fix(tools): write_file preserves the existing file's line-ending style (CRLF)

[EvanProgramming]

Summary

Fixes #5909. write_file silently converted an existing CRLF (Windows) file to LF on overwrite, while edit_file preserves the detected line ending — so the same file behaved differently depending on the tool used. Both write paths were affected:

• WriteFileTool::execute (full replace)
• WriteFileTool::execute_contract_write (the hidden write_file / File contract path)

Root cause

Both paths wrote file_content.as_bytes() verbatim. Model-produced content is LF, so every \r\n in the on-disk file became \n — changing byte content the model never intended to touch, corrupting diffs/blame on CRLF-committed repos and on Windows.

Changes

• New helper preserve_prior_line_endings(content, prior): normalizes incoming content to LF internally, then restores the prior file's detected ending (contract_line_ending), exactly mirroring edit_file's policy.
• Both write paths call it. A brand-new file (no prior content) is written verbatim — there is no style to honor, so creation behavior is unchanged.

Tests

• E2E: write_file over an existing CRLF file with LF content leaves the file CRLF.
• Helper unit coverage: CRLF prior → CRLF out; LF prior → LF out; empty prior → verbatim (including CRLF input).

Verification

• cargo check -p codewhale-tui --lib --tests passes.
• cargo fmt -p codewhale-tui clean (only the touched files changed).

[github-actions]

Thanks @EvanProgramming for taking the time to contribute.

This repository is observing a maintainer-managed PR intake gate in dry-run mode, so this pull request is staying open. This note helps maintainers prepare the allowlist before any enforcement is considered.

Please read CONTRIBUTING.md for the expected contribution shape. A maintainer can grant recurring PR access by commenting /lgtm on a pull request.

[Hmbown]

Verified the bug on main — both write paths wrote content verbatim (crates/tui/src/tools/file.rs:1388, :1507) while edit_file preserved endings, so parity here is exactly right, and thanks for updating the diff/summary to use written so "(no changes)" stays truthful. Two small follow-ups: (1) normalize_contract_line_endings (file.rs:1568) also rewrites bare \r to the detected ending, so a lone CR in the new content becomes CRLF when the prior is CRLF — same policy as edit_file, but a unit test would pin it; (2) the contract path (execute_contract_write, file.rs:~1358) is only covered via the shared helper — one E2E test there would close the loop. Housekeeping: the commit is missing its Signed-off-by: trailer; mind amending with git commit --amend -s? Nice work, and thanks for the well-written #5909.

[EvanProgramming]

Thanks for the review — all three follow-ups are addressed in the updated commit (5693fbd4d):

1. Lone-\r policy pinned: preserve_prior_line_endings_keeps_the_prior_style now asserts that a bare CR in the incoming content normalizes like edit_file (lone \r → \n → re-emitted as the prior ending: CRLF prior → \r\n, LF prior → \n).
2. Contract-path E2E added: contract_write_over_crlf_file_preserves_crlf_line_endings drives WriteFileTool::execute_contract_write over an existing CRLF file and asserts the on-disk result stays \r\n.
3. Signed-off-by: trailer added via git commit --amend -s.

Verification: cargo check -p codewhale-tui --lib --tests passes and cargo fmt -p codewhale-tui is clean.