@netgrep/search

The WASM porting of ripgrep. See the main README for more information.

This is the low-level core of @netgrep/netgrep, which is probably the package you want — it adds the streaming that makes the engine useful over HTTP. Used directly, this one exposes two functions, and its default export is an init you must await before calling either of them:

search_bytes(chunk: Uint8Array, pattern: string): boolean
search_block(chunk: Uint8Array, pattern: string, maxLineBytes: number): BlockHits

search_bytes answers membership and allocates nothing. search_block returns every matching line in the chunk, as two flat values rather than an array of objects — text, the matching lines joined by \n, and table, a Uint32Array holding [hitCount, linesInBlock] and then [lineNumber, nRanges, start, end, …] per hit. Two values cross the WASM boundary per call however many lines matched, which is what keeps memory flat on a file with hundreds of thousands of hits; the caller walks the table and builds each result at the moment it needs it.

Each line is terminator stripped, truncated to maxLineBytes on a UTF-8 character boundary and decoded lossily. Its start/end offsets are UTF-16 code units of that returned line, so line.slice(start, end) works without conversion. nRanges counts pairs, and can be zero: a line can match past the truncation cut, in which case the hit is still reported with nothing to highlight. Count hits rather than testing text for truthiness — a pattern matching an empty line is a real hit whose line is "", and a zero-hit block also has empty text.

Both throw if the pattern is not valid regex, so a stray ( from a search box is an ordinary catchable error. Both share one compiled matcher and one searcher configuration, so their matching semantics cannot differ: anything search_bytes calls a match, search_block returns the line for.

On its own this package is a ~1.17 MB WebAssembly binary (~500 KB gzipped) and nothing else. Its known limitations are worth reading first — netgrep does not detect binary files, results inside a line longer than 64 KB are approximate, and on input using bare CR line endings ^/$ anchor to a boundary the returned line does not.