[antonwolfy]

test_script_and_manifest_patterns_are_linear_on_hostile_strings[files2] failed on main in the py3.11 and py3.12 jobs (run): inspect_repo took 2.12 s and 2.36 s against a 2 s limit. The code under test hasn't changed since #66.

The case is a package.json script of "tsc " * 100_000, which guards the {0,256} bound in _NODE_TYPE_TOOL (dokimi/repos/environment/checks.py).

• With the bound, the cost is linear: 0.08 / 0.18 / 0.30 / 0.55 s at 25k / 50k / 100k / 200k on a workstation. The CI runners, with 4 xdist workers, are about 5–6× slower, which puts the 100k case at the limit.
• Without the bound, the pattern is quadratic: 0.06 / 0.27 / 0.94 s at 2k / 4k / 8k, about 150 s at 100k.

This PR raises the limit to 15 s, which still catches that regression by a wide margin. The test's other three cases are much cheaper.

test_manifest_readers_are_linear_on_repeated_tokens and test_makefile_rules_are_linear use the same 2 s limit. They have not failed, so this PR leaves them unchanged.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.38%. Comparing base (9ad5506) to head (bc4640d).
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #147   +/-   ##
=======================================
  Coverage   91.38%   91.38%           
=======================================
  Files         216      216           
  Lines       26722    26722           
  Branches     4373     4373           
=======================================
+ Hits        24419    24421    +2     
+ Misses       1637     1635    -2     
  Partials      666      666           

see 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[vchamarthi]

@exai-bot /review

[exai-bot]

Note

AI-generated review of bc4640dd8a06, requested by @vchamarthi, using the default review skill. It can be wrong: verify a finding before acting on it. React with a thumbs up or down on a finding to tell the bot whether it helped.
Model us.anthropic.claude-sonnet-5-5, effort default, 9 of 93 turns, 0.5 of 48 minutes, 230k in / 3k out tokens.

The change raises the time limit in test_script_and_manifest_patterns_are_linear_on_hostile_strings from 2 s to 15 s, because the "tsc " * 100_000 case sat close to 2 s on CI runners. Your own numbers back the choice: about 0.3 s on a workstation and roughly 5–6× that on CI with the bound, against about 150 s without it. That leaves a wide margin on both sides, so the flake fix is sound. I have one minor point about the other cases in the same test.

1 finding is on the changed lines, as comments below.

[exai-bot]

Minor (plausible, tests): 15 s limit is applied to every case, including the cheap ones

The 15 s limit applies to all four parametrized cases, but only the tsc case needs it. The rm --a×28 and git clean --a×28 cases are tiny inputs, and a regression there would show up as backtracking blow-up. That blow-up is less certain to be caught under a 15 s limit than under 2 s: a polynomial regression on a 28-token input could finish in under 15 s and pass. This is unconfirmed, because I could not run the patterns. Giving the tsc case its own limit would keep the other cases strict.

Suggested fix:

Add a limit parameter to the parametrization (for example (files, 15) for the tsc case and (files, 2) for the others), or move the tsc case into its own test with the 15 s limit.

[napetrov]

Fine as an unblock: the analysis (linear vs quadratic, ~150 s unbounded) shows 15 s still catches the regression.

Follow-up suggestion: an absolute wall-clock limit stays runner-dependent. Timing n and 2n and asserting the ratio < ~3 (linear ~2, quadratic ~4) is machine-independent. test_manifest_readers_are_linear_on_repeated_tokens and test_makefile_rules_are_linear have the same 2 s limit and are next to flake; worth converting all three together.