Skip to content

Spec self-check: the "changed paths" and "pins removed" numbers drift from the real diff #3218

Closed

Description

@Birfy

Written-by: Chief

What happens

Twice in a row the issue-spec comment published per-file numbers that did not survive contact with the implementation:

  1. Cut 20 (#3114 comment 7007863808): A10 says git status --porcelain | wc -l must be 13 (hand 7 + generated 6). The PR that was actually built from that spec (#3216) touches 11 paths (hand 6 + generated 5): i18n-plugins is untouched, so there is no mirror for it, and patch regenerate rewrote only two patches/ds-bot files. The reviewer had to rule "11 is not a gap" in the PASS, and the same correction had to be repeated in Correction from Reviewer (7008826388) for the Share / Share pins.
  2. The pin count itself: the spec said 13 pins would go; the correction raised it to 23; then the reviewer found Share had to go while Share stays, because Share: '分享' lands in the cut and the ratchet refuses a pin whose key is translated.

Why it matters

  • A reviewer following the spec literally writes Verdict: INC for a difference that is not a defect, which costs a round.
  • The numbers are the only cheap way to check a 250-line budget; if they drift, the budget check is decorative.
  • Both drifts are mechanical, so a checklist step catches them.

Candidate fix

Before posting the spec, the run recomputes, on the same scratch tree:

  • git diff --name-only | sort → path count, split into hand vs generated by the generatedPaths rule;
  • per-file git diff --numstat, summed into the hand budget;
  • for every pin it plans to remove, that the key really leaves the baseline (grep -c), not that it is listed in the plan.

Then the spec quotes the measured numbers, not the planned ones.

Acceptance

  • A spec comment's "changed paths" equals the path count of the PR built from it, or the spec states why generated files differ.
  • No spec ships a pin count that a grep -c on the same tree contradicts.

Tier: T1 (docs/tooling only).

Activity

  1. added
    bugSomething isn't working
    needs-humanThe agent loop stopped; a person must decide (reason in the last comment)
    area:automationScheduled tasks and automation
    on Nov 19, 2026
  2. Birfy commented on Nov 19, 2026

    @Birfy
    CollaboratorAuthor

    Written-by: Chief

    已落进流程模板,关闭:第二十一/二十二刀的 brief 里已经写死「贴规格前在同一棵树重算改动路径数、逐文件 numstat、逐条 pin 用 grep -c 验」,并要求把实测值写进规格;规格线这两刀的实测数字与 PR 一致(cut21 215 行 / 6 文件、cut22 229 行 / 9 文件)。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    area:automationScheduled tasks and automationbugSomething isn't workingneeds-humanThe agent loop stopped; a person must decide (reason in the last comment)

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions