Blaze

Early but usable. Expect breaking changes while the query langauge and CLI settles.

A Blazingly fast file searcher.

blaze is a fast, index-based file searcher for large codebases and home directories. It precomputes a compact on-disk index and answers queries in milliseconds, even on trees with millions of paths.

The goal is to provide a syntax-rich searcher that is competitive with plocate. blaze is designed for free-text-like queries. It is not a sophisticated regex engine and it is not a flag-driven find replacement. The query syntax is intentionally small and human-friendly.

Features

• Very fast substring search
• Not a recursive find
• Relevance ranking
• Rich query language
• Low memory overhead

Usage

blaze is designed for free-text-like queries. It is not a sophisticated regex engine and it is not a flag-driven find replacement. The syntax is intentionally small and human-friendly.

blaze query '<query>'

Common searches

Just type words:

blaze query 'invoice pdf'
blaze query 'src config'

Exact phrase (use quotes when there are spaces):

blaze query '"design doc"'
blaze query '"tax return" 2023'

Either / exclude:

blaze query 'resume or cv'
blaze query 'config not backup'
blaze query '(invoice or receipt) not draft'

Simple wildcards:

blaze query 'Cargo*.toml'
blaze query 'report-202?-final'

Filters

By extension:

blaze query 'ext:rs'
blaze query 'ext:.jpg vacation'

By time (examples):

blaze query 'modified:today'
blaze query 'modified:-7d'
blaze query 'created:2024-01-01'
blaze query 'modified:this_week ext:md'

By size:

blaze query 'size:>10MB'
blaze query 'size:<500K ext:log'

Bits and Bytes Smart casing

size: defaults to bytes. If you specifically want bits, use an uppercase unit with a lowercase b (like Mb).

# bytes (default)
blaze query 'size:>10MB'
blaze query 'size:>10mb'

# bits (uppercase + lowercase b)
blaze query 'size:>10Mb'   # megabits

Performance

Benchmarks were run with hyperfine on:

• Operating System: PopOS
• Dataset: $HOME
• blaze: release build with an index prebuilt over $HOME (daemon and CLI)

Each command was run 3 warmup + 10 measured iterations with warm caches.

blaze vs fdfind vs find vs plocate

Representative results (mean time, lower is better):

Query	blaze (daemon)	blaze (CLI)	fdfind	plocate	find
ext:rs	1.4 ms	2.0 ms	11.1 ms	18.2 ms	236.9 ms
Cargo.toml	0.7 ms	0.9 ms	9.1 ms	2.2 ms	223.6 ms
config	1.7 ms	2.1 ms	10.1 ms	48.5 ms	272.6 ms
src	4.8 ms	5.0 ms	9.7 ms	141.9 ms	252.6 ms
modified:today	0.7 ms	1.2 ms	10.3 ms	1125 ms	359.7 ms

Highlights:

• Daemon mode is 2-15x faster than fdfind and 50-540x faster than find, depending on query selectivity.
• Even against plocate, blaze holds a 3-30x advantage on filename/path queries (and >1000x on date filters) while providing a richer query language.
• Cold-start CLI mode is still 2-10x faster than fdfind and 50-300x faster than find.

License

MIT