docs(tests): stop calling pull requests issues in the const suites

[JarryShaw]

Please follow the guide below

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort)
• make test passes, and a test case covers the change
• Added a changelog entry under docs/source/changelog/ and regenerated CHANGELOG.md, if the change is user-visible — N/A, centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

What is the purpose of your pull request?

• fix — corrects a defect
• feat — adds a feature
• perf — changes performance, not behaviour
• refactor — changes neither behaviour nor performance
• test — tests only
• docs — documentation only
• ci — workflows or build tooling
• chore — anything else

Part of #719.

---

Description of your pull request and other information

16 sentences under tests/const/ described a pull request as a "GitHub issue". tests/ is exempt from the rule preferring the issue over the pull request (conventions/documentation.rst:203) — but no exemption makes calling a pull request an issue true, and that is the whole of this change.

All 18 changed lines are in-place replacements. Nothing is reflowed, so the marker-merge and inline-literal-split hazards are structurally absent rather than merely checked for; max width, the count of lines over 95 and the reStructuredText marker count are byte-identical per file.

Two of the sixteen were false twice over, not merely mislabelled.

test_const_enum_builtin_parity.py:1138 said #803 added a test. It did not. git log -S on that def finds it first in fc32d1b81, whose subject is PR #677's squash merge; #803's own commit carries the line only as diff context and adds two different tests. Relabelled to PR #677.

Six sites credited "#775/#847's ruling" with converting or sparing mint sites. #841 is the issue #847 landed against, not #775, and it converted nothing — it reordered Socket._missing_ range branches. So PR #847 converted… would restate the same false claim in correct grammar, and #775/#841 fails differently, because the criterion is not in #841's text either: conventions/mint-criterion.rst:52-56 records it as settled in review of #841's branch-order fix and reaffirmed on #775, which puts it in #847's review thread — exactly the case the tests/ exemption exists for. The false noun is dropped and the bare pair kept, matching six siblings already written that way in this directory and test_const_enum_no_mint.py:28's "settled on PR #847 and confirmed on #775".

Eight further candidates were left alone because they already label the pull request correctly — "GitHub issue #862, fixed by #865", "blocked on #859", "the PR #836 ruling" and similar. Separating them needs a scan keyed on the citation run the phrase actually governs, not on any number appearing nearby; the looser method reports 24 sites where 16 are real.

No test pin needed changing — no assertion anywhere in tests/ quotes a string this touches. The seven changed files pass: 244 passed, 40001 subtests, exit code 0 read from the process.

One site outside this tranche, for whoever takes it: tests/foundation/registry/test_protocols.py:535 calls PR #815 an issue.

[JarryShaw]

GOOD TO GO at 7a365d12a — sonnet cross-review, a different model from the author. One real defect, and it was in this description rather than the diff; fixed above, with no new head needed.

The description said "#847 closes #841" and GitHub parsed that as a closing keyword. This pull request was listed under closingIssuesReferences: [841], and #841's own closedBy showed both #847 and #999 — so my prose credited this change with closing an issue it has nothing to do with. #841 was already closed, so nothing broke; the provenance was simply wrong. Reworded to "the pull request that fixed #841", and the reference is gone.

The contestable call holds. I asked the review to attack the #847 decision hardest, and it independently confirmed each limb: #847 closes #841 and reordered Socket._missing_ range branches only, so PR #847 converted… would be false; and the criterion was settled in #847's own comment thread, which it located at 13:06Z, consistent with mint-criterion.rst:52-56 recording it as settled in review of #841's fix. It also found the precedent stronger than I claimed — nine or more bare #775/#847 sites already exist, not six. Its verdict on the outcome: keep the bare pair.

The count that matters is confirmed by a different method. It typed all distinct numbers cited in tests/const/ in one GraphQL call and scanned every issue(s) #N[/#M] run line-wrap-safely: 16 "issue"-labelled references to a pull request at base 3c59e2917, and 0 at head — across 783 ×1, 803 ×2, 815 ×1, 847 ×6, 865 ×2, 907 ×2, 913 ×2. It could not reproduce two of my figures: it counts 61 distinct numbers where I said 49, and did not rebuild the looser 24-site scan. Those are unreconciled and I am not claiming otherwise; the before-and-after figure, which is the one the change rests on, agrees.

It also checked the #677 relabel at the level that matters — not just that #677 is the right number, but that the claim is true of it: fc32d1b81 is #677's squash merge, and it introduces BESPOKE_TEMPLATES along with the tcp-flags render test. One pre-existing imprecision it flagged and I am leaving: "the other half" is loose, since the test covers one module rather than two.

Four of the eight left-alone candidates spot-checked, all genuinely already correct. No collateral damage: 18 insertions and 18 deletions, identical line counts, max widths, over-95 counts and markup-character counts per file, the pre-existing role split at test_const_enum_get.py:98-99 unchanged on both sides, and no assertion in tests/ quoting an edited string. Tests re-run: 244 passed, 40001 subtests, exit 0 read from the process.

Label is now review: good-to-go. Not yet ready to merge — CI is still in flight.

[JarryShaw]

Correction to the comment above: my first fix did not work, and I said it had. I wrote that the stray closing reference was gone. It was not — it survived, because I replaced "closes #841" with "fixed #841", and fix/fixes/fixed is a closing keyword in its own right. I had swapped one trigger for another and then asserted the result without re-reading it.

The sentence now avoids the construction entirely — "#841 is the issue #847 landed against" — and a scan of the whole description for close[sd]?|fix(e[sd])?|resolve[sd]? immediately before a number returns nothing. Verified after the edit rather than assumed: this pull request's closingIssuesReferences is empty, and #841's closedByPullRequestsReferences is back to #847 alone where it briefly read "847, 999".

Worth recording because the trap is broader than the one word: GitHub's closing keywords are a set, not a word, so removing one and leaving another in the same sentence changes nothing, and the only way to know is to re-query the reference after editing. Nothing about the diff changed; this was description prose throughout.

[JarryShaw]

Ready to merge at 7a365d12a — closing the "not yet" on my verdict comment. CI finished clean: 69 success, 3 skipped, 0 in flight, 0 failures, mergeStateStatus is CLEAN, and all six required contexts are green — Required checks passed with Compat Python 3.10 through 3.14, and Compat Python 3.15 (scheduled) skipped.

One commit on main at 3c59e2917. Part of #719; the issue stays open for the rest of its sweep.

Unpublished decisions are yours, and I have not merged. This and #998 are both ready and independent of each other — #998 touches docs/source/conf.py and the conventions pages, this one only tests/const/ — so neither needs the other first. Note that whichever merges first will flip the other to BEHIND, because main's ruleset has strict required checks; that costs the second one a re-run but nothing else.

Still queued behind #998 specifically: the pcapkit/corekit/ citation tranche, 100 conversions already verified, which cannot become a pull request until the roles in #998's conf.py are on main — a converted docstring built against main today errors on an unknown role. And #989 carries a needs: decision on whether the 19 plain # comment citations in that tranche convert at all, which decides 243 sites tree-wide.