A link gate printed PASS over zero files because git ls-files ran against an enclosing repository

A PASS over an empty set is worse than no gate at all. The hook's output is pasted into the commit message and the pull-request body, where it is the record that the tree was checked ^[1]. With no gate, a reviewer has nothing to trust. With this one, a reviewer had a verdict word, an exit code and counts, and every count was 0 ^[3].

A faithful `git archive` export excludes `.git`. So `git ls-files` walked up to an enclosing repository, and that repository's index held nothing under the directory it was standing in: zero lines, exit 0, no error ^[4].

Naming the root and the command costs one line of output, and it removes an ambiguity that would otherwise be permanent. `git ls-files` returning nothing means nothing until you know which directory it ran in and against which repository ^[6].

Each instrument's self-test grew the case asserted from both sides: the set is empty and it prints NOT RUN, the set is non-empty and it does not ^[8]. The four counts went 9/9 to 12/12, 17/17 to 20/20, 14/14 to 17/17 and 40/40 to 41/41 ^[9]. That is 80 assertions becoming 90 ^[10]. Three instruments needed three new cases each and the fourth needed one ^[10].

A third verdict is no use on the second case in the same codebase. A round's commit message and pull-request body both recorded `NUMGATE: PASS - sites=1023 carrier=410 package=613` ^[11]. A re-run at the head the receipt named printed 1025, 412 and 613, exiting 0 both plain and with `--check` ^[12]. The base's real figure was 1021/408 and the head's was 1025/412, with the recorded pair being the base's carried forward and the expected delta applied by hand ^[13]. That put the record two below the head on both figures and two above the base ^[15].

The pre-commit gate piped the three tools through `tail -6`, and that drops the tool's third line, the `sites=` line, because it falls outside the last six lines of a longer output ^[14]. The person writing the receipt had no line to copy. A reader could check that receipt against the posted text character for character and learn nothing about the state of the tree ^[16]. According to the post, any filter in front of a report chooses which figures exist, and the figures outside the window are never produced at all ^[17].

The third case is a count that was right when it ran. A disclosure comment stated the population of the project's own review threads, 188 comments and "the other 185", and it was posted on one of those threads ^[18]. Re-taken with the same instrument at that round's close, the figures were 189/3/186 ^[19]. The count was correct when it was computed and wrong when it was read, and posting the comment is the act in between ^[20]. The post draws a rule from this. A count over a live set either names the state it holds of, or it is re-taken after the write by an act that is not a member of the set it states ^[21]. Every "X users, Y open PRs, Z commits" line on a project page has the same problem whenever the page itself is one of the things counted ^[22].