Fingerprint numpy arrays by shape, dtype and contents - #19
Merged
Conversation
Arrays previously fell through to generic instance handling, which compares by repr and so reported equal arrays as different. Elements are routed back through fingerprint() rather than compared as raw floats, which keeps one answer to the tolerance question whatever the container holds, and names nan/inf the way the scalar path already does. Comparing raw floats would make nan equality depend on json.loads returning the same NaN singleton - true today, and a false positive on every nan array the moment the serialiser changes. No rounding tolerance. np.round(x, 12) is an absolute cut, so a relative error of 1e-15 compared equal at magnitude 1 and different at 1e6 and above, and it disagreed with how the same float is treated outside an array. Tolerance is a policy decision for every type at once, not one container. numpy is located through sys.modules instead of being imported, so a project that never uses it does not pay to load it; one that does has already imported it by the time its functions are replayed. Closes #1
LuShadowX
added a commit
that referenced
this pull request
Aug 8, 2026
* Bring the README's platform, comparator and overhead claims up to date Windows is tested in CI as of #20, numpy arrays are compared as of #19, and recording has cost ~8x rather than 15-20x since the back-off work. * Let check compare two arbitrary refs check took one ref and always compared it against the working tree, so reviewing someone else's branch or auditing a release meant checking it out first. A second optional positional ref is now the candidate: nodrift check HEAD~1 HEAD With no second ref the working tree is still the candidate, which is the common case. Closes #11 * Strip compiled bytecode from staged trees A repository with __pycache__ committed hands the replay bytecode compiled from other source. Python runs it in preference to the file git archive just materialised, so both versions behave identically and check reports no behaviour change while the code genuinely differs. Found while testing two-ref comparison: a tree staged from a commit containing POSITIVE replayed as positive, and check called it clean. The one-ref path had the same hole. Silent wrongness rather than an error, which is the worst failure this tool has. Compiled artifacts are now removed from every staged tree, exported or copied. The regression test commits stale bytecode deliberately and fails without the fix.
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.
Arrays previously fell through to generic instance handling, which compares
by repr and so reported equal arrays as different.
Elements are routed back through fingerprint() rather than compared as raw
floats, which keeps one answer to the tolerance question whatever the
container holds, and names nan/inf the way the scalar path already does.
Comparing raw floats would make nan equality depend on json.loads returning
the same NaN singleton - true today, and a false positive on every nan array
the moment the serialiser changes.
No rounding tolerance. np.round(x, 12) is an absolute cut, so a relative
error of 1e-15 compared equal at magnitude 1 and different at 1e6 and above,
and it disagreed with how the same float is treated outside an array.
Tolerance is a policy decision for every type at once, not one container.
numpy is located through sys.modules instead of being imported, so a project
that never uses it does not pay to load it; one that does has already
imported it by the time its functions are replayed.
Closes #1