ashlar-diff

Headless text diffs for ashlar apps: which lines changed, which words in them, and how to show them with context.

• FileDiff::from_texts(old, new): a line diff (Myers' algorithm) cut into hunks with three lines of context. A cost cap bounds the time and memory of one run; past it the diff splits at lines unique to both sides, and as a last resort replaces the rest wholesale (coarse is then set). from_texts_with takes the context size, the cap, and whether to compute word emphasis.
• Word-level emphasis: in a run of removed lines followed by added lines, the n-th removed line pairs with the n-th added one, and Line::emphasis holds the byte ranges that differ. word_diff(old, new) exposes the same for any two strings.
• FileDiff::gaps(): the unchanged ranges between hunks, which a view shows collapsed. They are expandable (source_line) when the new text is known: always for from_texts, and for a parsed patch after with_source(new_text).
• parse_patch(text): unified patches and git diff output, one FileDiff per file, with renames, copies, mode changes, binary markers (and GIT binary patch data), quoted paths and \ No newline at end of file. Malformed input is a PatchError with a line number, never a panic; nothing is allocated from a number the patch states. write_patch writes git diff output back byte for byte.
• diff_ops(a, b, max_cost): the underlying edit script for any Eq + Hash sequence.

No dependencies, no UI; builds for wasm32-unknown-unknown. The view is ashlar_ui::diff_view::DiffView.

use ashlar_diff::{FileDiff, LineKind, parse_patch, write_patch};

let d = FileDiff::from_texts("a\nb\n", "a\nc\n").with_paths("f.txt", "f.txt");
assert_eq!(d.stats(), (1, 1));
let files = parse_patch(&d.to_patch()).unwrap();
assert_eq!(files[0].hunks[0].lines[1].kind, LineKind::Removed);
assert_eq!(write_patch(&files), d.to_patch());