- Changed +m to --multi to enable multi-select in CTRL-R
- Changed exclude to exclude-multi and {1} to {+1} so
shift-delete removes all selected entries at once
When --walker=follow is used, symlinks like Wine's z: -> / cause fzf to
traverse the entire root filesystem. fastwalk's built-in loop detection
only catches this on the second pass, but a single pass through / already
causes severe CPU and memory exhaustion.
This fix resolves each symlink-to-directory target to its absolute real
path and skips it if it is an ancestor of (or equal to) the walker root.
Close#4710
When --walker=follow is used, symlink following is now handled by
fastwalk's IgnoreDuplicateDirs adapter which tracks visited directories
by device+inode. This prevents the same directory from being entered
more than once, avoiding effectively infinite traversal when a symlink
points outside the walker root.
Close#4710
Separate item identity from cursor tracking:
- Add --id-nth=NTH to define item identity fields for cross-reload ops
- --track reverts to a simple boolean flag
- track-current action no longer accepts nth argument
- With --multi, selections are preserved across reload-sync by matching
identity keys in the reloaded list
Close#4718Close#4701Close#4483Close#4409Close#3460Close#2441
Allow --track to accept an optional nth expression for cross-reload
tracking. When a reload is triggered, fzf extracts a tracking key from
the current item using the nth expression, blocks the UI, and searches
for a matching item in the reloaded list.
- --track=.. tracks by entire line, --track=1 by first field, etc.
- --track without NTH retains existing index-based behavior
- UI is blocked during search (dimmed query, hidden cursor, +T*/+t*)
- reload unblocks eagerly on match; reload-sync waits for stream end
- Escape/Ctrl-C cancels blocked state without quitting
- track-current action accepts optional nth: track-current(1)
- Validate nth expression at parse time for both --track and track()
- Cache trackKeyFor results per item to avoid redundant computation
- Rename executeRegexp to argActionRegexp
Close#4701Close#3460
Fix#4709
Use go-shellwords instead of strings.Fields to parse --with-shell,
so paths with spaces can be properly quoted.
ln -s /bin/bash "/tmp/ba sh"
fzf --with-shell='/tmp/ba\ sh -c' --preview 'echo hello world'
fzf --with-shell='"/tmp/ba sh" -c' --preview 'echo hello world'
Replace the per-chunk query cache from []Result slices to fixed-size
bitmaps (ChunkBitmap: [16]uint64 = 128 bytes per entry). Each bit
indicates whether the corresponding item in the chunk matched.
This reduces cache memory by 86 times in testing:
- Old []Result cache: ~22KB per chunk per query (for 500 matches)
- New bitmap cache: ~262 bytes per chunk per query (fixed)
With the reduced per-entry cost, queryCacheMax is raised from
chunkSize/5 to chunkSize/2, allowing broader queries (up to 50% match
rate) to be cached while still using far less memory.
Fixes bugs reported in https://github.com/junegunn/fzf/pull/4703:
* Clamp followOffset return value to avoid going past the end of lines
* Account for t.previewed.filled when determining scrollability
* Fix nth attr merge order to respect precedence hierarchy
nth attrs were merged ON TOP of current-fg/selected-fg attrs, so
nth:regular would clear attrs like underline from current-fg. Fix the
merge chain to apply nth BEFORE the line-type overlay:
fg < nth < selected-fg < current-fg < hl < selected-hl < current-hl
Store raw current-fg and selected-fg attrs in ColorTheme before they
get merged with fg/ListFg, then pass them as nthOverlay through
printHighlighted to colorOffsets where the correct merge chain is
computed: fgAttr.Merge(nthAttr).Merge(nthOverlay).
Fix#4687
* Make current-fg inherit from list-fg instead of fg
current-fg was inheriting from fg, not list-fg, so setting list-fg
had no effect on the current line. e.g.
fzf --color fg:dim,list-fg:underline,nth:regular -d / --nth -1
The non-nth part of the current line was dim (from fg) instead of
underline (from list-fg). Resolve ListFg/ListBg earlier in InitTheme
so Current can inherit from them.
* Make selected-fg inherit from list-fg via merge instead of override
selected-fg used o() which replaces the attr from list-fg entirely.
e.g. with fg:dim,selected-fg:italic, the dim was lost on selected
lines because o() replaced dim with italic instead of merging them.
Use ColorAttr.Merge() so attrs are combined additively, consistent
with how current-fg inherits from list-fg.
* Apply selected-fg attrs on current line when item is selected
When an item is both current and selected, the current-line rendering
ignored selected-fg attrs entirely. e.g.
echo foo | fzf --color fg:dim,nth:regular,current-fg:underline,selected-fg:italic --bind result:select --multi
The italic from selected-fg was lost. Merge NthSelectedAttr into the
overlay and rebuild colBase attr with the correct precedence chain:
fg < selected-fg < current-fg.
Replace comparison-based pdqsort with LSD radix sort on the uint64
sort key. Radix sort is O(n) vs O(n log n) and avoids pointer-chasing
cache misses in the comparison function. Sort scratch buffer is reused
across iterations to reduce GC pressure.
Benchmark (single-threaded, Chromium file list):
- linux query (180K matches): ~16% faster
- src query (high match count): ~31% faster
- Rare matches: equivalent (falls back to pdqsort for n < 128)
For the common case of a single fuzzy term with no nth transform,
call the algo function directly from matchChunk, bypassing the
MatchItem -> extendedMatch -> iter dispatch chain. This eliminates
3 function calls and the per-match []Offset heap allocation.
Extend the uint64 rank comparison trick (comparing [4]uint16 as a
single uint64) to arm64 builds. ARM64 is little-endian like x86, so
the same unsafe.Pointer cast produces correct lexicographic ordering.
This replaces a 4-iteration loop with a single uint64 comparison,
speeding up the sort phase.
Chromium file list, single-threaded:
linux: 126ms -> 126ms (sort not dominant)
src: 462ms -> 438ms (-5%, sort-heavy)
By default, fzf uses 8 * NumCPU goroutines (capped at 32) for
parallel matching. --threads N overrides this to use exactly N
goroutines, which is useful for benchmarking and profiling.
fzf --filter PATTERN --bench 3s < input
Repeats matcher.scan() for the given duration, clears cache between
iterations, and prints stats (iterations, avg, min, max) to stderr.
All input lines now enter the chunklist with sequential indices, and
header lines are excluded from matching via Pattern.startIndex and
PassMerger offset. This allows the number of header lines to be changed
at runtime with change-header-lines(N), transform-header-lines, and
bg-transform-header-lines actions.
- Remove EvtHeader event; header items are read directly from chunks
- Add startIndex to Pattern and PassMerger for skipping header items
- Add targetIndex field to Terminal for cursor repositioning across
header-lines changes
Close#4659
In Phase 3 of FuzzyMatchV2, when a cell's left neighbor score is <= 1
and the current character doesn't match the pattern character, the
cell's score is guaranteed to be 0 (since gap penalties are -1 and -3).
Skip the bonus/gap computation entirely and fast-forward through
consecutive non-matching characters in the dead zone.
This yields 6-11% faster fuzzy searches on typical workloads.
With chunkSize=100 and 10M items, 100K chunks cause ~300K mutex
lock/unlock operations per search across 32 goroutines competing
for a single sync.Mutex in ChunkCache.
Increasing to 1000 reduces chunks to 10K, cutting contention overhead.
Benchmarks on 10M items show 14-80% faster searches depending on query
selectivity.
Support double, curly, dotted, and dashed underline styles via --color
(e.g. underline-curly) and ANSI passthrough (SGR 4:N, 58, 59) with --ansi.
Close#4633Close#4678
Thanks to @shtse8 for the test cases.
This commit fixes incorrect coloring for items that contain zero-width
characters. It also makes ellipsis coloring consistent when text is
trimmed from either the left or the right.
Fix#4620Close#4646
Fix#4652
Code was reorganized so that it doesn't delete the current line if the
cursor is not at the front. This should be acceptable in almost all
cases.
Thanks to @phanen for reporting the issue and suggesting the fix.
Junegunn Choi