github-featured/triggerdotdev--trigger.dev
2
0
Fork 0
Code Issues Pull Requests Deployments Models Agent Notes Packages Projects Releases Wiki Activity

fix: keep all of a PR's changesets in the release PR summary (#4031)

Browse Source 
## Summary

The auto-generated changeset release PR (`changeset-release/main`)
builds its `## Improvements` / `## Bug fixes` summary with
`scripts/enhance-release-pr.mjs`. The script deduplicated summary
entries by PR number, so when a single PR shipped more than one
changeset, only the first entry survived and the rest were silently
dropped from the summary. The dropped entries still appeared in the raw
`<details>` block, which is how the mismatch surfaced (for example in
[#3998](https://github.com/triggerdotdev/trigger.dev/pull/3998), where
one PR's four changesets showed up as a single summary line).

## Fix

Deduplicate on the full entry text rather than the bare PR number. The
entry text embeds the PR link, so:

- the same changeset rendered once per package section still collapses
to one,
- distinct changesets from the same PR are each kept,
- identical descriptions from different PRs stay separate.

Verified against the raw changeset output from
[#3998](https://github.com/triggerdotdev/trigger.dev/pull/3998): that
PR's changesets went from 1 to all 4 in the generated summary.
This commit is contained in:
Eric Allam
2026-06-25 09:20:10 +01:00
committed by GitHub
parent 2fa84ea124
commit 422d8da339
1 changed files with 12 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/enhance-release-pr.mjs
+12 -10
View File
@@ -35,9 +35,12 @@ function parsePrBody(body) {
const entries = [];
if (!body) return entries;
// Deduplicate by PR number
// Deduplicate by entry content. A single changeset that targets multiple
// packages is rendered once per package section, so the same text repeats and
// we collapse it. But several distinct changesets from one PR have distinct
// text (and each still carries that PR's link), so keying on content keeps
// them all instead of dropping every entry after the first for that PR.
const seen = new Set();
const prPattern = /\[#(\d+)\]\(([^)]+)\)/;
// A standalone dependency-bump list item, e.g. "`@trigger.dev/core@4.5.0-rc.7`"
// or "trigger.dev@4.5.0-rc.7". These normally appear nested under
@@ -87,20 +90,19 @@ function parsePrBody(body) {
if (headLine.startsWith("Updated dependencies")) continue;
if (depBumpPattern.test(headLine)) continue;
// Deduplicate by PR number (the changeset link lives on the head line)
const prMatch = itemLines[0].match(prPattern);
if (prMatch) {
const prNumber = prMatch[1];
if (seen.has(prNumber)) continue;
seen.add(prNumber);
}
// Reconstruct the full item: head line + dedented continuation lines, so
// code blocks and sub-bullets survive. Continuation under a "- " item is
// indented 4 spaces; strip up to 4 to bring it back to the base level.
const continuation = itemLines.slice(1).map((l) => l.replace(/^ {1,4}/, ""));
const text = [headLine, ...continuation].join("\n").replace(/\s+$/, "");
// Deduplicate on the full entry text (which embeds the PR link). The same
// changeset echoed across package sections collapses to one, while multiple
// distinct changesets from a single PR are each preserved.
const dedupeKey = text.replace(/\s+/g, " ").trim();
if (seen.has(dedupeKey)) continue;
seen.add(dedupeKey);
// Categorize off the head line
const lower = headLine.toLowerCase();
let type = "improvement";