThreads, pageIndex, pageSize],
    freeResultMemory: false,
  }) as JsExternal;

@@ -1319,14 +1300,7 @@ export function ffiSearchDirectories(
      DataType.U32, // page_index
      DataType.U32, // page_size
    ],
-    paramsValue: [
-      handle,
-      query,
-      currentFile ?? "",
-      maxThreads,
-      pageIndex,
-      pageSize,
-    ],
+    paramsValue: [handle, query, currentFile ?? "", maxThreads, pageIndex, pageSize],
    freeResultMemory: false,
  }) as JsExternal;

@@ -1545,11 +1519,7 @@ export function ffiGetScanProgress(handle: NativeHandle): Result<{
  isWarmupComplete: boolean;
 }> {
  loadLibrary();
-  const res = readResultEnvelope(
-    "fff_get_scan_progress",
-    [DataType.External],
-    [handle],
-  );
+  const res = readResultEnvelope("fff_get_scan_progress", [DataType.External], [handle]);
  if ("ok" in res) return res;

  const handlePtr = res.struct.handle;
@@ -1584,10 +1554,7 @@ export function ffiGetScanProgress(handle: NativeHandle): Result<{
 /**
 * Wait for a tree scan to complete.
 */
-export function ffiWaitForScan(
-  handle: NativeHandle,
-  timeoutMs: number,
-): Result<boolean> {
+export function ffiWaitForScan(handle: NativeHandle, timeoutMs: number): Result<boolean> {
  return callBoolResult(
    "fff_wait_for_scan",
    [DataType.External, DataType.U64],
@@ -1598,10 +1565,7 @@ export function ffiWaitForScan(
 /**
 * Restart index in new path.
 */
-export function ffiRestartIndex(
-  handle: NativeHandle,
-  newPath: string,
-): Result<void> {
+export function ffiRestartIndex(handle: NativeHandle, newPath: string): Result<void> {
  return callVoidResult(
    "fff_restart_index",
    [DataType.External, DataType.String],
@@ -1772,8 +1736,7 @@ function ensureWatchTrampoline(): JsExternal {
 // fff watcher uses a single cross-boundary FFI callback to deliver all events which we then manually
 // mapping to the user's javascript functions
 function ensureWatchCallbackRegistered(handle: NativeHandle): Result<void> {
-  if (watchInstances.has(handle as unknown))
-    return { ok: true, value: undefined };
+  if (watchInstances.has(handle as unknown)) return { ok: true, value: undefined };
  const trampoline = ensureWatchTrampoline();
  const registered = callVoidResult(
    "fff_set_watch_callback",
@@ -1785,11 +1748,7 @@ function ensureWatchCallbackRegistered(handle: NativeHandle): Result<void> {
 }

 function releaseWatchTrampolineIfIdle(): void {
-  if (
-    watchHandlers.size > 0 ||
-    watchInstances.size > 0 ||
-    watchTrampoline === null
-  )
+  if (watchHandlers.size > 0 || watchInstances.size > 0 || watchTrampoline === null)
    return;
  freePointer({
    paramsType: [WATCH_TRAMPOLINE_TYPE],
@@ -1835,10 +1794,7 @@ export function ffiWatch(
 * this returns the callback can never run again (a late native tail batch
 * misses the map lookup and is dropped).
 */
-export function ffiUnwatch(
-  handle: NativeHandle,
-  watchId: number,
-): Result<boolean> {
+export function ffiUnwatch(handle: NativeHandle, watchId: number): Result<boolean> {
  const result = callBoolResult(
    "fff_unwatch",
    [DataType.External, DataType.U64],
@@ -2,13 +2,13 @@ import { after, before, describe, it } from "node:test";
 import { strict as assert } from "node:assert";
 import { dirname, resolve } from "node:path";
 import { fileURLToPath } from "node:url";
-import { FileFinder } from "../dist/src/index.js";
+import { FileFinder } from "../dist/index.js";

 const __dirname = dirname(fileURLToPath(import.meta.url));
 const REPO_ROOT = resolve(__dirname, "..", "..", "..");
 const normalizePath = (p) => p.replace(/\\/g, "/");

-/** @type {import("../dist/src/finder.js").FileFinder | null} */
+/** @type {import("../dist/finder.js").FileFinder | null} */
 let finder = null;

 describe("fff-node", { concurrency: 1 }, () => {
@@ -13,7 +13,7 @@ import { existsSync } from "node:fs";
 import { dirname, resolve } from "node:path";
 import process from "node:process";
 import { fileURLToPath } from "node:url";
-import { FileFinder } from "../dist/src/index.js";
+import { FileFinder } from "../dist/index.js";

 const __dirname = dirname(fileURLToPath(import.meta.url));
 const REPO_ROOT = resolve(__dirname, "..", "..", "..");
@@ -15,7 +15,7 @@ import { mkdtempSync, realpathSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join, sep } from "node:path";
 import { promisify } from "node:util";
-import { FileFinder } from "../dist/src/index.js";
+import { FileFinder } from "../dist/index.js";

 const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));

@@ -33,7 +33,7 @@ async function waitFor(cond, timeoutMs = 10_000) {
 /** All events delivered to a batch-callback mock, flattened across calls. */
 const deliveredEvents = (fn) => fn.mock.calls.flatMap((call) => call.arguments[0]);

-/** @type {import("../dist/src/finder.js").FileFinder | null} */
+/** @type {import("../dist/finder.js").FileFinder | null} */
 let finder = null;
 /** @type {string} */
 let baseDir = "";
@@ -237,7 +237,7 @@ describe("fff-node watch", { concurrency: 1 }, () => {
      import { mkdtempSync, writeFileSync } from "node:fs";
      import { tmpdir } from "node:os";
      import { join } from "node:path";
-      import { FileFinder } from ${JSON.stringify(new URL("../dist/src/index.js", import.meta.url).href)};
+      import { FileFinder } from ${JSON.stringify(new URL("../dist/index.js", import.meta.url).href)};

      const dir = mkdtempSync(join(tmpdir(), "fff-watch-exit-"));
      writeFileSync(join(dir, "seed.txt"), "seed");
@@ -0,0 +1,10 @@
+{
+  "extends": "./tsconfig.json",
+  "compilerOptions": {
+    "rootDir": "./src",
+    "outDir": "./dist",
+    "emitDeclarationOnly": true,
+    "noEmit": false
+  },
+  "include": ["src/**/*"]
+}
@@ -1,6 +1,6 @@
 # fff-search

-Python bindings for [FFF (Fast File Finder)](https://github.com/dmtrKovalenko/fff.nvim), built with [PyO3](https://pyo3.rs/) and [Maturin](https://www.maturin.rs/). Install with `pip install fff-search`; import as `fff`.
+Python bindings for [FFF (Fast File Finder)](https://github.com/dmtrKovalenko/fff), built with [PyO3](https://pyo3.rs/) and [Maturin](https://www.maturin.rs/). Install with `pip install fff-search`; import as `fff`.

 ## Requirements

@@ -1,6 +1,6 @@
 [project]
 name = "fff-search"
-version = "0.10.1"
+version = "0.10.4"
 description = "Python bindings for FFF (Fast File Finder)"
 readme = "README.md"
 license = { text = "MIT" }
@@ -15,8 +15,8 @@ dev = [
 ]

 [project.urls]
-Repository = "https://github.com/dmtrKovalenko/fff.nvim"
-Issues = "https://github.com/dmtrKovalenko/fff.nvim/issues"
+Repository = "https://github.com/dmtrKovalenko/fff"
+Issues = "https://github.com/dmtrKovalenko/fff/issues"

 [build-system]
 requires = ["maturin>=1.0,<2.0"]
@@ -50,7 +50,7 @@ class FileFinder(_FileFinder):
        return True


-__version__ = "0.10.1"
+__version__ = "0.10.4"

 __all__ = [
    "FFFException",
@@ -34,7 +34,7 @@ wheels = [

 [[package]]
 name = "fff-search"
-version = "0.10.1"
+version = "0.10.4"
 source = { editable = "." }

 [package.optional-dependencies]
@@ -1,6 +1,6 @@
 # @ff-labs/pi-fff

-A [pi](https://github.com/badlogic/pi-mono) extension that replaces the built-in `find` and `grep` tools with [FFF](https://github.com/dmtrKovalenko/fff.nvim) — a Rust-native, SIMD-accelerated file finder with built-in memory.
+A [pi](https://github.com/badlogic/pi-mono) extension that replaces the built-in `find` and `grep` tools with [FFF](https://github.com/dmtrKovalenko/fff) — a Rust-native, SIMD-accelerated file finder with built-in memory.

 ## What it does

@@ -44,20 +44,20 @@ pi install -l npm:@ff-labs/pi-fff
 **Via git:**

 ```bash
-pi install git:github.com/dmtrKovalenko/fff.nvim
+pi install git:github.com/dmtrKovalenko/fff
 ```

 Pin to a release:

 ```bash
-pi install git:github.com/dmtrKovalenko/fff.nvim@v0.3.0
+pi install git:github.com/dmtrKovalenko/fff@v0.3.0
 ```

 ### Local development / manual install

 ```bash
-git clone https://github.com/dmtrKovalenko/fff.nvim.git
-cd fff.nvim/packages/pi-fff
+git clone https://github.com/dmtrKovalenko/fff.git
+cd fff/packages/pi-fff
 npm install
 ```

@@ -65,14 +65,14 @@ Then add to your pi `settings.json`:

 ```json
 {
-  "extensions": ["/path/to/fff.nvim/packages/pi-fff/src/index.ts"]
+  "extensions": ["/path/to/fff/packages/pi-fff/src/index.ts"]
 }
 ```

 Or test directly:

 ```bash
-pi -e /path/to/fff.nvim/packages/pi-fff/src/index.ts
+pi -e /path/to/fff/packages/pi-fff/src/index.ts
 ```

 This extension registers FFF-powered tools (`fffind`, `ffgrep`, `fff-multi-grep`) alongside pi's built-in tools.
@@ -134,7 +134,8 @@ Mode precedence:
 - `--fff-mode <mode>` — set mode (see above)
 - `--fff-frecency-db <path>` — path to frecency database (also: `FFF_FRECENCY_DB` env)
 - `--fff-history-db <path>` — path to query history database (also: `FFF_HISTORY_DB` env)
- `--fff-enable-root-scan` — allow indexing when launched from `/` (also: `FFF_ENABLE_ROOT_SCAN=1` env). FFF refuses to init at the filesystem root by default. Home directory scanning is always enabled for pi.
+- `--fff-enable-root-scan` — allow indexing when launched from `/` (also: `FFF_ENABLE_ROOT_SCAN=1` env). FFF refuses to init at the filesystem root by default.
+- `--fff-enable-home-scan` — index the home directory when launched from `$HOME` (also: `FFF_ENABLE_HOME_SCAN` env). Enabled by default. Disable with `--fff-enable-home-scan=false` or `FFF_ENABLE_HOME_SCAN=0` if your `$HOME` contains huge trees (toolchains, kernel sources, build outputs) that make the background index run for a long time. When launched from `$HOME` with this enabled, pi shows a warning that the whole home tree is being indexed.

 ## Data

@@ -1,7 +1,7 @@
 import fs from "node:fs";
-import os from "node:os";
 import path from "node:path";
 import type { FileFinderApi } from "@ff-labs/fff-node";
+import { HOME_DIR } from "./paths";
 import { loadSdk, SCAN_TIMEOUT_MS } from "./sdk";

 export const MAX_AUX = 3;
@@ -15,6 +15,11 @@ interface AuxPicker {

 export interface AuxOpts {
  enableFsRootScanning: boolean;
+  enableHomeDirScanning?: boolean;
+  // Called before a newly spawned aux picker starts a scan that covers $HOME.
+  onHomeDirScan?: (root: string) => void;
+  frecencyDbPath?: string;
+  historyDbPath?: string;
 }

 export class AuxFinderPool {
@@ -84,26 +89,31 @@ export class AuxFinderPool {
  private async create(root: string): Promise<AuxPicker> {
    if (this.entries.length >= MAX_AUX) {
      let oldest = this.entries[0];
-      for (const e of this.entries)
-        if (e.lastUsed < oldest.lastUsed) oldest = e;
+      for (const e of this.entries) if (e.lastUsed < oldest.lastUsed) oldest = e;
      if (!oldest.finder.isDestroyed) oldest.finder.destroy();
      this.entries = this.entries.filter((e) => e !== oldest);
    }

+    const enableHomeDirScanning = this.opts.enableHomeDirScanning ?? true;
+    // A fresh picker rooted at (or above) $HOME walks the whole home tree, so
+    // the user gets told every time the agent spawns one — see issue #743.
+    if (enableHomeDirScanning && rootCovers(root, HOME_DIR)) {
+      this.opts.onHomeDirScan?.(root);
+    }
+
    const { FileFinder } = await loadSdk();
-    // LMDB env can only be opened once per process; the main finder already
-    // owns the frecency/history DBs. Aux finders are transient and run without
-    // persistent scoring — see issue #700.
    const result = FileFinder.create({
      basePath: root,
+      frecencyDbPath: this.opts.frecencyDbPath,
+      historyDbPath: this.opts.historyDbPath,
      aiMode: true,
-      enableHomeDirScanning: true,
+      enableHomeDirScanning,
      enableFsRootScanning: this.opts.enableFsRootScanning,
    });
-    if (!result.ok)
-      throw new Error(
-        `Failed to create aux file finder for ${root}: ${result.error}`,
-      );
+
+    if (!result.ok) {
+      throw new Error(`Failed to create aux file finder for ${root}: ${result.error}`);
+    }

    await result.value.waitForScan(SCAN_TIMEOUT_MS);
    const entry: AuxPicker = {
@@ -125,9 +135,7 @@ export class AuxFinderPool {
 // remainder usable as a fuzzy path constraint relative to that root. Glob and
 // nonexistent segments both go into the suffix: we walk up to the nearest
 // existing ancestor so partially-wrong paths still resolve to a search root.
-export function resolveAuxRoot(
-  absPath: string,
-): { root: string; suffix: string } | null {
+export function resolveAuxRoot(absPath: string): { root: string; suffix: string } | null {
  const trimmed = path.normalize(absPath.trim()).replace(/\/+$/, "") || "/";
  if (!path.isAbsolute(trimmed)) return null;
  if (trimmed === path.sep) return { root: path.sep, suffix: "" };
@@ -176,7 +184,7 @@ export function routePathConstraint(
  let candidate = pathConstraint.trim();
  if (!candidate) return null;
  if (candidate === "~" || candidate.startsWith("~/"))
-    candidate = path.join(os.homedir(), candidate.slice(1));
+    candidate = path.join(HOME_DIR, candidate.slice(1));
  if (!path.isAbsolute(candidate)) {
    // Plain workspace-relative constraints stay on the workspace finder.
    if (candidate !== ".." && !candidate.startsWith("../")) return null;
@@ -187,7 +195,6 @@ export function routePathConstraint(
  return resolveAuxRoot(candidate);
 }

-
 export function rootCovers(root: string, target: string): boolean {
  if (root === target) return true;
  const prefix = root.endsWith(path.sep) ? root : root + path.sep;
@@ -0,0 +1,9 @@
+import os from "node:os";
+import path from "node:path";
+
+// Resolved once per process: os.homedir() hits the env/passwd on every call.
+export const HOME_DIR = path.resolve(os.homedir());
+
+export function isHomeDir(dir: string): boolean {
+  return path.resolve(dir) === HOME_DIR;
+}
@@ -10,11 +10,7 @@ export function normalizePathConstraint(
  if (path.isAbsolute(trimmed)) {
    const relative = path.relative(cwd, trimmed).replaceAll(path.sep, "/");
    if (relative === "") return null;
-    if (
-      relative.startsWith("../") ||
-      relative === ".." ||
-      path.isAbsolute(relative)
-    ) {
+    if (relative.startsWith("../") || relative === ".." || path.isAbsolute(relative)) {
      throw new Error(
        `Path constraint must be relative to the workspace: ${pathConstraint}`,
      );
@@ -22,8 +22,21 @@ function detectRuntime(): "bun" | "node" {
 export function loadSdk(): Promise<{ FileFinder: FileFinderStatic }> {
  if (sdkPromise) return sdkPromise;

+  // Pi reloads extension modules with jiti moduleCache:false, so this module
+  // is re-executed on every /reload. Re-importing the fff-bun module graph
+  // (which top-level awaits a `type: "file"` import of the native .so) hangs
+  // forever inside the Bun-compiled pi binary. Cache the first import on
+  // globalThis so reloads reuse the resolved module instead of re-importing.
+  const g = globalThis as Record<string, unknown>;
+  if (g.__fffSdkPromiseGlobal) {
+    sdkPromise = g.__fffSdkPromiseGlobal as Promise<{ FileFinder: FileFinderStatic }>;
+    return sdkPromise;
+  }
+
  // default to node as it seems like default option
  const pkg = detectRuntime() === "bun" ? "@ff-labs/fff-bun" : "@ff-labs/fff-node";
-  sdkPromise = import(pkg) as Promise<{ FileFinder: FileFinderStatic }>;
-  return sdkPromise;
+  const p = import(pkg) as Promise<{ FileFinder: FileFinderStatic }>;
+  sdkPromise = p;
+  (globalThis as Record<string, unknown>).__fffSdkPromiseGlobal = p;
+  return p;
 }
@@ -98,9 +98,7 @@ describe("routePathConstraint", () => {
    });

    test("returns null when .. resolves back inside the workspace", () => {
-      expect(
-        routePathConstraint("../workspace/src", workspace),
-      ).toBeNull();
+      expect(routePathConstraint("../workspace/src", workspace)).toBeNull();
    });
  });
 });
@@ -1,4 +1,6 @@
 import { describe, expect, mock, test } from "bun:test";
+import os from "node:os";
+import path from "node:path";

 interface MockFinder {
  isDestroyed: boolean;
@@ -40,10 +42,10 @@ mock.module("@ff-labs/fff-bun", () => finderModule);

 const { AuxFinderPool } = await import("../src/aux-finders");

-function makePool() {
+function makePool(opts: Record<string, unknown> = {}) {
  created.length = 0;
  createOptions.length = 0;
-  return new AuxFinderPool({ enableFsRootScanning: false });
+  return new AuxFinderPool({ enableFsRootScanning: false, ...opts });
 }

 describe("AuxFinderPool covering reuse", () => {
@@ -92,16 +94,51 @@ describe("AuxFinderPool covering reuse", () => {
    expect(created.length).toBe(2);
  });

-  // Regression for #700: aux finders must not reopen the main frecency/history
-  // LMDB envs, or heed fails with "environment already open in this program".
-  test("aux finders are created without frecency/history db paths", async () => {
-    const pool = makePool();
+  // Regression for #743: the agent spawning an aux picker over $HOME must warn
+  // the user every time, not silently walk the home tree.
+  test("notifies on every aux picker that covers $HOME", async () => {
+    const onHomeDirScan = mock(() => undefined);
+    const pool = makePool({ onHomeDirScan });
+    const home = os.homedir();
+
+    await pool.acquire(home);
+    await pool.acquire(path.dirname(home));
+    expect(onHomeDirScan.mock.calls).toEqual([[home], [path.dirname(home)]]);
+
+    // Project-scoped roots below $HOME do not walk the whole home tree.
+    await pool.acquire(path.join(home, "dev", "some-project"));
+    expect(onHomeDirScan).toHaveBeenCalledTimes(2);
+  });
+
+  test("no aux notification when home scanning is disabled", async () => {
+    const onHomeDirScan = mock(() => undefined);
+    const pool = makePool({ enableHomeDirScanning: false, onHomeDirScan });
+
+    await pool.acquire(os.homedir());
+    expect(onHomeDirScan).not.toHaveBeenCalled();
+    expect(createOptions[0].enableHomeDirScanning).toBe(false);
+  });
+
+  // #700 is fixed by the process-wide LMDB env pool: same-path opens share one
+  // env, so aux finders now reuse the session's frecency/history DBs.
+  test("aux finders receive the pool's frecency/history db paths", async () => {
+    const pool = makePool({
+      frecencyDbPath: "/dbs/frecency",
+      historyDbPath: "/dbs/history",
+    });
    await pool.acquire("/a/b/c");
    await pool.acquire("/x/y");
    expect(createOptions.length).toBe(2);
    for (const opts of createOptions) {
-      expect(opts.frecencyDbPath).toBeUndefined();
-      expect(opts.historyDbPath).toBeUndefined();
+      expect(opts.frecencyDbPath).toBe("/dbs/frecency");
+      expect(opts.historyDbPath).toBe("/dbs/history");
    }
  });
+
+  test("aux finders stay db-less when the session has no db paths", async () => {
+    const pool = makePool();
+    await pool.acquire("/a/b/c");
+    expect(createOptions[0].frecencyDbPath).toBeUndefined();
+    expect(createOptions[0].historyDbPath).toBeUndefined();
+  });
 });
@@ -1,20 +1,35 @@
 import { beforeEach, describe, expect, mock, test } from "bun:test";
+import os from "node:os";

 type MockFinder = {
  isDestroyed: boolean;
  waitForScan: ReturnType<typeof mock>;
  mixedSearch: ReturnType<typeof mock>;
+  getScanProgress: ReturnType<typeof mock>;
  destroy: ReturnType<typeof mock>;
 };

 const createCalls: unknown[] = [];
 let finders: MockFinder[] = [];
 let mixedSearchImpl: ((query: string, options: unknown) => unknown) | undefined;
+let scanProgressImpl: (() => unknown) | undefined;

 function createMockFinder(): MockFinder {
  return {
    isDestroyed: false,
    waitForScan: mock(async () => undefined),
+    getScanProgress: mock(() => {
+      if (scanProgressImpl) return scanProgressImpl();
+      return {
+        ok: true,
+        value: {
+          scannedFilesCount: 0,
+          isScanning: false,
+          isWatcherReady: true,
+          isWarmupComplete: true,
+        },
+      };
+    }),
    mixedSearch: mock((query: string, options: unknown) => {
      if (mixedSearchImpl) return mixedSearchImpl(query, options);
      return {
@@ -102,20 +117,21 @@ function createPi(mode?: string) {
  return { pi, events, commands };
 }

-function createContext() {
+function createContext(cwd = "/tmp/workspace") {
  return {
-    cwd: "/tmp/workspace",
+    cwd,
    ui: {
      addAutocompleteProvider: mock(() => undefined),
      notify: mock(() => undefined),
      setEditorComponent: mock(() => undefined),
+      setStatus: mock(() => undefined),
    },
  };
 }

-async function start(mode?: string) {
+async function start(mode?: string, cwd?: string) {
  const setup = createPi(mode);
-  const ctx = createContext();
+  const ctx = createContext(cwd);
  fffExtension(setup.pi as any);

  const sessionStart = setup.events.get("session_start");
@@ -125,6 +141,10 @@ async function start(mode?: string) {
  return { ...setup, ctx };
 }

+async function shutdown(setup: { events: Map<string, EventHandler> }) {
+  await setup.events.get("session_shutdown")?.({}, undefined);
+}
+
 function currentProvider(
  result = { items: [{ value: "base", label: "base" }], prefix: "ba" },
 ) {
@@ -143,7 +163,71 @@ beforeEach(() => {
  createCalls.length = 0;
  finders = [];
  mixedSearchImpl = undefined;
+  scanProgressImpl = undefined;
  delete process.env.PI_FFF_MODE;
+  delete process.env.FFF_ENABLE_HOME_SCAN;
+});
+
+// Regression for #743: launching from $HOME must be visible and interruptible.
+describe("pi-fff $HOME scan warning", () => {
+  test("warns and pins a status when cwd is $HOME", async () => {
+    const setup = await start(undefined, os.homedir());
+
+    expect(setup.ctx.ui.notify).toHaveBeenCalledTimes(1);
+    const [message, level] = setup.ctx.ui.notify.mock.calls[0];
+    expect(message).toContain(os.homedir());
+    expect(level).toBe("warning");
+    expect(setup.ctx.ui.setStatus).toHaveBeenCalledWith(
+      "fff",
+      "Agent is indexing $HOME, this can lead to high CPU",
+    );
+    await shutdown(setup);
+  });
+
+  test("stays silent outside $HOME", async () => {
+    const { ctx } = await start();
+
+    expect(ctx.ui.notify).not.toHaveBeenCalled();
+    expect(ctx.ui.setStatus).not.toHaveBeenCalled();
+  });
+
+  test("clears the status once the scan settles", async () => {
+    const setup = await start(undefined, os.homedir());
+
+    expect(setup.ctx.ui.setStatus).toHaveBeenLastCalledWith("fff", undefined);
+    await shutdown(setup);
+  });
+
+  // waitForScan resolves on timeout, so a slow $HOME walk keeps the footer up.
+  test("keeps reporting live progress while the scan is still running", async () => {
+    scanProgressImpl = () => ({
+      ok: true,
+      value: {
+        scannedFilesCount: 12345,
+        isScanning: true,
+        isWatcherReady: false,
+        isWarmupComplete: false,
+      },
+    });
+    const setup = await start(undefined, os.homedir());
+
+    const [key, text] = setup.ctx.ui.setStatus.mock.calls.at(-1) as [string, string];
+    expect(key).toBe("fff");
+    expect(text).toContain("12345 files");
+
+    // session_shutdown must stop the poller and clear the footer.
+    await shutdown(setup);
+    expect(setup.ctx.ui.setStatus).toHaveBeenLastCalledWith("fff", undefined);
+  });
+
+  test("no warning when home scanning is disabled", async () => {
+    process.env.FFF_ENABLE_HOME_SCAN = "0";
+    const setup = await start(undefined, os.homedir());
+
+    expect(setup.ctx.ui.notify).not.toHaveBeenCalled();
+    expect(setup.ctx.ui.setStatus).not.toHaveBeenCalled();
+    await shutdown(setup);
+  });
 });

 describe("pi-fff autocomplete registration", () => {
@@ -164,6 +248,14 @@ describe("pi-fff autocomplete registration", () => {
    ]);
  });

+  test("FFF_ENABLE_HOME_SCAN=0 disables home dir scanning", async () => {
+    process.env.FFF_ENABLE_HOME_SCAN = "0";
+    await start();
+
+    const opts = createCalls[0] as { enableHomeDirScanning: boolean };
+    expect(opts.enableHomeDirScanning).toBe(false);
+  });
+
  test("session_start survives hosts without addAutocompleteProvider", async () => {
    const setup = createPi();
    const ctx = {
@@ -567,16 +567,10 @@ export interface FileFinderApi {
  glob(pattern: string, options?: GlobOptions): Result<SearchResult>;

  /** Fuzzy directory search. */
-  directorySearch(
-    query: string,
-    options?: DirSearchOptions,
-  ): Result<DirSearchResult>;
+  directorySearch(query: string, options?: DirSearchOptions): Result<DirSearchResult>;

  /** Fuzzy search over files and directories interleaved by score. */
-  mixedSearch(
-    query: string,
-    options?: SearchOptions,
-  ): Result<MixedSearchResult>;
+  mixedSearch(query: string, options?: SearchOptions): Result<MixedSearchResult>;

  /** Content search (live grep). */
  grep(query: string, options?: GrepOptions): Result<GrepResult>;
@@ -641,10 +635,7 @@ export interface FileFinderApi {
   * Events are debounced and submitted in batches per 100-ms window at most 128 events.
   * Gitignored and other ignored files are never triggering watcher.
   */
-  watch(
-    callback: WatchBatchCallback,
-    options?: WatchOptions,
-  ): Result<WatchUnsubscribe>;
+  watch(callback: WatchBatchCallback, options?: WatchOptions): Result<WatchUnsubscribe>;
  watch(
    pattern: string,
    callback: WatchBatchCallback,
@@ -77,4 +77,4 @@ git push origin "$TAG"

 echo ""
 echo "Release $VERSION created and pushed."
-echo "CI will build and publish: https://github.com/dmtrKovalenko/fff.nvim/actions"
+echo "CI will build and publish: https://github.com/dmtrKovalenko/fff/actions"
@@ -1,5 +1,5 @@
 ---@diagnostic disable: undefined-field, missing-fields
-- Regression test for https://github.com/dmtrKovalenko/fff.nvim/issues/389
+-- Regression test for https://github.com/dmtrKovalenko/fff/issues/389
 -- When find_files_in_dir(dir) runs with dir != neovim's cwd, the Rust indexer
 -- reports paths relative to `dir`, but the Lua side used to resolve them
 -- against neovim's cwd when calling :edit / preview / quickfix — so files
@@ -31,7 +31,7 @@ local function find_result_by_name(items, name)
 end

 describe('programmatic search APIs', function()
-  describe('against the actual fff.nvim repo', function()
+  describe('against the actual fff repo', function()
    before_each(function()
      pcall(vim.api.nvim_del_augroup_by_name, 'fff_file_tracking')
      vim.g.fff = {}
@@ -181,9 +181,9 @@ describe('programmatic search APIs', function()
      fd:write('-- only lives in the other sandbox\n')
      fd:close()

-      -- Sanity: the file does not exist in the primary (fff.nvim) index.
+      -- Sanity: the file does not exist in the primary (fff) index.
      local before = fff.file_search(other_filename)
-      assert.are.equal(0, #before.items, 'sandbox file leaked into primary fff.nvim index')
+      assert.are.equal(0, #before.items, 'sandbox file leaked into primary fff index')

      local result = fff.file_search(other_filename, { cwd = sandbox_root })
      assert.is_true(#result.items > 0, 'cwd switch did not surface file from the new root')
@@ -202,7 +202,7 @@ describe('programmatic search APIs', function()
      sandbox_root = vim.fn.tempname() .. '_other_grep'
      vim.fn.mkdir(sandbox_root, 'p')
      -- Build the marker by concatenation so the literal string doesn't
-      -- appear anywhere in the fff.nvim tree (otherwise the "before" grep
+      -- appear anywhere in the fff tree (otherwise the "before" grep
      -- would find this very test file via its own marker constant).
      local marker = 'isolated_grep' .. '_marker_xyzzy'
      local fd = assert(io.open(sandbox_root .. '/grep_target.lua', 'w'))
@@ -214,9 +214,9 @@ describe('programmatic search APIs', function()
      -- this is fine in practice cause this is not a real use case for fff to search RIGHT AFTER mkdir
      if vim.fn.has('win32') == 1 then vim.wait(250, function() return false end) end

-      -- Marker must not exist anywhere in the primary fff.nvim tree.
+      -- Marker must not exist anywhere in the primary fff tree.
      local before = fff.content_search(marker)
-      assert.are.equal(0, #before.items, 'marker leaked into primary fff.nvim tree')
+      assert.are.equal(0, #before.items, 'marker leaked into primary fff tree')

      local result = fff.content_search(marker, { cwd = sandbox_root })
      assert.is_true(#result.items > 0, 'cwd switch did not surface match from the new root')
Author	SHA1	Message	Date
Dmitriy Kovalenko	7b2864e4b1	chore(ci): Attempt to optimize CI build time	2026-08-15 17:33:05 -07:00
dmtrKovalenko	2c9cae2e0f	chore: bump fff-mcp release artifacts to v0.10.4	2026-08-15 22:16:40 +00:00
Dmitriy Kovalenko	a166255d8f	chore: release 0.10.4 Build & Publish / Build Neovim aarch64-apple-darwin (push) Has been cancelled Details Build & Publish / Build Neovim aarch64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build Neovim aarch64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-apple-darwin (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-linux-android (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build Neovim aarch64-linux-android (push) Has been cancelled Details Build & Publish / Build Neovim aarch64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build Neovim x86_64-apple-darwin (push) Has been cancelled Details Build & Publish / Build Neovim x86_64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build Neovim x86_64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build Neovim x86_64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build C FFI x86_64-apple-darwin (push) Has been cancelled Details Build & Publish / Build C FFI x86_64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build C FFI x86_64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build C FFI x86_64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build MCP aarch64-apple-darwin (push) Has been cancelled Details Build & Publish / Build MCP aarch64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build MCP aarch64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build MCP aarch64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build MCP x86_64-apple-darwin (push) Has been cancelled Details Build & Publish / Build MCP x86_64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build MCP x86_64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build MCP x86_64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build Python wheels aarch64 (macos-latest) (push) Has been cancelled Details Build & Publish / Build Python wheels x86_64 (macos-latest) (push) Has been cancelled Details Build & Publish / Build Python wheels aarch64 (ubuntu-latest) (push) Has been cancelled Details Build & Publish / Build Python wheels x86_64 (ubuntu-latest) (push) Has been cancelled Details Build & Publish / Build Python wheels x86_64 (windows-latest) (push) Has been cancelled Details Build & Publish / Build Python sdist (push) Has been cancelled Details Build & Publish / Release (push) Has been cancelled Details Build & Publish / Publish Python wheels to PyPI (push) Has been cancelled Details Build & Publish / Publish Rust crates (push) Has been cancelled Details Build & Publish / Publish npm packages (push) Has been cancelled Details	2026-08-15 14:35:20 -07:00
Kh05ifr4nD	ffd8eb9c08	fix(mcp): standalone constraints in multi_grep (#753 ) * fix: apply standalone constraints in multi_grep * chore: use pub(crate) --------- Co-authored-by: Dmitriy Kovalenko <dmitriy@iusevimbtw.com>	2026-08-14 22:11:24 -07:00
Dmitriy Kovalenko	ed6b40a591	chore(deps): Update zlob to 1.6.3 (#777 )	2026-08-14 21:47:10 -07:00
Dmitriy Kovalenko	c4f6c600d6	fix(core): share one LMDB env per path within a process (#775 ) because LMDB usee posix lock one process can not hold many open databases	2026-08-14 21:46:46 -07:00
Gustav the Bot	787a9b1cb7	fix: rebuild file picker after FFFClearCache files (#772 ) (#773 ) clear_cache('files') drops the Rust picker but core.lua kept state.file_picker_initialized set, so ensure_initialized short-circuited on state.initialized and never rebuilt it. Every later FFFScan/search then ran against a dropped picker (watcher logs "File picker not initialized"), which SIGSEGVs on Linux. Split ensure_initialized: one-time setup stays gated on state.initialized, picker creation is gated separately on state.file_picker_initialized. clear_cache now clears that flag via mark_file_picker_uninitialized so the next ensure_initialized rebuilds the picker. Closes #772 Co-authored-by: gustav-fff <286169375+gustav-fff@users.noreply.github.com>	2026-08-14 09:10:03 -07:00
Dmitriy Kovalenko	6398d32c0c	fix: Do not exit MCP if the parent process is alive (#770 ) closes #703 Bumped inactivity timeout to an hour and make it actually check every minute if parent is alive and working	2026-08-13 17:39:14 -07:00
Gustav the Bot	c6194b848d	fix(mcp): declare readOnlyHint on all tools (#771 ) (#774 ) find_files, grep, and multi_grep only read the filesystem but declared no MCP tool annotations, so plan-mode / read-only clients blocked them. Add annotations(read_only_hint = true, destructive_hint = false, open_world_hint = false) to each #[tool]. Closes #771 Co-authored-by: gustav-fff <286169375+gustav-fff@users.noreply.github.com>	2026-08-13 08:42:55 -07:00
Gustav the Bot	b71b7cf4ac	fix(pi-fff): cap grep total results with pageSize, clamp context (#768 ) (#769 ) ffgrep/fff-multi-grep passed the documented `limit` only as `maxMatchesPerFile`, so matches spread across files could return a full SDK page (default 50) instead of `limit`. Pass `limit` as `pageSize` too, on both grep tools and the fuzzy fallback, so the cap applies to total matches per page; excess stays retrievable via the existing cursor. Also clamp `context` to a bounded non-negative integer so a large value cannot multiply output size past the model window. Closes #768 Co-authored-by: gustav-fff <286169375+gustav-fff@users.noreply.github.com>	2026-08-13 07:45:48 -07:00
Joe Taber	cc289f0f93	chore: rename old fff.nvim repo refs (#762 ) * chore: point residual repo refs at dmtrKovalenko/fff Update install URLs, download targets, and docs after the GitHub rename from fff.nvim to fff. Preserve the existing Neovim package name (fff.nvim) in lazy/vim.pack snippets so upgraders keep the same install dir and lockfile identity. * chore(docs): align Neovim package name with repo basename Drop the explicit fff.nvim package name so lazy/vim.pack use fff from the repository URL. Mild migration cost for existing installs (new plugin dir + clean of the old one). Drop this commit to keep the lowest-pain name=fff.nvim install snippets from the previous commit.	2026-08-11 21:40:21 -07:00
Gustav the Bot	2107f038dc	fix(grep): keep FilePath scope in regex/literal fallback (#756 ) (#764 ) * fix(grep): keep FilePath scope in regex/literal fallback (#756) The literal/regex fallback rebuilt the query with empty constraints, dropping an explicit inline FilePath scope. In regex mode a top-level alternation then leaked matches into files outside the pinned path. Preserve FilePath constraints in the fallback query. Closes #756 * chore: cargo fmt (#756) --------- Co-authored-by: gustav-fff <286169375+gustav-fff@users.noreply.github.com>	2026-08-11 20:54:36 -07:00
Dmitriy Kovalenko	9441cbc5ae	fix(bun): Add a proper build step for fff-bun (#766 ) We never build it and this seems to be not a good idea	2026-08-11 20:45:40 -07:00
Gustav the Bot	dd87489d6b	fix: add missing clap attribute to no_content_indexing (#754 ) (#765 ) The no_content_indexing field lacked its #[arg(long = ...)] attribute, so clap treated it as a positional with a SetTrue action. Debug builds panicked on any invocation; release builds exposed [NO_CONTENT_INDEXING] as a positional and rejected --no-content-indexing. Closes #754 Co-authored-by: gustav-fff <286169375+gustav-fff@users.noreply.github.com>	2026-08-11 11:28:58 -07:00
chenydev	d4c416c2fc	fix(pi-fff): cache sdk import across reloads to avoid hang (#758 ) Pi reloads extension modules with jiti moduleCache:false, so loadSdk() re-executes a dynamic import of the fff-bun module graph on every /reload. The fff-bun graph top-level awaits a type:file import of the native .so, which never resolves when re-imported inside the Bun-compiled pi binary, leaving the reload screen stuck forever (pi awaits session_start handlers without a timeout). Cache the first import on globalThis so reloads reuse it. Closes #757 Co-authored-by: chenydev <chenydev@users.noreply.github.com>	2026-08-08 19:28:16 -07:00
dmtrKovalenko	b6f351d729	chore: bump fff-mcp release artifacts to v0.10.3	2026-08-07 04:55:45 +00:00
Dmitriy Kovalenko	e2cad2f09e	chore: release 0.10.3 Build & Publish / Build Neovim aarch64-apple-darwin (push) Has been cancelled Details Build & Publish / Build Neovim aarch64-linux-android (push) Has been cancelled Details Build & Publish / Build Neovim aarch64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build Neovim aarch64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build Neovim aarch64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build Neovim x86_64-apple-darwin (push) Has been cancelled Details Build & Publish / Build Neovim x86_64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build Neovim x86_64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build Neovim x86_64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-apple-darwin (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-linux-android (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build C FFI aarch64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build C FFI x86_64-apple-darwin (push) Has been cancelled Details Build & Publish / Build C FFI x86_64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build C FFI x86_64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build C FFI x86_64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build MCP aarch64-apple-darwin (push) Has been cancelled Details Build & Publish / Build MCP aarch64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build MCP aarch64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build MCP aarch64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build MCP x86_64-apple-darwin (push) Has been cancelled Details Build & Publish / Build MCP x86_64-pc-windows-msvc (push) Has been cancelled Details Build & Publish / Build MCP x86_64-unknown-linux-gnu (push) Has been cancelled Details Build & Publish / Build Python wheels aarch64 (macos-latest) (push) Has been cancelled Details Build & Publish / Build Python wheels x86_64 (macos-latest) (push) Has been cancelled Details Build & Publish / Build Python wheels x86_64 (windows-latest) (push) Has been cancelled Details Build & Publish / Build Python sdist (push) Has been cancelled Details Build & Publish / Build MCP x86_64-unknown-linux-musl (push) Has been cancelled Details Build & Publish / Build Python wheels aarch64 (ubuntu-latest) (push) Has been cancelled Details Build & Publish / Build Python wheels x86_64 (ubuntu-latest) (push) Has been cancelled Details Build & Publish / Release (push) Has been cancelled Details Build & Publish / Publish Python wheels to PyPI (push) Has been cancelled Details Build & Publish / Publish Rust crates (push) Has been cancelled Details Build & Publish / Publish npm packages (push) Has been cancelled Details	2026-08-06 20:26:51 -07:00
Gustav the Bot	4afad820b4	fix(nvim): refuse fs-root/home index at Lua level before FFI (#745 ) (#748 ) * fix(nvim): refuse fs-root/home index at Lua level before FFI (#745) Opening nvim at `/` with lazy=false crashed the whole neovim process on CI-cross-compiled aarch64 .so binaries: the init_file_picker FFI call SIGSEGVs instead of returning Error::FilesystemRoot cleanly. A SIGSEGV is a hardware signal that the pcall around init_file_picker cannot catch. Mirror the Rust refusal (file_picker.rs:862) in Lua and bail before crossing the FFI boundary, so the crashing path is never reached. Same guard added to change_indexing_directory for the :cd-into-root case. Honors enable_fs_root_scanning / enable_home_dir_scanning overrides. * chore(nvim): trim comments in fs-root/home refusal guard --------- Co-authored-by: gustav-fff <286169375+gustav-fff@users.noreply.github.com>	2026-08-06 20:25:43 -07:00
Gustav the Bot	7af3a9dab6	fix(pi-fff): make home-dir scanning configurable, warn when indexing $HOME (#743 ) (#749 ) Expose home-dir scanning as --fff-enable-home-scan flag + FFF_ENABLE_HOME_SCAN env, default true. resolveBoolOpt takes a fallback and accepts 0/false. Threaded through the main finder and AuxFinderPool. Cache os.homedir() once per process in src/paths.ts. AuxOpts.onHomeDirScan fires whenever the agent spawns an aux picker rooted at or above $HOME. On session start from $HOME, notify the user and track scan progress with a 1s poller that clears the footer once the scan settles; the interval is unref'd and cleared on shutdown and in destroyFinder(). Co-authored-by: gustav-fff <286169375+gustav-fff@users.noreply.github.com> Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: dmtrKovalenko <16926049+dmtrKovalenko@users.noreply.github.com>	2026-08-06 20:25:33 -07:00
Dmitriy Kovalenko	031005e227	fix: Reduce amount of rescans in giant /Users/neogoose like folders (#751 )	2026-08-06 19:47:41 -07:00