Fix panic on short "diff --git" args

[samarth70]

Problem

parseDiffGitArgs panics on a diff --git line whose unquoted, space-containing arguments are shorter than two bytes. It is reachable from the public ParseFileDiff / ParseMultiFileDiff on malformed input:

_, err := diff.ParseFileDiff([]byte("diff --git x  \n"))
// panic: runtime error: slice bounds out of range [:2] with length 1

In the ambiguous "unquoted filenames containing spaces" branch, the two candidate names are compared against the a/ and b/ prefixes:

// If the name minus the a/ b/ prefixes is equal, proceed.
if len(first) >= 3 && first[1] == '/' && first[1:] == second[1:] {
	return first, second, true
}
// If the names don't have the a/ and b/ prefixes and they're equal, proceed.
if !(first[:2] == "a/" && second[:2] == "b/") && first == second {
	return first, second, true
}

The first check guards len(first) >= 3 before indexing, but the second slices first[:2] and second[:2] with no length guard. When the args are short enough that first is a single byte (e.g. x ), first[:2] is out of range and the parser panics instead of returning cleanly.

Since go-diff is commonly used to parse diffs from untrusted or arbitrary sources, a malformed header should be handled, not crash the caller.

Fix

Guard the prefix comparison with len(first) >= 2, mirroring the len(first) >= 3 guard on the sibling check. On this branch first and second always have equal length, so guarding first covers second too. Short, unextractable args now return the existing ("", "", true) result (valid syntax, no filename extracted) rather than panicking.

Testing

TestParseDiffGitArgs_ShortAmbiguous covers x , a , and ab . Each panics on current master and, with the fix, returns ("", "", true). The existing TestParseDiffGitArgs_Success / _Unsuccessful tables still pass, and gofmt/go vet are clean.

Note: a few pre-existing TestParseMultiFileDiffHeaders testdata cases fail in my local Windows environment on an unmodified checkout as well, so they are unrelated to this change; worth confirming on CI.

[keegancsmith]

Thanks for the fix! I pushed a slight cleanup using strings.HasPrefix instead and added a testcase using a public API which would of paniced before.