fix(tests): stop git corrupting the decmpfs fixtures on Windows checkout - #8
Closed
h4x0r wants to merge 2 commits into
Closed
fix(tests): stop git corrupting the decmpfs fixtures on Windows checkout#8h4x0r wants to merge 2 commits into
h4x0r wants to merge 2 commits into
Conversation
The shared workflow's `deny-config-repo` input defaulted to SecurityRonin/fleet-config, so adopting it silently replaced this repo's supply-chain policy with the shared one. That is a real change to what `cargo deny` accepts, and no adoption PR disclosed it — nobody involved knew. Measured across the fleet when it was found: 61 of 65 adopted repos had a stricter local config. The shared one carries 21 advisory ignores against their `ignore = []`, every one a bare RUSTSEC id with no reason and no removal condition — which the fleet's own suppression rule forbids. Among them RUSTSEC-2023-0071 (rsa, Marvin timing attack). Several repos also went from `bans.multiple-versions = "deny"` to `"warn"` and gained nine allowed licences. fleet-ci now defaults to the repository's own deny.toml; opting into the shared config is explicit. This repin picks that up, restoring the policy this repo actually wrote. Reassuring rather than alarming: before this fix all 65 adopted repos were re-checked against their OWN deny.toml and all 65 passed. The gate was weakened, but nothing was hiding behind it. If this repin does turn a check red, that is a true finding this repo's own policy always meant to catch — fix it rather than re-pointing at the shared config. Only the pinned SHA changes.
Seven decmpfs tests failed only on windows-latest. Decompression itself was fine -- the output was the expected perl script -- so the corruption was in the fixture bytes, not the decoder. Cause: the repo had no .gitattributes, so git applied its text/binary autodetection to byte-exact forensic evidence. core.autocrlf is true by default on windows-latest, so every file git judged "text" had its LF bytes rewritten to CRLF at checkout, and the tests then compared against corrupted evidence. Git scans only the first 8000 bytes for a NUL when making that call, so this was not confined to the .expected files -- it also hit a compressed INPUT whose first NUL falls past that window. Comparing a simulated windows-latest checkout against the stored blobs shows 7 of 15 decmpfs fixtures altered: lzvn.expected 80000 -> 87855 zlib.expected 150000 -> 164122 real_zlib_inline.expected 2000 -> 2218 zlib_inline.expected 3000 -> 3325 tahoe_type8.expected 1936 -> 1994 zlib_type3_stored.payload tahoe_type9.expected 63 -> 71 3001 -> 3326 Marking the data trees -text fixes the class rather than the seven failing tests, and covers fixtures added later. No test or source file is touched. Verified against git's own mechanism rather than by waiting on CI, since the mechanism is git's and not the OS's -- `git -c core.autocrlf=true clone` reproduces the runner's checkout on any platform: before 7 fixtures differ from their stored blobs; 19 passed, 7 failed after 0 fixtures differ; 26 passed, 0 failed The seven altered fixtures correspond one-to-one with the seven failing tests. Both runs use --no-fail-fast: without it cargo abandons the remaining targets after the first failure, and the broken tree reports 19 tests against the fixed tree's 26, which would have compared two different test sets. Corruption is detected by comparing each file against `git cat-file blob`, NOT by counting CRLF sequences -- compressed data legitimately contains 0D 0A, and a counting check falsely flagged lzvn.rsrc and real_zlib_rsrc.rsrc, which are intact. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Contributor
Author
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Seven decmpfs tests failed only on windows-latest. Decompression itself was
fine -- the output was the expected perl script -- so the corruption was in
the fixture bytes, not the decoder.
Cause: the repo had no .gitattributes, so git applied its text/binary
autodetection to byte-exact forensic evidence. core.autocrlf is true by
default on windows-latest, so every file git judged "text" had its LF bytes
rewritten to CRLF at checkout, and the tests then compared against corrupted
evidence.
Git scans only the first 8000 bytes for a NUL when making that call, so this
was not confined to the .expected files -- it also hit a compressed INPUT
whose first NUL falls past that window. Comparing a simulated windows-latest
checkout against the stored blobs shows 7 of 15 decmpfs fixtures altered:
lzvn.expected 80000 -> 87855 zlib.expected 150000 -> 164122
real_zlib_inline.expected 2000 -> 2218 zlib_inline.expected 3000 -> 3325
tahoe_type8.expected 1936 -> 1994 zlib_type3_stored.payload
tahoe_type9.expected 63 -> 71 3001 -> 3326
Marking the data trees -text fixes the class rather than the seven failing
tests, and covers fixtures added later. No test or source file is touched.
Verified against git's own mechanism rather than by waiting on CI, since the
mechanism is git's and not the OS's --
git -c core.autocrlf=true clonereproduces the runner's checkout on any platform:
before 7 fixtures differ from their stored blobs; 19 passed, 7 failed
after 0 fixtures differ; 26 passed, 0 failed
The seven altered fixtures correspond one-to-one with the seven failing
tests. Both runs use --no-fail-fast: without it cargo abandons the remaining
targets after the first failure, and the broken tree reports 19 tests against
the fixed tree's 26, which would have compared two different test sets.
Corruption is detected by comparing each file against
git cat-file blob,NOT by counting CRLF sequences -- compressed data legitimately contains 0D 0A,
and a counting check falsely flagged lzvn.rsrc and real_zlib_rsrc.rsrc, which
are intact.
Co-Authored-By: Claude Opus 5 noreply@anthropic.com