fix: credit co-authors of a PR in the generated changelog

[andygrove]

Which issue does this PR close?

There is no issue for this. It came up while generating the 1.1.0 changelog in #6282.

Rationale for this change

#6219 carries the commits from #6092 by Michael Taranov (mixermt), but the generated 1.1.0 changelog credits #6219 only to andygrove, and Michael is missing from the credits. The script credits each PR to the author of the commit that merged it, on the PR's line (commit.author.login) and in the credits (git shortlog). When a squash-merged PR has commits from more than one person, GitHub records the others only as Co-authored-by trailers on that commit, and neither of those reads them. The four backports on branch-1.1 (#6266, #6305, #6318 and #6323) have the same problem: they are credited to whoever opened the backport, and not to the author of the original change.

What changes are included in this PR?

For a PR whose merge commit has a Co-authored-by trailer, generate-changelog.py now fetches the PR's commits and credits the GitHub account of each commit's author, on the PR's line and in the credits. It leaves out:

• Merges of the base branch into the PR. Most of the trailers that name a maintainer come from these (the "Update branch" button), not from commits they wrote.
• Commits whose email is not linked to a GitHub account. Across the repository's history, every one of these is a second address of the PR's own author, or a machine-local identity such as Hung <hung@Hungs-Mac-mini.local>.
• Accounts that have never opened an issue or PR in the repository. GitHub links a commit to whichever account has its email address, so the sandbox identity test <a@b.c> in three of the 1.1.0 PRs resolves to an unrelated account, as do a guessed erik@users.noreply.github.com and a Cursor Agent commit. The script prints each account it leaves out this way to stderr, so the release manager can check them.

AI assistants that only appear as a Co-authored-by trailer on a commit are not credited, because only commit authors are.

In the credits, a co-author counts once for each such PR, under the name git shortlog already lists them under, and counts toward the number of contributors in the header. As in git shortlog, a name counts once per PR, so someone who committed from two accounts under the same name is not counted twice. The credits sentence now says that a PR with commits from more than one person counts for each of them.

Only PRs with a trailer need the extra calls: 107 of the 399 commits in the 1.1.0 range, plus one issue lookup per co-author account. The issue lookup reads the first page rather than totalCount, which PyGithub reports as 0 for this endpoint because GitHub pages it with a cursor.

How are these changes tested?

The release scripts have no test harness, so I generated changelogs before and after the change and compared them.

For 1.1.0 (1.0.0..branch-1.1 at 39779e3), the only differences are:

• fix: park the native scan loop instead of busy-polling while waiting on native I/O #6219 is credited to andygrove, mixermt, fix: refresh S3 policy locations when a location's credential fails #6223 to andygrove, snmvaughan and fix: decode invalid UTF-8 at the JVM to native FFI import boundary #5310 to manuzhang, andygrove, and the four backports now also credit dwsmith1983, liupoyi-1031 and parthchandra.
• The credits gain Michael Taranov (1), and Andy Grove, dustin, Parth Chandra, Steve Vaughan and liupoyi-1031 each gain the PRs they co-authored. Steve's lands on his existing Steve Vaughan line, although his commits say Steve Vaughan Jr. The header says 41 contributors instead of 40.
• cincrement (test <a@b.c>) on perf: vectorize the native map lookup behind element_at and GetMapValue #5806, ci: gate the Delta contrib build on symbols, not on libcomet size #5827 and ci: move the Spark 3.4/3.5/4.0, Iceberg, macOS and benchmark suites behind a merge queue #5843, erik on fix: keep Iceberg complex null checks on native scans #5732 and cursoragent on test: enable native path in lower/upper_enabled sql fixtures #5619 are left out, with a line on stderr for each.

To check the rules beyond one release, I applied them to all 430 PRs on main whose merge commit has a trailer. They credit 14 accounts, each of them someone who wrote commits in the PR. The issue-or-PR check leaves out 6: the three above, and three second accounts of the PR's own author. Generating 0.6.0 covers the two-account case: #1392 is credited to EmilyMatt, EmilyFlarionIO, and Emily Matheys still counts once for it.

[sunchao]

Summary

• Prior state and problem: Changelogs credited squash-merge authors while omitting contributors whose commits were included in those PRs.
• Design approach: pull_authors collects commit authors for PRs with co-author trailers. A Counter incorporates their contributions into release credits.
• Correctness / compatibility analysis: Live base/head comparisons verified attribution for #6219 and #6223. Additional API checks verified backport #6266 and two-account attribution in #1392. Spark semantics and supported-version behavior are unaffected by this release-tooling change.
• Key design decisions: Merge commits and unlinked identities are excluded. Additional accounts must have opened an issue or PR. Trailer-gated requests and cached participation lookups bound the added API work.
• Implementation sketch: The localized helper supplies author mappings to existing categorization and formatting. Credits reuse known names and deduplicate names per PR.
• Behavioral changes worth calling out: PR entries list multiple accounts, co-authors increase credit totals, and rejected accounts are reported to stderr.
• Suggested improvements: None meeting the P1/P2 reporting threshold. No introduced P1/P2 issues found within this review.

Reviewed the entire diff from e1d2c11729c2fc60a5def4e87bb17e5b28df2a29 to dfe0c8a2070e927d89d4ea22349fb5e08fbc5528. The PR is not a draft. Read AGENTS.md and applied review-comet-pr; no sibling skill applies. The snapshot and live GitHub checks contained no existing reviews, comments, or threads.

Exact-head CI: 7 checks passed, 15 jobs skipped, no failures. Builds and Spark/Iceberg suites were skipped.

Validation: Six disposable tests passed, alongside the live checks above, using PyGithub 2.10.0. Full-release changelog regeneration and JVM/native builds were not run. Project source remains unchanged.