Sparkles

Appearance

Build a small fuzzy file search ​

Add sparkles:fuzzy as a DUB dependency, then parse the prompt once and reuse one matcher workspace for every candidate:

d
import sparkles.fuzzy;

auto workspace = new MatcherWorkspace!();
auto parsed = parseQuery(`app ext:d`, workspace.textWorkspace);
assert(parsed.hasValue);

CandidateView candidate;
candidate.id.low = 1;
candidate.path = "src/app.d";
candidate.filenameOffset = 4;

auto result = match(parsed.value, candidate, *workspace);
assert(result.hasValue && result.value.admitted);
assert(result.value.score > 0);

TextRange[16] highlights;
auto count = positions(parsed.value, candidate, MatchConfig.init,
    FuzzyLimits.init, *workspace, highlights);
assert(count.hasValue);

QueryStorage borrows the prompt, and CandidateView borrows the path. Keep both byte buffers alive and unchanged for the whole operation. Each worker needs its own MatcherWorkspace; workspaces are deliberately not synchronized. Allocate the workspace once, before interactive parsing. textWorkspace exposes its reusable Unicode scratch: parsing finishes before matching overwrites that scratch, while QueryStorage retains only its own bounded data and prompt borrow. The default matcher is about 13.1 MiB on x86-64; do not put it on a worker stack.

The default codePath profile is NFC with smart case. Use QueryParseOptions.profile = AnalysisProfile.generalLanguage() for NFKC, full folding, accent removal, and optional caller-owned stopwords.