dcdpr / jp / 27841623249

69%

push

github

chore(tools): Fix non-UTF-8 bytes silently swallowing diff output (#772)

`DuctProcessRunner` converted captured stdout/stderr with
`String::from_utf8(...).unwrap_or_default()`. When a staged diff
contained even a single non-UTF-8 byte — a latin-1 source file, a
near-binary blob git still treats as text — the entire capture was
discarded and replaced with an empty string. An empty stdout with a zero
exit code fell through to the `"No changes."` branch in `git_diff_impl`,
making a non-empty staged diff completely invisible.

Switching to `String::from_utf8_lossy(...).into_owned()` preserves the
output and replaces only the invalid bytes with the Unicode replacement
character, which is the correct behaviour for a tool that needs to
display diff content rather than parse it precisely.

A regression test stages a file containing byte `0xe9` (no NUL, so git
treats it as text and emits the raw byte in the diff) and asserts the
result is neither `"No changes."` nor missing the filename.

Signed-off-by: rgrant <rgrant@contract.design>

2 of 2 new or added lines in 1 file covered. (100.0%)

35690 of 53570 relevant lines covered (66.62%)

406.39 hits per line