fix: don't treat a literal CR at EOF as a Windows line ending

[maximilliangrand]

Bug

applyPatch corrupts output when a Unix source is patched with a patch whose final line's content ends in a literal carriage return and has no trailing newline. The \r is silently dropped.

import{structuredPatch,applyPatch}from'diff';constoldFile='line1\nline2\n';constnewFile='line1\nline2\nline3\r';constpatch=structuredPatch('f','f',oldFile,newFile,undefined,undefined,{context: 0});applyPatch(oldFile,patch);// => 'line1\nline2\nline3' (the '\r' is lost)

I found this by fuzzing random source/target pairs through structuredPatch -> applyPatch; 80 of 200k pairs failed to round-trip, all of this shape.

Cause

isWin and winToUnix treat any patch line ending in \r as a Windows CRLF line ending. But a line ending in \r that is immediately followed by a \ No newline at end of file marker is not a CRLF ending (Windows endings are \r\n) - it's a literal carriage return in the final line's content. A genuine Windows no-newline-at-EOF line never ends in \r.

So isWin misclassifies such a patch as Windows, and when applyPatch (with autoConvertLineEndings on) converts it to match the Unix source, winToUnix strips the \r.

unixToWin already guards this exact case (hunk.lines[i + 1]?.startsWith('\\')), so it won't add a \r to a no-newline final line. This change adds the symmetric guard to isWin and winToUnix.

Fix

Ignore a trailing \r that sits immediately before a \ No newline at end of file marker when deciding whether a patch is Windows / when stripping CRs.

Added tests: a winToUnix unit test, an isWin unit test, and an applyPatch round-trip regression test. Full suite passes with 100% coverage.

[ExplodingCabbage]

when a Unix source is patched with a patch whose final line's content ends in a literal carriage return and has no trailing newline

It's more specific than that, no? Roughly, the patch also needs to not feature any other lines, in order for isWin to return true - i.e. it needs to be a single-line patch. Still a legit bug, though!

[ExplodingCabbage]

LGTM, thank you! Will just see if Copilot offers any useful input before I merge.

[Copilot]

Pull request overview

Fixes applyPatch preserving literal carriage returns at EOF during line-ending conversion.

Changes:

• Guards Windows detection/conversion around no-newline markers.
• Adds unit and regression coverage.
• Documents the fix in release notes.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
src/patch/line-endings.ts	Preserves literal EOF carriage returns.
test/patch/line-endings.js	Tests conversion and detection behavior.
test/patch/apply.js	Adds an applyPatch regression test.
release-notes.md	Documents the bug fix.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.