Skip to content

fix(helm): remove ':' from test fixture paths so repo is cloneable on Windows#201

Open
Mandark-droid wants to merge 1 commit into
evaleval:mainfrom
Mandark-droid:fix/windows-colon-fixture-paths
Open

fix(helm): remove ':' from test fixture paths so repo is cloneable on Windows#201
Mandark-droid wants to merge 1 commit into
evaleval:mainfrom
Mandark-droid:fix/windows-colon-fixture-paths

Conversation

@Mandark-droid

@Mandark-droid Mandark-droid commented Jul 7, 2026

Copy link
Copy Markdown

Fixes #200

Problem

The HELM test fixtures use HELM run-spec names verbatim as directory names, and those contain : — an illegal character in Windows filenames. As a result, git clone fails at checkout on Windows:

error: invalid path 'tests/data/helm/commonsense:dataset=hellaswag,method=multiple_choice_joint,model=eleutherai_pythia-1b-v0/per_instance_stats.json'
fatal: unable to checkout working tree

Change

  • Rename the three fixture directories under tests/data/helm/, replacing : with _ (pure renames, 100% similarity — file contents untouched).
  • Update the 17 path references: HELM test files, the example command in every_eval_ever/converters/README.md, and the default --log_path in every_eval_ever/converters/helm/__main__.py.

Nothing at runtime parses these directory names — the adapter reads run_spec.json from inside each run directory — so real HELM output directories (which do contain : on Linux/macOS) are unaffected.

Verification

On Windows 11 + Python 3.12 (where this repo previously could not even be checked out):

  • pytest tests/test_helm_adapter.py tests/test_helm_instance_level_adapter.py19 passed, including all 16 fixture-dependent tests.
  • The README example command (every_eval_ever convert helm --log_path tests/data/helm/commonsense_dataset=...) converts successfully.

(Aside noticed while testing, unrelated to this change: on Windows the test run needs PYTHONUTF8=1 because crfm-helm reads files with the locale encoding, and on Python 3.14 crfm-helm fails to import entirely due to its pydantic-v1 usage — the tests then skip gracefully.)

A follow-up could add a CI check rejecting committed paths containing characters invalid on Windows (: * ? " < > |) to prevent regressions — happy to do that in a separate PR if wanted.

… Windows

The HELM test fixture directories used HELM run-spec names verbatim as
directory names, which contain ':' — an invalid character in Windows
filenames. This made `git clone` fail at checkout on Windows
(error: invalid path 'tests/data/helm/commonsense:dataset=...').

Rename the three fixture directories, replacing ':' with '_', and update
the path references in the HELM tests, the converters README example, and
the helm converter's default --log_path. File contents are unchanged
(pure renames); nothing at runtime parses these directory names — the
adapter reads run_spec.json from inside each run directory.

Fixes evaleval#200
@Mandark-droid Mandark-droid force-pushed the fix/windows-colon-fixture-paths branch from bcbfdc6 to 7679e88 Compare July 7, 2026 21:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Windows: git clone fails because HELM test fixture directories contain ":" (invalid NTFS character)

1 participant