Search syntax

In extended-search mode (the default), the user's query is split on whitespace into terms, and each term is interpreted as a small operator. Implemented in src/pattern.go. The README's "Search syntax" section is the user-facing reference; this page focuses on the code.

Operators

Token	termType (src/pattern.go)	Algorithm	Description
foo	termFuzzy	algo.FuzzyMatchV2 (or V1 with --algo=v1)	Fuzzy match.
'foo	termExact	algo.ExactMatchNaive	Exact substring match.
^foo	termPrefix	algo.PrefixMatch	Match must start at position 0.
foo$	termSuffix	algo.SuffixMatch	Match must end at the last position.
'foo$	termExactBoundary	algo.ExactMatchBoundary	Exact match at a word boundary.
=foo	termEqual	algo.EqualMatch	Whole-string equality.
!term	inverted	(any)	Negation of any of the above.
|	OR separator	—	Splits a termSet into multiple OR groups.

The grammar is documented at the top of src/pattern.go:

// fuzzy
// 'exact
// ^prefix-exact
// suffix-exact$
// !inverse-exact
// !'inverse-fuzzy
// !^inverse-prefix-exact
// !inverse-suffix-exact$

Compilation

BuildPattern in src/pattern.go runs the query through these steps:

1. Trim whitespace, taking care not to drop trailing escaped spaces (\ ).
2. Look up the cache by string — re-typing the same query is free.
3. Split the trimmed query into tokens (_splitRegex matches one or more spaces).
4. For each token:
   • Strip the leading !, set inv.
   • Strip the leading ', ^, = and decide the termType.
   • Strip the trailing $.
   • If the result is empty, skip.
   • Choose the corresponding algo.Algo from Pattern.procFun and wrap it as a term.
5. Group consecutive terms into termSets, splitting on |.

The result is a Pattern whose MatchItem(item) returns the best (Result, position) tuple across the OR groups, with all AND terms within a group required to match.

Smart case

By default fzf uses smart-case: case-sensitive when the query contains an uppercase letter, otherwise case-insensitive. --ignore-case and --no-ignore-case force the behavior. The decision is made per term in BuildPattern.

Normalization

The default behavior normalizes Latin-script accented characters (so typing ascii matches ãśčíí). --literal disables this. Implemented in src/algo/normalize.go. Disabled per term as well as globally.

Schemes (--scheme)

--scheme selects a preset:

• default — general-purpose tuning.
• path — adds bonus for matches after /; first character of a path component is treated like a word start.
• history — disables sort and tweaks scoring so that recent (input order) wins ties.

The implementation is a few branches in BuildPattern and a couple of constants in src/algo/algo.go.

Disabled mode

--disabled short-circuits the pattern to a no-op term that matches everything. Useful when fzf is being driven entirely by transform-search actions over --listen.

Where it lives

Concern	File
Query parsing & term grouping	src/pattern.go
Algorithm implementations	src/algo/algo.go
Normalization tables	src/algo/normalize.go
Cache key	Pattern.cacheKey in src/pattern.go
Tests	src/pattern_test.go, src/algo/algo_test.go, test/test_filter.rb

See systems/matching for the broader picture.