fix: add Windows .bat + expand .gitattributes to handle CRLF in *.sh and *.env#6
Open
sami-marreed wants to merge 5 commits into
Open
fix: add Windows .bat + expand .gitattributes to handle CRLF in *.sh and *.env#6sami-marreed wants to merge 5 commits into
sami-marreed wants to merge 5 commits into
Conversation
Prevents Windows clones (with core.autocrlf=true) from checking out *.sh files with CRLF, which breaks bash under WSL. Closes #74 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Expands the CRLF fix beyond the .gitattributes pin so existing Windows
clones can recover without re-cloning:
* .gitattributes now also pins *.env, *.yaml, *.yml, *.toml to LF
(only *.sh and *.bash were covered before, so *.env files like
appworld.env were still checked out as CRLF on Windows).
* Adds fix_line_endings.bat at repo root: a self-contained Windows
batch + PowerShell script that walks the repo and strips CRLF from
*.sh and *.env files, skipping .git/.venv/vendor/node_modules.
* Pins *.bat/*.cmd/*.ps1 to CRLF (cmd.exe expects CRLF in batch files).
* README: short Windows/WSL note pointing users at the .bat.
A bash self-heal prelude was attempted but is fundamentally unworkable:
bash cannot parse a script whose own keywords end in \r ("fi\r" != "fi"),
so the heal code never gets a chance to run.
Refs #74
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The exclusion regex used \\(...)\\ which only matches Windows backslash separators. Verified with pwsh on macOS: vendor/ and node_modules/ paths were silently being normalized because the regex never matched their forward-slash paths. Switch to [\\/](...)[\\/] so exclusions work whether the script is invoked under Windows cmd.exe (backslash paths) or pwsh on any OS (which is how this was discovered). Tested via pwsh on macOS against a fixture with .git/, .venv/, vendor/, node_modules/ paths -- all correctly skipped now. Refs #74 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Every .sh in the repo now has a sibling .bat (35 scripts + 1 shared _delegate_to_bash.bat helper) so Windows users have a path that doesn't require Git Bash for the simple cases. Two translation styles: - Pure cmd.exe ports (15 files): the simple wrappers and env-loaders — setup_cuga, setup_m3, setup_appworld, load_env, run_registry (helper + 4 per-benchmark stubs), model_profiles, viz, the analyze thin-stubs, all run_app/run_eval wrappers. These are usable on a vanilla Windows install. - bash-delegate shims (19 files): for the heavy scripts that use POSIX-only features (lsof, pkill, signal traps, process substitution, sourceable function libraries, embedded Python heredocs, mktemp, comm, find -mindepth/-maxdepth). Each shim is ~6 lines and calls _delegate_to_bash.bat, which tries Git Bash (well-known install paths) -> bash on PATH -> WSL -> friendly install instructions. benchmarks/helpers/common.bat is a placeholder noting that the bash function library it mirrors can't be sourced into cmd.exe; callers that delegate to bash source the .sh version directly. .secrets.baseline gets one new entry for the openai/gpt-oss-120b model name in scripts/model_profiles.bat (false positive — same string is already in scripts/model_profiles.sh; cmd.exe's lack of inline comment syntax means the standard pragma can't be embedded on the line, so the baseline is the cleaner workaround). The longer-term cleanup — move logic into Python so .sh and .bat both become ~5-line wrappers around `uv run python -m ...` — is tracked in research-rpa/cuga-internal-evaluation#88.
README changes: - Expand the Windows note in §Installation to mention that every .sh has a .bat sibling, and that the simple wrappers run on stock cmd.exe (no WSL/Git Bash needed) while the heavy scripts delegate to bash. - New §Running on Windows in Quick Start with cmd.exe/PowerShell examples mirroring the bash examples above it. - Reference the smoke test and the long-term Python migration (#88). scripts/test_bat_scripts.ps1 — new pwsh 7 smoke test that verifies structural invariants of the .bat layer: 1. every in-scope .sh has a sibling .bat 2. every .bat starts with `@echo off` 3. every .bat has an `exit /b` terminator 4. every delegate shim points to an existing .sh 5. every shim's `_delegate_to_bash.bat` exists where expected It doesn't execute the .bat files (cmd.exe isn't available on mac/linux), but it caught a real bug on its first run — three orphan `analyze.bat` files delegating to .sh files that don't exist on this branch (they were added on master after this branch was cut). Test runs on any host with pwsh: `pwsh scripts/test_bat_scripts.ps1`. Orphan deletions: benchmarks/appworld/analyze.bat, benchmarks/m3/analyze.bat, scripts/analyze.bat. Their .sh counterparts will arrive when master is merged in; the .bat files can come back at that point. .secrets.baseline auto-updated by the detect-secrets hook to refresh the timestamp after the line-number shift in scripts/model_profiles.bat.
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
2 participants
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.
Summary
.gitattributespinning*.sh,*.bash,*.env,*.yaml,*.yml,*.tomlto LF (only*.sh/*.bashwere covered in the first commit, soappworld.envwas still checked out as CRLF on Windows).*.bat/*.cmd/*.ps1to CRLF (cmd.exe expects CRLF in batch files).fix_line_endings.batat the repo root: a Windows batch + PowerShell script that walks the repo and strips CRLF from*.shand*.envfiles (skipping.git,.venv,vendor,node_modules). This recovers existing Windows clones without forcing a re-clone..bat.Why not a bash self-heal prelude?
I tried this first and it doesn't work: bash cannot parse a script whose own keywords end in
\r(fi\r≠fi), so aif ... fiself-heal block fails with a parse error before it can run. Even a single-line prelude fails. The issue must be fixed from a tool that doesn't choke on CRLF in its own source — hence the.bat.Migration
.gitattributesensures LF on checkout.fix_line_endings.batonce (double-click or from cmd/PowerShell), then runsetup_*.shunder WSL.Test plan
fix_line_endings.batpresent after merge (master)setup_cuga.sh/appworld.env: double-clickfix_line_endings.bat, then runsetup_cuga.shunder WSL — confirm both succeedsetup_cuga.shandbenchmarks/appworld/config/appworld.envalready have LF (no.batrun needed)