Improve squash parsing to avoid linking issues/PRs that were already shipped

src/extractors.test.ts

@@ -546,14 +546,132 @@ describe("getRevertMessageDepth", () => {
   });
 });
 
+describe("squash sub-commit blocks", () => {
+  // `git merge --squash` followed by `git commit` writes a body that begins
+  // with a "Squashed commit of the following:" header and dumps every commit
+  // pulled in via the squash, including upstream commits merged into the
+  // feature branch. Those references describe branch history, not the change
+  // landing here, so they must not be re-attributed to this release.
+
+  it("ignores PR refs inside a squash sub-commit dump (no real title)", () => {
+    const message = `Squashed commit of the following:
+
+commit aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+
+    Older shipped PR title (#85)
+
+    Fixes LIN-50`;
+    expect(extractPullRequestNumbersForCommit({ sha: "abc", message })).toEqual([]);
+    expect(ids(extractLinearIssueIdentifiersForCommit({ sha: "abc", message }))).toEqual([]);
+  });
+
+  it("ignores PR refs inside a squash dump but keeps a real title prepended", () => {
+    const message = `New dashboard widget (#100)
+
+Squashed commit of the following:
+
+commit aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+
+    Older shipped PR title (#85)
+
+    Fixes LIN-50
+
+commit bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb
+
+    Bug fix for graph render
+
+    Fixes LIN-100`;
+    // Title's PR # is the legitimate one; body's #85 must not leak.
+    expect(extractPullRequestNumbersForCommit({ sha: "abc", message })).toEqual([100]);
+    // LIN-100 is in a sub-commit body — also nested history. With the squash
+    // dump stripped, neither LIN-50 (already shipped) nor LIN-100 (whose
+    // attribution belongs to a different commit) are re-extracted from here.
+    expect(ids(extractLinearIssueIdentifiersForCommit({ sha: "abc", message }))).toEqual([]);
+  });
+
+  it("keeps magic-word refs from the PR description body above the squash dump", () => {
+    const message = `New dashboard widget (#100)
+
+Closes LIN-100
+
+Squashed commit of the following:
+
+commit aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+
+    Fixes LIN-50`;
+    expect(extractPullRequestNumbersForCommit({ sha: "abc", message })).toEqual([100]);
+    expect(ids(extractLinearIssueIdentifiersForCommit({ sha: "abc", message }))).toEqual(["LIN-100"]);
+  });
+
+  it("does not extract PR # from body cross-reference like 'builds on #85'", () => {
+    const message = `Add settings page (#100)
+
+This builds on #85 and #87. Closes LIN-200.`;
+    expect(extractPullRequestNumbersForCommit({ sha: "abc", message })).toEqual([100]);
+    expect(ids(extractLinearIssueIdentifiersForCommit({ sha: "abc", message }))).toEqual(["LIN-200"]);
+  });
+
+  it("preserves a user-authored footer appended after the squash dump", () => {
+    // Default `git merge --squash` puts the dump at the top, so any footer
+    // (e.g., trailers like `Closes LIN-X`, `Co-authored-by: ...`) typically
+    // ends up below the dump after the developer edits the message.
+    const message = `Squashed commit of the following:
+
+commit aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+Author: Dev <dev@example.com>
+Date:   Wed May 6 16:45:34 2026 +0000
+
+    Edge cases
+
+commit bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb
+Author: Dev <dev@example.com>
+Date:   Wed May 6 16:45:34 2026 +0000
+
+    Implement search filter
+
+Closes LIN-200`;
+    expect(ids(extractLinearIssueIdentifiersForCommit({ sha: "abc", message }))).toEqual(["LIN-200"]);
+  });
+
+  it("preserves both a prepended title and an appended footer around the dump", () => {
+    const message = `Search filter (#100)
+
+Squashed commit of the following:
+
+commit aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
+
+    Older shipped PR title (#85)
+
+    Fixes LIN-50
+
+Closes LIN-200`;
+    expect(extractPullRequestNumbersForCommit({ sha: "abc", message })).toEqual([100]);
+    expect(ids(extractLinearIssueIdentifiersForCommit({ sha: "abc", message })).sort()).toEqual(["LIN-200"]);
+  });
+
+  it("handles the marker without recognizable commit blocks below", () => {
+    const message = `Squashed commit of the following:
+
+Closes LIN-200`;
+    expect(ids(extractLinearIssueIdentifiersForCommit({ sha: "abc", message }))).toEqual(["LIN-200"]);
+  });
+
+  it("does not strip squash blocks from revert add-extraction (revert path is already blocked)", () => {
+    // A revert message that wraps a squash dump shouldn't add anything.
+    const message = `Revert "Squashed commit of the following:
+
+    Fixes LIN-50"`;
+    expect(ids(extractLinearIssueIdentifiersForCommit({ sha: "abc", message }))).toEqual([]);
+  });
+});
+
 describe("extractPullRequestNumbersForCommit", () => {
   // Messages that should extract PR numbers
   it.each([
     ["For issue #111, fix (#123)", [123], "squash merge skips fallback"],
     ["Fix #124 with better handling", [124], "hash in middle of title"],
     ["Merge pull request #431 from org/branch", [431], "GitHub merge format"],
     ["Merge pull request #42 from owner/branch\n\nDescription", [42], "merge with multiline description"],
-    ["Fix bug\n\nRelated to (#999)", [999], "PR in commit body"],
   ])("message %j should yield %j (%s)", (message, expected) => {
     const result = extractPullRequestNumbersForCommit({ sha: "abc", message });
     expect(result).toEqual(expected);
@@ -577,15 +695,15 @@ describe("extractPullRequestNumbersForCommit", () => {
     [
       "FLEX-2816: fix something\n\nTwo issues from cursor[bot] review #4211934690.",
       [],
-      "fallback drops oversized #NNN (cursor review id)",
+      "title has no PR number and body scan is no longer attempted",
     ],
     [
       "Fix something (#51876)\n\nTwo issues from cursor[bot] review #4211934690.",
       [51876],
-      "squash match keeps valid PR; fallback would have grabbed the oversized id but is skipped",
+      "squash match keeps valid PR; body cross-references are ignored",
     ],
-    ["Fix bug\n\nRelated to (#4211934690)", [], "fallback drops oversized parens form"],
-    ["Fix bug\n\nSee #123 and sentry #9999999999", [123], "fallback keeps small numbers and drops oversized"],
+    ["Fix bug\n\nRelated to (#4211934690)", [], "body fallback removed: cross-reference in body is ignored"],
+    ["Fix bug\n\nSee #123 and sentry #9999999999", [], "body fallback removed: numbers in body don't leak"],
     ["Title (#4211934690)", [], "squash format with oversized number is dropped"],
     ["Merge pull request #4211934690 from x/y", [], "merge format with oversized number is dropped"],
     [`Title (#${2_147_483_647})`, [2_147_483_647], "Int32 max is allowed"],

src/extractors.ts

@@ -29,6 +29,45 @@ const ISSUE_IDENTIFIER_REGEX = new RegExp(
 
 const LINEAR_ISSUE_URL_REGEX = /https?:\/\/linear\.app\/[\w-]+\/issue\/(\w{1,7}-[0-9]{1,9})(?:\/[\w-]*)*/gi;
 
+/**
+ * `git merge --squash` followed by `git commit` writes a body containing this
+ * header and then dumps the full message of every commit pulled in via the
+ * squash — including upstream history merged into the feature branch via
+ * `git merge`. Issue / PR references inside that dump describe branch history,
+ * not the change being squashed, so they must not feed release association.
+ *
+ * We excise *only* the dump itself: any real subject the developer prepended
+ * and any footer they appended (e.g. `Closes LIN-X`, `Co-authored-by: …`) are
+ * preserved. The dump is bounded by recognizable structural lines (`commit
+ * <sha>`, `Author:`, `Date:`, `Merge:`, blank lines, and indented body
+ * content); the first non-indented, non-empty line that doesn't match those
+ * patterns marks the start of user-authored footer content.
+ */
+const SQUASH_BLOCK_MARKER = /^Squashed commit of the following:/i;
+const SQUASH_COMMIT_HEADER = /^commit [0-9a-f]{7,40}\b/i;
+const SQUASH_METADATA_HEADER = /^(?:Author|AuthorDate|Commit|CommitDate|Date|Merge):\s/i;
+
+function stripSquashBlock(message: string): string {
+  const lines = message.split(/\r?\n/);
+  const markerIdx = lines.findIndex((l) => SQUASH_BLOCK_MARKER.test(l));
+  if (markerIdx === -1) return message;
+
+  let i = markerIdx + 1;
+  for (; i < lines.length; i++) {
+    const line = lines[i]!;
+    if (line === "") continue;
+    if (/^[ \t]/.test(line)) continue;
+    if (SQUASH_COMMIT_HEADER.test(line)) continue;
+    if (SQUASH_METADATA_HEADER.test(line)) continue;
+    break;
+  }
+
+  const before = lines.slice(0, markerIdx).join("\n").trimEnd();
+  const after = lines.slice(i).join("\n").trim();
+  if (before && after) return `${before}\n\n${after}`;
+  return before || after;
+}
+
 function normalizeLinearUrls(text: string): string {
   return text.replace(LINEAR_ISSUE_URL_REGEX, "$1");
 }
@@ -174,8 +213,10 @@ export function extractLinearIssueIdentifiersForCommit(commit: CommitContext): E
     }
   }
 
-  // Commit message: only extract when preceded by a magic word
-  const message = commit.message ?? "";
+  // Commit message: only extract when preceded by a magic word.
+  // Strip any squashed sub-commit dump first so references that came from
+  // already-merged branch history don't get re-attributed to this commit.
+  const message = stripSquashBlock(commit.message ?? "");
   if (message.length > 0) {
     for (const match of matchMagicWordIdentifiers(message)) {
       if (!found.has(match.identifier)) {
@@ -192,11 +233,11 @@ export function extractPullRequestNumbersForCommit(commit: CommitContext): numbe
     return [];
   }
 
-  const message = commit.message ?? "";
+  const rawMessage = commit.message ?? "";
 
   // Skip reverts - they reference the original PR, not a new one
-  if (/^Revert "/i.test(message)) {
-    verbose(`Skipping revert commit ${commit.sha} with message: "${message}"`);
+  if (/^Revert "/i.test(rawMessage)) {
+    verbose(`Skipping revert commit ${commit.sha} with message: "${rawMessage}"`);
     return [];
   }
 
@@ -207,6 +248,11 @@ export function extractPullRequestNumbersForCommit(commit: CommitContext): numbe
     return [];
   }
 
+  // Drop nested squash sub-commit dumps before scanning so `(#NNN)` references
+  // from already-shipped commits pulled in via `git merge` don't get attributed
+  // to this commit's release.
+  const message = stripSquashBlock(rawMessage);
+
   const prNumbers: number[] = [];
   const pushIfValid = (raw: string, source: string): void => {
     const number = Number.parseInt(raw, 10);
@@ -233,10 +279,13 @@ export function extractPullRequestNumbersForCommit(commit: CommitContext): numbe
     pushIfValid(mergeMatch[1]!, "merge format");
   }
 
-  // Only use fallback if no matches from squash/merge formats
+  // Fallback for non-canonical merge titles (e.g. a direct push that put the PR
+  // number somewhere other than the trailing parens). Restrict to the title line
+  // — scanning the body would re-pick up cross-references like "builds on #85"
+  // and stale references inside squashed-in sub-commit history.
   if (prNumbers.length === 0) {
-    for (const match of message.matchAll(/#(\d+)/g)) {
-      pushIfValid(match[1]!, "message scan");
+    for (const match of title.matchAll(/#(\d+)/g)) {
+      pushIfValid(match[1]!, "title scan");
     }
   }
 
@@ -311,7 +360,8 @@ export function extractRevertedIssueIdentifiersForCommit(commit: CommitContext):
   // Use magic-word gating on the inner message, same as the add path, to avoid
   // false positives from generic word-number tokens (e.g. "Bump v1-2 to v1-3").
   if (messageDepth % 2 === 1) {
-    for (const match of matchMagicWordIdentifiers(innerMessage)) {
+    const innerStripped = stripSquashBlock(innerMessage);
+    for (const match of matchMagicWordIdentifiers(innerStripped)) {
       if (!found.has(match.identifier)) {
         found.set(match.identifier, { identifier: match.identifier, source: "commit_message" });
       }

src/git.test.ts

@@ -588,16 +588,39 @@ describe("getCommitContextsBetweenShas", () => {
     expect(result[0]?.sha).toBe(repo.commits.first);
   });
 
-  it("should normalize commit message whitespace", () => {
+  it("should collapse horizontal whitespace but preserve newlines", () => {
     const result = getCommitContextsBetweenShas(repo.commits.first, repo.commits.first, {
       cwd: repo.cwd,
     });
     expect(result).toHaveLength(1);
-    // The first commit has "feat: add src file  with  extra  spaces" - multiple spaces should be normalized
-    expect(result[0]?.message).not.toMatch(/\s{2,}/);
+    // Multiple spaces in the subject should be collapsed
     expect(result[0]?.message).toBe("feat: add src file with extra spaces");
   });
 
+  it("should preserve newlines so extractors can distinguish title from body", () => {
+    // Standalone tempdir so the multiline body is independent of the shared fixture.
+    const cwd = mkdtempSync(join(tmpdir(), "linear-release-multiline-"));
+    try {
+      runGit("init", cwd);
+      runGit('config user.email "test@example.com"', cwd);
+      runGit('config user.name "Test User"', cwd);
+      writeFileSync(join(cwd, "file.txt"), "x");
+      runGit("add .", cwd);
+      runGit('commit -m "Add feature (#100)" -m "Closes LIN-200" -m "Co-authored-by: Other <other@example.com>"', cwd);
+      const sha = runGit("rev-parse HEAD", cwd);
+
+      const result = getCommitContextsBetweenShas(sha, sha, { cwd });
+      expect(result).toHaveLength(1);
+      expect(result[0]?.message).toBe(
+        "Add feature (#100)\n\nCloses LIN-200\n\nCo-authored-by: Other <other@example.com>",
+      );
+      // First line is the actual title (not the entire flattened body)
+      expect(result[0]!.message!.split("\n")[0]).toBe("Add feature (#100)");
+    } finally {
+      rmSync(cwd, { recursive: true, force: true });
+    }
+  });
+
   it("should return empty array when no commits in range", () => {
     // third..first is empty because first is an ancestor of third
     const result = getCommitContextsBetweenShas(repo.commits.third, repo.commits.first, {

src/git.ts

@@ -195,7 +195,9 @@ export function extractBranchNameFromMergeMessage(message: string | null | undef
  */
 function parseCommitChunk(chunk: string): CommitContext {
   const [sha, rawMessage, rawDecorations] = chunk.split("\x1f");
-  const message = (rawMessage ?? "").trim().replace(/\s+/g, " ");
+  // Collapse runs of horizontal whitespace, but keep newlines so downstream
+  // extractors can tell the title from the body and skip nested commit blocks.
+  const message = (rawMessage ?? "").trim().replace(/[ \t]+/g, " ");
   const branchName = extractBranchNameFromMergeMessage(message) ?? extractBranchName(rawDecorations);
 
   return { sha: sha.trim(), branchName, message };