Use LSD radix sort for Result sorting in matcher

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)
Add direct algo fast path in matchChunk
2026-03-13 02:10:51 +08:00 · 2026-03-01 11:58:16 +09:00 · 2026-03-01 11:58:16 +09:00 · 2026-03-01 11:18:43 +09:00
44 changed files with 429 additions and 2098 deletions
--- a/.github/workflows/labeler.yml
+++ b/.github/workflows/labeler.yml
@@ -12,6 +12,6 @@ jobs:
  label:
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/labeler@v6
+      - uses: actions/labeler@v5
        with:
          configuration-path: .github/labeler.yml
--- a/.github/workflows/linux.yml
+++ b/.github/workflows/linux.yml
@@ -44,10 +44,5 @@ jobs:
    - name: Unit test
      run: make test

-    - name: Fuzz test
-      run: |
-        go test ./src/algo/ -fuzz=FuzzIndexByteTwo -fuzztime=5s
-        go test ./src/algo/ -fuzz=FuzzLastIndexByteTwo -fuzztime=5s
-
    - name: Integration test
      run: make install && ./install --all && tmux new-session -d && ruby test/runner.rb --verbose
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,62 +1,6 @@
 CHANGELOG
 =========

-0.70.1
------
- Performance improvements
-    - The search performance now scales linearly with the number of CPU cores, as we dropped static partitioning to allow better load balancing across threads.
-      ```
-      === query: 'linux' ===
-        [all]   baseline:    17.12ms  current:    14.28ms  (1.20x)  matches: 179966 (12.79%)
-        [1T]    baseline:   136.49ms  current:   137.25ms  (0.99x)  matches: 179966 (12.79%)
-        [2T]    baseline:    75.74ms  current:    68.75ms  (1.10x)  matches: 179966 (12.79%)
-        [4T]    baseline:    41.16ms  current:    34.97ms  (1.18x)  matches: 179966 (12.79%)
-        [8T]    baseline:    32.82ms  current:    17.79ms  (1.84x)  matches: 179966 (12.79%)
-      ```
-    - Improved the cache structure, reducing memory footprint per entry by 86x.
-        - With the reduced per-entry cost, the cache now has broader coverage.
- fish: Improved command history (CTRL-R) (#44703) (@bitraid)
- Bug fixes
-    - Fixed AWK tokenizer not treating a new line character as whitespace
-    - Fixed `--{accept,with}-nth` removing trailing whitespaces with a non-default `--delimiter`
-    - Fixed OSC8 hyperlinks being mangled when the URL contains unicode characters (#4707)
-    - Fixed `--with-shell` not handling quoted arguments correctly (#4709)
-
-0.70.0
------
- Added `change-with-nth` action for dynamically changing the `--with-nth` option.
-    - Requires `--with-nth` to be set initially.
-    - Multiple options separated by `|` can be given to cycle through.
-  ```sh
-  echo -e "a b c\nd e f\ng h i" | fzf --with-nth .. \
-    --bind 'space:change-with-nth(1|2|3|1,3|2,3|)'
-  ```
- Added `change-header-lines` action for dynamically changing the `--header-lines` option
- Performance improvements (1.3x to 1.9x faster filtering depending on query)
-  ```
-  === query: 'l' ===
-    [all]   baseline:   168.87ms  current:    95.21ms  (1.77x)  matches: 5069891 (94.78%)
-    [1T]    baseline:  1652.22ms  current:   841.40ms  (1.96x)  matches: 5069891 (94.78%)
-
-  === query: 'lin' ===
-    [all]   baseline:   343.27ms  current:   252.59ms  (1.36x)  matches: 3516507 (65.74%)
-    [1T]    baseline:  3199.89ms  current:  2230.64ms  (1.43x)  matches: 3516507 (65.74%)
-
-  === query: 'linux' ===
-    [all]   baseline:    85.47ms  current:    63.72ms  (1.34x)  matches: 307229 (5.74%)
-    [1T]    baseline:   774.64ms  current:   589.32ms  (1.31x)  matches: 307229 (5.74%)
-
-  === query: 'linuxlinux' ===
-    [all]   baseline:    55.13ms  current:    35.67ms  (1.55x)  matches: 12230 (0.23%)
-    [1T]    baseline:   461.99ms  current:   332.38ms  (1.39x)  matches: 12230 (0.23%)
-
-  === query: 'linuxlinuxlinux' ===
-    [all]   baseline:    51.77ms  current:    32.53ms  (1.59x)  matches: 865 (0.02%)
-    [1T]    baseline:   409.99ms  current:   296.33ms  (1.38x)  matches: 865 (0.02%)
-  ```
- Fixed `nth` attribute merge order to respect precedence hierarchy (#4697)
- bash: Replaced `printf` with builtin `printf` to bypass local indirections (#4684) (@DarrenBishop)
-
 0.68.0
 ------
 - Implemented word wrapping in the list section
--- a/README.md
+++ b/README.md
--- a/2
+++ b/2
@@ -2,7 +2,7 @@

 set -u

-version=0.70.0
+version=0.68.0
 auto_completion=
 key_bindings=
 update_config=2
--- a/install.ps1
+++ b/install.ps1
@@ -1,4 +1,4 @@
-$version="0.70.0"
+$version="0.68.0"

 $fzf_base=Split-Path -Parent $MyInvocation.MyCommand.Definition

--- a/main.go
+++ b/main.go
@@ -11,7 +11,7 @@ import (
 	"github.com/junegunn/fzf/src/protector"
 )

-var version = "0.70"
+var version = "0.68"
 var revision = "devel"

 //go:embed shell/key-bindings.bash
--- a/man/man1/fzf-tmux.1
+++ b/man/man1/fzf-tmux.1
@@ -21,7 +21,7 @@ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 THE SOFTWARE.
 ..
-.TH fzf\-tmux 1 "Mar 2026" "fzf 0.70.0" "fzf\-tmux - open fzf in tmux split pane"
+.TH fzf\-tmux 1 "Feb 2026" "fzf 0.68.0" "fzf\-tmux - open fzf in tmux split pane"

 .SH NAME
 fzf\-tmux - open fzf in tmux split pane
--- a/man/man1/fzf.1
+++ b/man/man1/fzf.1
@@ -21,7 +21,7 @@ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 THE SOFTWARE.
 ..
-.TH fzf 1 "Mar 2026" "fzf 0.70.0" "fzf - a command-line fuzzy finder"
+.TH fzf 1 "Feb 2026" "fzf 0.68.0" "fzf - a command-line fuzzy finder"

 .SH NAME
 fzf - a command-line fuzzy finder
@@ -134,14 +134,6 @@ e.g.
     # Use template to rearrange fields
     echo foo,bar,baz | fzf --delimiter , --with-nth '{n},{1},{3},{2},{1..2}'
 .RE
-.RS
-
-\fBchange\-with\-nth\fR action is only available when \fB\-\-with\-nth\fR is set.
-When \fB\-\-with\-nth\fR is used, fzf retains the original input lines in memory
-so they can be re\-transformed on the fly (e.g. \fB\-\-with\-nth ..\fR to keep
-the original presentation). This increases memory usage, so only use
-\fB\-\-with\-nth\fR when you actually need field transformation.
-.RE
 .TP
 .BI "\-\-accept\-nth=" "N[,..] or TEMPLATE"
 Define which fields to print on accept. The last delimiter is stripped from the
@@ -1237,18 +1229,6 @@ Here is an example script that uses a Unix socket instead of a TCP port.
    curl --unix-socket /tmp/fzf.sock http -d up
    \fR

-.TP
-.BI "\-\-threads=" "N"
-Number of matcher threads to use. The default value is
-\fBmin(8 * NUM_CPU, 32)\fR.
-.TP
-.BI "\-\-bench=" "DURATION"
-Repeatedly run \fB\-\-filter\fR for the given duration and print timing
-statistics. Must be used with \fB\-\-filter\fR.
-
-e.g.
-     \fBcat /usr/share/dict/words | fzf \-\-filter abc \-\-bench 10s\fR
-
 .SS DIRECTORY TRAVERSAL
 .TP
 .B "\-\-walker=[file][,dir][,follow][,hidden]"
@@ -1418,8 +1398,6 @@ fzf exports the following environment variables to its child processes.
 .br
 .BR FZF_NTH "             Current \-\-nth option"
 .br
-.BR FZF_WITH_NTH "        Current \-\-with\-nth option"
-.br
 .BR FZF_PROMPT "          Prompt string"
 .br
 .BR FZF_GHOST "           Ghost string"
@@ -1910,7 +1888,6 @@ A key or an event can be bound to one or more of the following actions.
    \fBchange\-multi\fR                 (enable multi-select mode with no limit)
    \fBchange\-multi(...)\fR            (enable multi-select mode with a limit or disable it with 0)
    \fBchange\-nth(...)\fR              (change \fB\-\-nth\fR option; rotate through the multiple options separated by '|')
-    \fBchange\-with\-nth(...)\fR         (change \fB\-\-with\-nth\fR option; rotate through the multiple options separated by '|')
    \fBchange\-pointer(...)\fR          (change \fB\-\-pointer\fR option)
    \fBchange\-preview(...)\fR          (change \fB\-\-preview\fR option)
    \fBchange\-preview\-label(...)\fR    (change \fB\-\-preview\-label\fR to the given string)
@@ -2016,7 +1993,6 @@ A key or an event can be bound to one or more of the following actions.
    \fBtransform\-input\-label(...)\fR   (transform input label using an external command)
    \fBtransform\-list\-label(...)\fR    (transform list label using an external command)
    \fBtransform\-nth(...)\fR           (transform nth using an external command)
-    \fBtransform\-with\-nth(...)\fR      (transform with-nth using an external command)
    \fBtransform\-pointer(...)\fR       (transform pointer using an external command)
    \fBtransform\-preview\-label(...)\fR (transform preview label using an external command)
    \fBtransform\-prompt(...)\fR        (transform prompt string using an external command)
--- a/shell/common.sh
+++ b/shell/common.sh
@@ -1,9 +1,9 @@
 __fzf_defaults() {
  # $1: Prepend to FZF_DEFAULT_OPTS_FILE and FZF_DEFAULT_OPTS
  # $2: Append to FZF_DEFAULT_OPTS_FILE and FZF_DEFAULT_OPTS
-  builtin printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
+  printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
  command cat "${FZF_DEFAULT_OPTS_FILE-}" 2> /dev/null
-  builtin printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
+  printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
 }

 __fzf_exec_awk() {
--- a/shell/completion.bash
+++ b/shell/completion.bash
@@ -38,9 +38,9 @@ if [[ $- =~ i ]]; then
 # the changes. See code comments in "common.sh" for the implementation details.

 __fzf_defaults() {
-  builtin printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
+  printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
  command cat "${FZF_DEFAULT_OPTS_FILE-}" 2> /dev/null
-  builtin printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
+  printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
 }

 __fzf_exec_awk() {
@@ -81,7 +81,7 @@ __fzf_orig_completion() {
      f="${BASH_REMATCH[2]}"
      cmd="${BASH_REMATCH[3]}"
      [[ $f == _fzf_* ]] && continue
-      builtin printf -v "_fzf_orig_completion_${cmd//[^A-Za-z0-9_]/_}" "%s" "${comp} %s ${cmd} #${f}"
+      printf -v "_fzf_orig_completion_${cmd//[^A-Za-z0-9_]/_}" "%s" "${comp} %s ${cmd} #${f}"
      if [[ $l == *" -o nospace "* ]] && [[ ${__fzf_nospace_commands-} != *" $cmd "* ]]; then
        __fzf_nospace_commands="${__fzf_nospace_commands-} $cmd "
      fi
@@ -111,7 +111,7 @@ __fzf_orig_completion_instantiate() {
  orig="${!orig_var-}"
  orig="${orig%#*}"
  [[ $orig == *' %s '* ]] || return 1
-  builtin printf -v REPLY "$orig" "$func"
+  printf -v REPLY "$orig" "$func"
 }

 _fzf_opts_completion() {
@@ -376,7 +376,7 @@ __fzf_generic_path_completion() {
            eval "rest=(${FZF_COMPLETION_PATH_OPTS-})"
          fi
          if declare -F "$1" > /dev/null; then
-            eval "$1 $(builtin printf %q "$dir")" | __fzf_comprun "$4" -q "$leftover" "${rest[@]}"
+            eval "$1 $(printf %q "$dir")" | __fzf_comprun "$4" -q "$leftover" "${rest[@]}"
          else
            if [[ $1 =~ dir ]]; then
              walker=dir,follow
@@ -385,7 +385,7 @@ __fzf_generic_path_completion() {
            fi
            __fzf_comprun "$4" -q "$leftover" --walker "$walker" --walker-root="$dir" "${rest[@]}"
          fi | while read -r item; do
-            builtin printf "%q " "${item%$3}$3"
+            printf "%q " "${item%$3}$3"
          done
        )
        matches=${matches% }
@@ -395,9 +395,9 @@ __fzf_generic_path_completion() {
        else
          COMPREPLY=("$cur")
        fi
-        # To redraw line after fzf closes (builtin printf '\e[5n')
+        # To redraw line after fzf closes (printf '\e[5n')
        bind '"\e[0n": redraw-current-line' 2> /dev/null
-        builtin printf '\e[5n'
+        printf '\e[5n'
        return 0
      fi
      dir=$(command dirname "$dir")
@@ -455,7 +455,7 @@ _fzf_complete() {
      COMPREPLY=("$cur")
    fi
    bind '"\e[0n": redraw-current-line' 2> /dev/null
-    builtin printf '\e[5n'
+    printf '\e[5n'
    return 0
  else
    _fzf_handle_dynamic_completion "$cmd" "${rest[@]}"
@@ -527,7 +527,7 @@ _fzf_proc_completion_post() {
 #     # Set the local attribute for any non-local variable that is set by _known_hosts_real()
 #     local COMPREPLY=()
 #     _known_hosts_real ''
-#     builtin printf '%s\n' "${COMPREPLY[@]}" | command sort -u --version-sort
+#     printf '%s\n' "${COMPREPLY[@]}" | command sort -u --version-sort
 #   }
 if ! declare -F __fzf_list_hosts > /dev/null; then
  __fzf_list_hosts() {
--- a/shell/completion.zsh
+++ b/shell/completion.zsh
@@ -102,9 +102,9 @@ if [[ -o interactive ]]; then
 # the changes. See code comments in "common.sh" for the implementation details.

 __fzf_defaults() {
-  builtin printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
+  printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
  command cat "${FZF_DEFAULT_OPTS_FILE-}" 2> /dev/null
-  builtin printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
+  printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
 }

 __fzf_exec_awk() {
--- a/shell/key-bindings.bash
+++ b/shell/key-bindings.bash
@@ -25,9 +25,9 @@ if [[ $- =~ i ]]; then
 # the changes. See code comments in "common.sh" for the implementation details.

 __fzf_defaults() {
-  builtin printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
+  printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
  command cat "${FZF_DEFAULT_OPTS_FILE-}" 2> /dev/null
-  builtin printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
+  printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
 }

 __fzf_exec_awk() {
--- a/shell/key-bindings.zsh
+++ b/shell/key-bindings.zsh
@@ -45,9 +45,9 @@ if [[ -o interactive ]]; then
 # the changes. See code comments in "common.sh" for the implementation details.

 __fzf_defaults() {
-  builtin printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
+  printf '%s\n' "--height ${FZF_TMUX_HEIGHT:-40%} --min-height 20+ --bind=ctrl-z:ignore $1"
  command cat "${FZF_DEFAULT_OPTS_FILE-}" 2> /dev/null
-  builtin printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
+  printf '%s\n' "${FZF_DEFAULT_OPTS-} $2"
 }

 __fzf_exec_awk() {
--- a/src/actiontype_string.go
+++ b/src/actiontype_string.go
@@ -38,159 +38,156 @@ func _() {
 	_ = x[actChangeListLabel-27]
 	_ = x[actChangeMulti-28]
 	_ = x[actChangeNth-29]
-	_ = x[actChangeWithNth-30]
-	_ = x[actChangePointer-31]
-	_ = x[actChangePreview-32]
-	_ = x[actChangePreviewLabel-33]
-	_ = x[actChangePreviewWindow-34]
-	_ = x[actChangePrompt-35]
-	_ = x[actChangeQuery-36]
-	_ = x[actClearScreen-37]
-	_ = x[actClearQuery-38]
-	_ = x[actClearSelection-39]
-	_ = x[actClose-40]
-	_ = x[actDeleteChar-41]
-	_ = x[actDeleteCharEof-42]
-	_ = x[actEndOfLine-43]
-	_ = x[actFatal-44]
-	_ = x[actForwardChar-45]
-	_ = x[actForwardWord-46]
-	_ = x[actForwardSubWord-47]
-	_ = x[actKillLine-48]
-	_ = x[actKillWord-49]
-	_ = x[actKillSubWord-50]
-	_ = x[actUnixLineDiscard-51]
-	_ = x[actUnixWordRubout-52]
-	_ = x[actYank-53]
-	_ = x[actBackwardKillWord-54]
-	_ = x[actBackwardKillSubWord-55]
-	_ = x[actSelectAll-56]
-	_ = x[actDeselectAll-57]
-	_ = x[actToggle-58]
-	_ = x[actToggleSearch-59]
-	_ = x[actToggleAll-60]
-	_ = x[actToggleDown-61]
-	_ = x[actToggleUp-62]
-	_ = x[actToggleIn-63]
-	_ = x[actToggleOut-64]
-	_ = x[actToggleTrack-65]
-	_ = x[actToggleTrackCurrent-66]
-	_ = x[actToggleHeader-67]
-	_ = x[actToggleWrap-68]
-	_ = x[actToggleWrapWord-69]
-	_ = x[actToggleMultiLine-70]
-	_ = x[actToggleHscroll-71]
-	_ = x[actToggleRaw-72]
-	_ = x[actEnableRaw-73]
-	_ = x[actDisableRaw-74]
-	_ = x[actTrackCurrent-75]
-	_ = x[actToggleInput-76]
-	_ = x[actHideInput-77]
-	_ = x[actShowInput-78]
-	_ = x[actUntrackCurrent-79]
-	_ = x[actDown-80]
-	_ = x[actDownMatch-81]
-	_ = x[actUp-82]
-	_ = x[actUpMatch-83]
-	_ = x[actPageUp-84]
-	_ = x[actPageDown-85]
-	_ = x[actPosition-86]
-	_ = x[actHalfPageUp-87]
-	_ = x[actHalfPageDown-88]
-	_ = x[actOffsetUp-89]
-	_ = x[actOffsetDown-90]
-	_ = x[actOffsetMiddle-91]
-	_ = x[actJump-92]
-	_ = x[actJumpAccept-93]
-	_ = x[actPrintQuery-94]
-	_ = x[actRefreshPreview-95]
-	_ = x[actReplaceQuery-96]
-	_ = x[actToggleSort-97]
-	_ = x[actShowPreview-98]
-	_ = x[actHidePreview-99]
-	_ = x[actTogglePreview-100]
-	_ = x[actTogglePreviewWrap-101]
-	_ = x[actTogglePreviewWrapWord-102]
-	_ = x[actTransform-103]
-	_ = x[actTransformBorderLabel-104]
-	_ = x[actTransformGhost-105]
-	_ = x[actTransformHeader-106]
-	_ = x[actTransformHeaderLines-107]
-	_ = x[actTransformFooter-108]
-	_ = x[actTransformHeaderLabel-109]
-	_ = x[actTransformFooterLabel-110]
-	_ = x[actTransformInputLabel-111]
-	_ = x[actTransformListLabel-112]
-	_ = x[actTransformNth-113]
-	_ = x[actTransformWithNth-114]
-	_ = x[actTransformPointer-115]
-	_ = x[actTransformPreviewLabel-116]
-	_ = x[actTransformPrompt-117]
-	_ = x[actTransformQuery-118]
-	_ = x[actTransformSearch-119]
-	_ = x[actTrigger-120]
-	_ = x[actBgTransform-121]
-	_ = x[actBgTransformBorderLabel-122]
-	_ = x[actBgTransformGhost-123]
-	_ = x[actBgTransformHeader-124]
-	_ = x[actBgTransformHeaderLines-125]
-	_ = x[actBgTransformFooter-126]
-	_ = x[actBgTransformHeaderLabel-127]
-	_ = x[actBgTransformFooterLabel-128]
-	_ = x[actBgTransformInputLabel-129]
-	_ = x[actBgTransformListLabel-130]
-	_ = x[actBgTransformNth-131]
-	_ = x[actBgTransformWithNth-132]
-	_ = x[actBgTransformPointer-133]
-	_ = x[actBgTransformPreviewLabel-134]
-	_ = x[actBgTransformPrompt-135]
-	_ = x[actBgTransformQuery-136]
-	_ = x[actBgTransformSearch-137]
-	_ = x[actBgCancel-138]
-	_ = x[actSearch-139]
-	_ = x[actPreview-140]
-	_ = x[actPreviewTop-141]
-	_ = x[actPreviewBottom-142]
-	_ = x[actPreviewUp-143]
-	_ = x[actPreviewDown-144]
-	_ = x[actPreviewPageUp-145]
-	_ = x[actPreviewPageDown-146]
-	_ = x[actPreviewHalfPageUp-147]
-	_ = x[actPreviewHalfPageDown-148]
-	_ = x[actPrevHistory-149]
-	_ = x[actPrevSelected-150]
-	_ = x[actPrint-151]
-	_ = x[actPut-152]
-	_ = x[actNextHistory-153]
-	_ = x[actNextSelected-154]
-	_ = x[actExecute-155]
-	_ = x[actExecuteSilent-156]
-	_ = x[actExecuteMulti-157]
-	_ = x[actSigStop-158]
-	_ = x[actBest-159]
-	_ = x[actFirst-160]
-	_ = x[actLast-161]
-	_ = x[actReload-162]
-	_ = x[actReloadSync-163]
-	_ = x[actDisableSearch-164]
-	_ = x[actEnableSearch-165]
-	_ = x[actSelect-166]
-	_ = x[actDeselect-167]
-	_ = x[actUnbind-168]
-	_ = x[actRebind-169]
-	_ = x[actToggleBind-170]
-	_ = x[actBecome-171]
-	_ = x[actShowHeader-172]
-	_ = x[actHideHeader-173]
-	_ = x[actBell-174]
-	_ = x[actExclude-175]
-	_ = x[actExcludeMulti-176]
-	_ = x[actAsync-177]
+	_ = x[actChangePointer-30]
+	_ = x[actChangePreview-31]
+	_ = x[actChangePreviewLabel-32]
+	_ = x[actChangePreviewWindow-33]
+	_ = x[actChangePrompt-34]
+	_ = x[actChangeQuery-35]
+	_ = x[actClearScreen-36]
+	_ = x[actClearQuery-37]
+	_ = x[actClearSelection-38]
+	_ = x[actClose-39]
+	_ = x[actDeleteChar-40]
+	_ = x[actDeleteCharEof-41]
+	_ = x[actEndOfLine-42]
+	_ = x[actFatal-43]
+	_ = x[actForwardChar-44]
+	_ = x[actForwardWord-45]
+	_ = x[actForwardSubWord-46]
+	_ = x[actKillLine-47]
+	_ = x[actKillWord-48]
+	_ = x[actKillSubWord-49]
+	_ = x[actUnixLineDiscard-50]
+	_ = x[actUnixWordRubout-51]
+	_ = x[actYank-52]
+	_ = x[actBackwardKillWord-53]
+	_ = x[actBackwardKillSubWord-54]
+	_ = x[actSelectAll-55]
+	_ = x[actDeselectAll-56]
+	_ = x[actToggle-57]
+	_ = x[actToggleSearch-58]
+	_ = x[actToggleAll-59]
+	_ = x[actToggleDown-60]
+	_ = x[actToggleUp-61]
+	_ = x[actToggleIn-62]
+	_ = x[actToggleOut-63]
+	_ = x[actToggleTrack-64]
+	_ = x[actToggleTrackCurrent-65]
+	_ = x[actToggleHeader-66]
+	_ = x[actToggleWrap-67]
+	_ = x[actToggleWrapWord-68]
+	_ = x[actToggleMultiLine-69]
+	_ = x[actToggleHscroll-70]
+	_ = x[actToggleRaw-71]
+	_ = x[actEnableRaw-72]
+	_ = x[actDisableRaw-73]
+	_ = x[actTrackCurrent-74]
+	_ = x[actToggleInput-75]
+	_ = x[actHideInput-76]
+	_ = x[actShowInput-77]
+	_ = x[actUntrackCurrent-78]
+	_ = x[actDown-79]
+	_ = x[actDownMatch-80]
+	_ = x[actUp-81]
+	_ = x[actUpMatch-82]
+	_ = x[actPageUp-83]
+	_ = x[actPageDown-84]
+	_ = x[actPosition-85]
+	_ = x[actHalfPageUp-86]
+	_ = x[actHalfPageDown-87]
+	_ = x[actOffsetUp-88]
+	_ = x[actOffsetDown-89]
+	_ = x[actOffsetMiddle-90]
+	_ = x[actJump-91]
+	_ = x[actJumpAccept-92]
+	_ = x[actPrintQuery-93]
+	_ = x[actRefreshPreview-94]
+	_ = x[actReplaceQuery-95]
+	_ = x[actToggleSort-96]
+	_ = x[actShowPreview-97]
+	_ = x[actHidePreview-98]
+	_ = x[actTogglePreview-99]
+	_ = x[actTogglePreviewWrap-100]
+	_ = x[actTogglePreviewWrapWord-101]
+	_ = x[actTransform-102]
+	_ = x[actTransformBorderLabel-103]
+	_ = x[actTransformGhost-104]
+	_ = x[actTransformHeader-105]
+	_ = x[actTransformHeaderLines-106]
+	_ = x[actTransformFooter-107]
+	_ = x[actTransformHeaderLabel-108]
+	_ = x[actTransformFooterLabel-109]
+	_ = x[actTransformInputLabel-110]
+	_ = x[actTransformListLabel-111]
+	_ = x[actTransformNth-112]
+	_ = x[actTransformPointer-113]
+	_ = x[actTransformPreviewLabel-114]
+	_ = x[actTransformPrompt-115]
+	_ = x[actTransformQuery-116]
+	_ = x[actTransformSearch-117]
+	_ = x[actTrigger-118]
+	_ = x[actBgTransform-119]
+	_ = x[actBgTransformBorderLabel-120]
+	_ = x[actBgTransformGhost-121]
+	_ = x[actBgTransformHeader-122]
+	_ = x[actBgTransformHeaderLines-123]
+	_ = x[actBgTransformFooter-124]
+	_ = x[actBgTransformHeaderLabel-125]
+	_ = x[actBgTransformFooterLabel-126]
+	_ = x[actBgTransformInputLabel-127]
+	_ = x[actBgTransformListLabel-128]
+	_ = x[actBgTransformNth-129]
+	_ = x[actBgTransformPointer-130]
+	_ = x[actBgTransformPreviewLabel-131]
+	_ = x[actBgTransformPrompt-132]
+	_ = x[actBgTransformQuery-133]
+	_ = x[actBgTransformSearch-134]
+	_ = x[actBgCancel-135]
+	_ = x[actSearch-136]
+	_ = x[actPreview-137]
+	_ = x[actPreviewTop-138]
+	_ = x[actPreviewBottom-139]
+	_ = x[actPreviewUp-140]
+	_ = x[actPreviewDown-141]
+	_ = x[actPreviewPageUp-142]
+	_ = x[actPreviewPageDown-143]
+	_ = x[actPreviewHalfPageUp-144]
+	_ = x[actPreviewHalfPageDown-145]
+	_ = x[actPrevHistory-146]
+	_ = x[actPrevSelected-147]
+	_ = x[actPrint-148]
+	_ = x[actPut-149]
+	_ = x[actNextHistory-150]
+	_ = x[actNextSelected-151]
+	_ = x[actExecute-152]
+	_ = x[actExecuteSilent-153]
+	_ = x[actExecuteMulti-154]
+	_ = x[actSigStop-155]
+	_ = x[actBest-156]
+	_ = x[actFirst-157]
+	_ = x[actLast-158]
+	_ = x[actReload-159]
+	_ = x[actReloadSync-160]
+	_ = x[actDisableSearch-161]
+	_ = x[actEnableSearch-162]
+	_ = x[actSelect-163]
+	_ = x[actDeselect-164]
+	_ = x[actUnbind-165]
+	_ = x[actRebind-166]
+	_ = x[actToggleBind-167]
+	_ = x[actBecome-168]
+	_ = x[actShowHeader-169]
+	_ = x[actHideHeader-170]
+	_ = x[actBell-171]
+	_ = x[actExclude-172]
+	_ = x[actExcludeMulti-173]
+	_ = x[actAsync-174]
 }

-const _actionType_name = "actIgnoreactStartactClickactInvalidactBracketedPasteBeginactBracketedPasteEndactCharactMouseactBeginningOfLineactAbortactAcceptactAcceptNonEmptyactAcceptOrPrintQueryactBackwardCharactBackwardDeleteCharactBackwardDeleteCharEofactBackwardWordactBackwardSubWordactCancelactChangeBorderLabelactChangeGhostactChangeHeaderactChangeHeaderLinesactChangeFooteractChangeHeaderLabelactChangeFooterLabelactChangeInputLabelactChangeListLabelactChangeMultiactChangeNthactChangeWithNthactChangePointeractChangePreviewactChangePreviewLabelactChangePreviewWindowactChangePromptactChangeQueryactClearScreenactClearQueryactClearSelectionactCloseactDeleteCharactDeleteCharEofactEndOfLineactFatalactForwardCharactForwardWordactForwardSubWordactKillLineactKillWordactKillSubWordactUnixLineDiscardactUnixWordRuboutactYankactBackwardKillWordactBackwardKillSubWordactSelectAllactDeselectAllactToggleactToggleSearchactToggleAllactToggleDownactToggleUpactToggleInactToggleOutactToggleTrackactToggleTrackCurrentactToggleHeaderactToggleWrapactToggleWrapWordactToggleMultiLineactToggleHscrollactToggleRawactEnableRawactDisableRawactTrackCurrentactToggleInputactHideInputactShowInputactUntrackCurrentactDownactDownMatchactUpactUpMatchactPageUpactPageDownactPositionactHalfPageUpactHalfPageDownactOffsetUpactOffsetDownactOffsetMiddleactJumpactJumpAcceptactPrintQueryactRefreshPreviewactReplaceQueryactToggleSortactShowPreviewactHidePreviewactTogglePreviewactTogglePreviewWrapactTogglePreviewWrapWordactTransformactTransformBorderLabelactTransformGhostactTransformHeaderactTransformHeaderLinesactTransformFooteractTransformHeaderLabelactTransformFooterLabelactTransformInputLabelactTransformListLabelactTransformNthactTransformWithNthactTransformPointeractTransformPreviewLabelactTransformPromptactTransformQueryactTransformSearchactTriggeractBgTransformactBgTransformBorderLabelactBgTransformGhostactBgTransformHeaderactBgTransformHeaderLinesactBgTransformFooteractBgTransformHeaderLabelactBgTransformFooterLabelactBgTransformInputLabelactBgTransformListLabelactBgTransformNthactBgTransformWithNthactBgTransformPointeractBgTransformPreviewLabelactBgTransformPromptactBgTransformQueryactBgTransformSearchactBgCancelactSearchactPreviewactPreviewTopactPreviewBottomactPreviewUpactPreviewDownactPreviewPageUpactPreviewPageDownactPreviewHalfPageUpactPreviewHalfPageDownactPrevHistoryactPrevSelectedactPrintactPutactNextHistoryactNextSelectedactExecuteactExecuteSilentactExecuteMultiactSigStopactBestactFirstactLastactReloadactReloadSyncactDisableSearchactEnableSearchactSelectactDeselectactUnbindactRebindactToggleBindactBecomeactShowHeaderactHideHeaderactBellactExcludeactExcludeMultiactAsync"
+const _actionType_name = "actIgnoreactStartactClickactInvalidactBracketedPasteBeginactBracketedPasteEndactCharactMouseactBeginningOfLineactAbortactAcceptactAcceptNonEmptyactAcceptOrPrintQueryactBackwardCharactBackwardDeleteCharactBackwardDeleteCharEofactBackwardWordactBackwardSubWordactCancelactChangeBorderLabelactChangeGhostactChangeHeaderactChangeHeaderLinesactChangeFooteractChangeHeaderLabelactChangeFooterLabelactChangeInputLabelactChangeListLabelactChangeMultiactChangeNthactChangePointeractChangePreviewactChangePreviewLabelactChangePreviewWindowactChangePromptactChangeQueryactClearScreenactClearQueryactClearSelectionactCloseactDeleteCharactDeleteCharEofactEndOfLineactFatalactForwardCharactForwardWordactForwardSubWordactKillLineactKillWordactKillSubWordactUnixLineDiscardactUnixWordRuboutactYankactBackwardKillWordactBackwardKillSubWordactSelectAllactDeselectAllactToggleactToggleSearchactToggleAllactToggleDownactToggleUpactToggleInactToggleOutactToggleTrackactToggleTrackCurrentactToggleHeaderactToggleWrapactToggleWrapWordactToggleMultiLineactToggleHscrollactToggleRawactEnableRawactDisableRawactTrackCurrentactToggleInputactHideInputactShowInputactUntrackCurrentactDownactDownMatchactUpactUpMatchactPageUpactPageDownactPositionactHalfPageUpactHalfPageDownactOffsetUpactOffsetDownactOffsetMiddleactJumpactJumpAcceptactPrintQueryactRefreshPreviewactReplaceQueryactToggleSortactShowPreviewactHidePreviewactTogglePreviewactTogglePreviewWrapactTogglePreviewWrapWordactTransformactTransformBorderLabelactTransformGhostactTransformHeaderactTransformHeaderLinesactTransformFooteractTransformHeaderLabelactTransformFooterLabelactTransformInputLabelactTransformListLabelactTransformNthactTransformPointeractTransformPreviewLabelactTransformPromptactTransformQueryactTransformSearchactTriggeractBgTransformactBgTransformBorderLabelactBgTransformGhostactBgTransformHeaderactBgTransformHeaderLinesactBgTransformFooteractBgTransformHeaderLabelactBgTransformFooterLabelactBgTransformInputLabelactBgTransformListLabelactBgTransformNthactBgTransformPointeractBgTransformPreviewLabelactBgTransformPromptactBgTransformQueryactBgTransformSearchactBgCancelactSearchactPreviewactPreviewTopactPreviewBottomactPreviewUpactPreviewDownactPreviewPageUpactPreviewPageDownactPreviewHalfPageUpactPreviewHalfPageDownactPrevHistoryactPrevSelectedactPrintactPutactNextHistoryactNextSelectedactExecuteactExecuteSilentactExecuteMultiactSigStopactBestactFirstactLastactReloadactReloadSyncactDisableSearchactEnableSearchactSelectactDeselectactUnbindactRebindactToggleBindactBecomeactShowHeaderactHideHeaderactBellactExcludeactExcludeMultiactAsync"

-var _actionType_index = [...]uint16{0, 9, 17, 25, 35, 57, 77, 84, 92, 110, 118, 127, 144, 165, 180, 201, 225, 240, 258, 267, 287, 301, 316, 336, 351, 371, 391, 410, 428, 442, 454, 470, 486, 502, 523, 545, 560, 574, 588, 601, 618, 626, 639, 655, 667, 675, 689, 703, 720, 731, 742, 756, 774, 791, 798, 817, 839, 851, 865, 874, 889, 901, 914, 925, 936, 948, 962, 983, 998, 1011, 1028, 1046, 1062, 1074, 1086, 1099, 1114, 1128, 1140, 1152, 1169, 1176, 1188, 1193, 1203, 1212, 1223, 1234, 1247, 1262, 1273, 1286, 1301, 1308, 1321, 1334, 1351, 1366, 1379, 1393, 1407, 1423, 1443, 1467, 1479, 1502, 1519, 1537, 1560, 1578, 1601, 1624, 1646, 1667, 1682, 1701, 1720, 1744, 1762, 1779, 1797, 1807, 1821, 1846, 1865, 1885, 1910, 1930, 1955, 1980, 2004, 2027, 2044, 2065, 2086, 2112, 2132, 2151, 2171, 2182, 2191, 2201, 2214, 2230, 2242, 2256, 2272, 2290, 2310, 2332, 2346, 2361, 2369, 2375, 2389, 2404, 2414, 2430, 2445, 2455, 2462, 2470, 2477, 2486, 2499, 2515, 2530, 2539, 2550, 2559, 2568, 2581, 2590, 2603, 2616, 2623, 2633, 2648, 2656}
+var _actionType_index = [...]uint16{0, 9, 17, 25, 35, 57, 77, 84, 92, 110, 118, 127, 144, 165, 180, 201, 225, 240, 258, 267, 287, 301, 316, 336, 351, 371, 391, 410, 428, 442, 454, 470, 486, 507, 529, 544, 558, 572, 585, 602, 610, 623, 639, 651, 659, 673, 687, 704, 715, 726, 740, 758, 775, 782, 801, 823, 835, 849, 858, 873, 885, 898, 909, 920, 932, 946, 967, 982, 995, 1012, 1030, 1046, 1058, 1070, 1083, 1098, 1112, 1124, 1136, 1153, 1160, 1172, 1177, 1187, 1196, 1207, 1218, 1231, 1246, 1257, 1270, 1285, 1292, 1305, 1318, 1335, 1350, 1363, 1377, 1391, 1407, 1427, 1451, 1463, 1486, 1503, 1521, 1544, 1562, 1585, 1608, 1630, 1651, 1666, 1685, 1709, 1727, 1744, 1762, 1772, 1786, 1811, 1830, 1850, 1875, 1895, 1920, 1945, 1969, 1992, 2009, 2030, 2056, 2076, 2095, 2115, 2126, 2135, 2145, 2158, 2174, 2186, 2200, 2216, 2234, 2254, 2276, 2290, 2305, 2313, 2319, 2333, 2348, 2358, 2374, 2389, 2399, 2406, 2414, 2421, 2430, 2443, 2459, 2474, 2483, 2494, 2503, 2512, 2525, 2534, 2547, 2560, 2567, 2577, 2592, 2600}

 func (i actionType) String() string {
 	if i < 0 || i >= actionType(len(_actionType_index)-1) {
--- a/src/algo/SIMD.md
+++ b/src/algo/SIMD.md
@@ -1,99 +0,0 @@
-# SIMD byte search: `indexByteTwo` / `lastIndexByteTwo`
-
-## What these functions do
-
-`indexByteTwo(s []byte, b1, b2 byte) int` — returns the index of the
-**first** occurrence of `b1` or `b2` in `s`, or `-1`.
-
-`lastIndexByteTwo(s []byte, b1, b2 byte) int` — returns the index of the
-**last** occurrence of `b1` or `b2` in `s`, or `-1`.
-
-They are used by the fuzzy matching algorithm (`algo.go`) to skip ahead
-during case-insensitive search. Instead of calling `bytes.IndexByte` twice
-(once for lowercase, once for uppercase), a single SIMD pass finds both at
-once.
-
-## File layout
-
-| File                  | Purpose                                                           |
-| ------                | ---------                                                         |
-| `indexbyte2_arm64.go` | Go declarations (`//go:noescape`) for ARM64                       |
-| `indexbyte2_arm64.s`  | ARM64 NEON assembly (32-byte aligned blocks, syndrome extraction) |
-| `indexbyte2_amd64.go` | Go declarations + AVX2 runtime detection for AMD64                |
-| `indexbyte2_amd64.s`  | AMD64 AVX2/SSE2 assembly with CPUID dispatch                      |
-| `indexbyte2_other.go` | Pure Go fallback for all other architectures                      |
-| `indexbyte2_test.go`  | Unit tests, exhaustive tests, fuzz tests, and benchmarks          |
-
-## How the SIMD implementations work
-
-**ARM64 (NEON):**
- Broadcasts both needle bytes into NEON registers (`VMOV`).
- Processes 32-byte aligned chunks. For each chunk, compares all bytes
-  against both needles (`VCMEQ`), ORs the results (`VORR`), and builds a
-  64-bit syndrome with 2 bits per byte.
- `indexByteTwo` uses `RBIT` + `CLZ` to find the lowest set bit (first match).
- `lastIndexByteTwo` scans backward and uses `CLZ` on the raw syndrome to
-  find the highest set bit (last match).
- Handles alignment and partial first/last blocks with bit masking.
- Adapted from Go's `internal/bytealg/indexbyte_arm64.s`.
-
-**AMD64 (AVX2 with SSE2 fallback):**
- At init time, `cpuHasAVX2()` checks CPUID + XGETBV for AVX2 and OS YMM
-  support. The result is cached in `_useAVX2`.
- **AVX2 path** (inputs >= 32 bytes, when available):
-  - Broadcasts both needles via `VPBROADCASTB`.
-  - Processes 32-byte blocks: `VPCMPEQB` against both needles, `VPOR`, then
-    `VPMOVMSKB` to get a 32-bit mask.
-  - 5 instructions per loop iteration (vs 7 for SSE2) at 2x the throughput.
-  - `VZEROUPPER` before every return to avoid SSE/AVX transition penalties.
- **SSE2 fallback** (inputs < 32 bytes, or CPUs without AVX2):
-  - Broadcasts via `PUNPCKLBW` + `PSHUFL`.
-  - Processes 16-byte blocks: `PCMPEQB`, `POR`, `PMOVMSKB`.
-  - Small inputs (<16 bytes) are handled with page-boundary-safe loads.
- Both paths use `BSFL` (forward) / `BSRL` (reverse) for bit scanning.
- Adapted from Go's `internal/bytealg/indexbyte_amd64.s`.
-
-**Fallback (other platforms):**
- `indexByteTwo` uses two `bytes.IndexByte` calls with scope-limiting
-  (search `b1` first, then limit the `b2` search to `s[:i1]`).
- `lastIndexByteTwo` uses a simple backward for loop.
-
-## Running tests
-
-```bash
-# Unit + exhaustive tests
-go test ./src/algo/ -run 'TestIndexByteTwo|TestLastIndexByteTwo' -v
-
-# Fuzz tests (run for 10 seconds each)
-go test ./src/algo/ -run '^$' -fuzz FuzzIndexByteTwo -fuzztime 10s
-go test ./src/algo/ -run '^$' -fuzz FuzzLastIndexByteTwo -fuzztime 10s
-
-# Cross-architecture: test amd64 on an arm64 Mac (via Rosetta)
-GOARCH=amd64 go test ./src/algo/ -run 'TestIndexByteTwo|TestLastIndexByteTwo' -v
-GOARCH=amd64 go test ./src/algo/ -run '^$' -fuzz FuzzIndexByteTwo -fuzztime 10s
-GOARCH=amd64 go test ./src/algo/ -run '^$' -fuzz FuzzLastIndexByteTwo -fuzztime 10s
-```
-
-## Running micro-benchmarks
-
-```bash
-# All indexByteTwo / lastIndexByteTwo benchmarks
-go test ./src/algo/ -bench 'IndexByteTwo' -benchmem
-
-# Specific size
-go test ./src/algo/ -bench 'IndexByteTwo_1000'
-```
-
-Each benchmark compares the SIMD `asm` implementation against reference
-implementations (`2xIndexByte` using `bytes.IndexByte`, and a simple `loop`).
-
-## Correctness verification
-
-The assembly is verified by three layers of testing:
-
-1. **Table-driven tests** — known inputs with expected outputs.
-2. **Exhaustive tests** — all lengths 0–256, every match position, no-match
-   cases, and both-bytes-present cases, compared against a simple loop
-   reference.
-3. **Fuzz tests** — randomized inputs via `testing.F`, compared against the
-   same loop reference.
--- a/src/algo/algo.go
+++ b/src/algo/algo.go
@@ -321,15 +321,22 @@ type Algo func(caseSensitive bool, normalize bool, forward bool, input *util.Cha

 func trySkip(input *util.Chars, caseSensitive bool, b byte, from int) int {
 	byteArray := input.Bytes()[from:]
-	// For case-insensitive search of a letter, search for both cases in one pass
-	if !caseSensitive && b >= 'a' && b <= 'z' {
-		idx := IndexByteTwo(byteArray, b, b-32)
-		if idx < 0 {
-			return -1
-		}
-		return from + idx
-	}
 	idx := bytes.IndexByte(byteArray, b)
+	if idx == 0 {
+		// Can't skip any further
+		return from
+	}
+	// We may need to search for the uppercase letter again. We don't have to
+	// consider normalization as we can be sure that this is an ASCII string.
+	if !caseSensitive && b >= 'a' && b <= 'z' {
+		if idx > 0 {
+			byteArray = byteArray[:idx]
+		}
+		uidx := bytes.IndexByte(byteArray, b-32)
+		if uidx >= 0 {
+			idx = uidx
+		}
+	}
 	if idx < 0 {
 		return -1
 	}
@@ -373,17 +380,14 @@ func asciiFuzzyIndex(input *util.Chars, pattern []rune, caseSensitive bool) (int
 	}

 	// Find the last appearance of the last character of the pattern to limit the search scope
-	scope := input.Bytes()[lastIdx:]
-	if len(scope) > 1 {
-		tail := scope[1:]
-		var end int
+	bu := b
 	if !caseSensitive && b >= 'a' && b <= 'z' {
-			end = lastIndexByteTwo(tail, b, b-32)
-		} else {
-			end = bytes.LastIndexByte(tail, b)
+		bu = b - 32
 	}
-		if end >= 0 {
-			return firstIdx, lastIdx + 1 + end + 1
+	scope := input.Bytes()[lastIdx:]
+	for offset := len(scope) - 1; offset > 0; offset-- {
+		if scope[offset] == b || scope[offset] == bu {
+			return firstIdx, lastIdx + offset + 1
 		}
 	}
 	return firstIdx, lastIdx + 1
--- a/src/algo/indexbyte2_amd64.go
+++ b/src/algo/indexbyte2_amd64.go
@@ -1,24 +0,0 @@
-//go:build amd64
-
-package algo
-
-var _useAVX2 bool
-
-func init() {
-	_useAVX2 = cpuHasAVX2()
-}
-
-//go:noescape
-func cpuHasAVX2() bool
-
-// indexByteTwo returns the index of the first occurrence of b1 or b2 in s,
-// or -1 if neither is present. Uses AVX2 when available, SSE2 otherwise.
-//
-//go:noescape
-func IndexByteTwo(s []byte, b1, b2 byte) int
-
-// lastIndexByteTwo returns the index of the last occurrence of b1 or b2 in s,
-// or -1 if neither is present. Uses AVX2 when available, SSE2 otherwise.
-//
-//go:noescape
-func lastIndexByteTwo(s []byte, b1, b2 byte) int
--- a/src/algo/indexbyte2_amd64.s
+++ b/src/algo/indexbyte2_amd64.s
@@ -1,377 +0,0 @@
-#include "textflag.h"
-
-// func cpuHasAVX2() bool
-//
-// Checks CPUID and XGETBV for AVX2 + OS YMM support.
-TEXT ·cpuHasAVX2(SB),NOSPLIT,$0-1
-	MOVQ	BX, R8          // save BX (callee-saved, clobbered by CPUID)
-
-	// Check max CPUID leaf >= 7
-	MOVL	$0, AX
-	CPUID
-	CMPL	AX, $7
-	JL	cpuid_no
-
-	// Check OSXSAVE (CPUID.1:ECX bit 27)
-	MOVL	$1, AX
-	CPUID
-	TESTL	$(1<<27), CX
-	JZ	cpuid_no
-
-	// Check AVX2 (CPUID.7.0:EBX bit 5)
-	MOVL	$7, AX
-	MOVL	$0, CX
-	CPUID
-	TESTL	$(1<<5), BX
-	JZ	cpuid_no
-
-	// Check OS YMM state support via XGETBV
-	MOVL	$0, CX
-	BYTE	$0x0F; BYTE $0x01; BYTE $0xD0  // XGETBV → EDX:EAX
-	ANDL	$6, AX          // bits 1 (XMM) and 2 (YMM)
-	CMPL	AX, $6
-	JNE	cpuid_no
-
-	MOVQ	R8, BX          // restore BX
-	MOVB	$1, ret+0(FP)
-	RET
-
-cpuid_no:
-	MOVQ	R8, BX
-	MOVB	$0, ret+0(FP)
-	RET
-
-// func IndexByteTwo(s []byte, b1, b2 byte) int
-//
-// Returns the index of the first occurrence of b1 or b2 in s, or -1.
-// Uses AVX2 (32 bytes/iter) when available, SSE2 (16 bytes/iter) otherwise.
-TEXT ·IndexByteTwo(SB),NOSPLIT,$0-40
-	MOVQ	s_base+0(FP), SI
-	MOVQ	s_len+8(FP), BX
-	MOVBLZX	b1+24(FP), AX
-	MOVBLZX	b2+25(FP), CX
-	LEAQ	ret+32(FP), R8
-
-	TESTQ	BX, BX
-	JEQ	fwd_failure
-
-	// Try AVX2 for inputs >= 32 bytes
-	CMPQ	BX, $32
-	JLT	fwd_sse2
-	CMPB	·_useAVX2(SB), $1
-	JNE	fwd_sse2
-
-	// ====== AVX2 forward search ======
-	MOVD	AX, X0
-	VPBROADCASTB	X0, Y0       // Y0 = splat(b1)
-	MOVD	CX, X1
-	VPBROADCASTB	X1, Y1       // Y1 = splat(b2)
-
-	MOVQ	SI, DI
-	LEAQ	-32(SI)(BX*1), AX    // AX = last valid 32-byte chunk
-	JMP	fwd_avx2_entry
-
-fwd_avx2_loop:
-	VMOVDQU	(DI), Y2
-	VPCMPEQB	Y0, Y2, Y3
-	VPCMPEQB	Y1, Y2, Y4
-	VPOR	Y3, Y4, Y3
-	VPMOVMSKB	Y3, DX
-	BSFL	DX, DX
-	JNZ	fwd_avx2_success
-	ADDQ	$32, DI
-
-fwd_avx2_entry:
-	CMPQ	DI, AX
-	JB	fwd_avx2_loop
-
-	// Last 32-byte chunk (may overlap with previous)
-	MOVQ	AX, DI
-	VMOVDQU	(AX), Y2
-	VPCMPEQB	Y0, Y2, Y3
-	VPCMPEQB	Y1, Y2, Y4
-	VPOR	Y3, Y4, Y3
-	VPMOVMSKB	Y3, DX
-	BSFL	DX, DX
-	JNZ	fwd_avx2_success
-
-	MOVQ	$-1, (R8)
-	VZEROUPPER
-	RET
-
-fwd_avx2_success:
-	SUBQ	SI, DI
-	ADDQ	DX, DI
-	MOVQ	DI, (R8)
-	VZEROUPPER
-	RET
-
-	// ====== SSE2 forward search (< 32 bytes or no AVX2) ======
-
-fwd_sse2:
-	// Broadcast b1 into X0
-	MOVD	AX, X0
-	PUNPCKLBW	X0, X0
-	PUNPCKLBW	X0, X0
-	PSHUFL	$0, X0, X0
-
-	// Broadcast b2 into X4
-	MOVD	CX, X4
-	PUNPCKLBW	X4, X4
-	PUNPCKLBW	X4, X4
-	PSHUFL	$0, X4, X4
-
-	CMPQ	BX, $16
-	JLT	fwd_small
-
-	MOVQ	SI, DI
-	LEAQ	-16(SI)(BX*1), AX
-	JMP	fwd_sseloopentry
-
-fwd_sseloop:
-	MOVOU	(DI), X1
-	MOVOU	X1, X2
-	PCMPEQB	X0, X1
-	PCMPEQB	X4, X2
-	POR	X2, X1
-	PMOVMSKB	X1, DX
-	BSFL	DX, DX
-	JNZ	fwd_ssesuccess
-	ADDQ	$16, DI
-
-fwd_sseloopentry:
-	CMPQ	DI, AX
-	JB	fwd_sseloop
-
-	// Search the last 16-byte chunk (may overlap)
-	MOVQ	AX, DI
-	MOVOU	(AX), X1
-	MOVOU	X1, X2
-	PCMPEQB	X0, X1
-	PCMPEQB	X4, X2
-	POR	X2, X1
-	PMOVMSKB	X1, DX
-	BSFL	DX, DX
-	JNZ	fwd_ssesuccess
-
-fwd_failure:
-	MOVQ	$-1, (R8)
-	RET
-
-fwd_ssesuccess:
-	SUBQ	SI, DI
-	ADDQ	DX, DI
-	MOVQ	DI, (R8)
-	RET
-
-fwd_small:
-	// Check if loading 16 bytes from SI would cross a page boundary
-	LEAQ	16(SI), AX
-	TESTW	$0xff0, AX
-	JEQ	fwd_endofpage
-
-	MOVOU	(SI), X1
-	MOVOU	X1, X2
-	PCMPEQB	X0, X1
-	PCMPEQB	X4, X2
-	POR	X2, X1
-	PMOVMSKB	X1, DX
-	BSFL	DX, DX
-	JZ	fwd_failure
-	CMPL	DX, BX
-	JAE	fwd_failure
-	MOVQ	DX, (R8)
-	RET
-
-fwd_endofpage:
-	MOVOU	-16(SI)(BX*1), X1
-	MOVOU	X1, X2
-	PCMPEQB	X0, X1
-	PCMPEQB	X4, X2
-	POR	X2, X1
-	PMOVMSKB	X1, DX
-	MOVL	BX, CX
-	SHLL	CX, DX
-	SHRL	$16, DX
-	BSFL	DX, DX
-	JZ	fwd_failure
-	MOVQ	DX, (R8)
-	RET
-
-// func lastIndexByteTwo(s []byte, b1, b2 byte) int
-//
-// Returns the index of the last occurrence of b1 or b2 in s, or -1.
-// Uses AVX2 (32 bytes/iter) when available, SSE2 (16 bytes/iter) otherwise.
-TEXT ·lastIndexByteTwo(SB),NOSPLIT,$0-40
-	MOVQ	s_base+0(FP), SI
-	MOVQ	s_len+8(FP), BX
-	MOVBLZX	b1+24(FP), AX
-	MOVBLZX	b2+25(FP), CX
-	LEAQ	ret+32(FP), R8
-
-	TESTQ	BX, BX
-	JEQ	back_failure
-
-	// Try AVX2 for inputs >= 32 bytes
-	CMPQ	BX, $32
-	JLT	back_sse2
-	CMPB	·_useAVX2(SB), $1
-	JNE	back_sse2
-
-	// ====== AVX2 backward search ======
-	MOVD	AX, X0
-	VPBROADCASTB	X0, Y0
-	MOVD	CX, X1
-	VPBROADCASTB	X1, Y1
-
-	// DI = start of last 32-byte chunk
-	LEAQ	-32(SI)(BX*1), DI
-
-back_avx2_loop:
-	CMPQ	DI, SI
-	JBE	back_avx2_first
-
-	VMOVDQU	(DI), Y2
-	VPCMPEQB	Y0, Y2, Y3
-	VPCMPEQB	Y1, Y2, Y4
-	VPOR	Y3, Y4, Y3
-	VPMOVMSKB	Y3, DX
-	BSRL	DX, DX
-	JNZ	back_avx2_success
-	SUBQ	$32, DI
-	JMP	back_avx2_loop
-
-back_avx2_first:
-	// First 32 bytes (DI <= SI, load from SI)
-	VMOVDQU	(SI), Y2
-	VPCMPEQB	Y0, Y2, Y3
-	VPCMPEQB	Y1, Y2, Y4
-	VPOR	Y3, Y4, Y3
-	VPMOVMSKB	Y3, DX
-	BSRL	DX, DX
-	JNZ	back_avx2_firstsuccess
-
-	MOVQ	$-1, (R8)
-	VZEROUPPER
-	RET
-
-back_avx2_success:
-	SUBQ	SI, DI
-	ADDQ	DX, DI
-	MOVQ	DI, (R8)
-	VZEROUPPER
-	RET
-
-back_avx2_firstsuccess:
-	MOVQ	DX, (R8)
-	VZEROUPPER
-	RET
-
-	// ====== SSE2 backward search (< 32 bytes or no AVX2) ======
-
-back_sse2:
-	// Broadcast b1 into X0
-	MOVD	AX, X0
-	PUNPCKLBW	X0, X0
-	PUNPCKLBW	X0, X0
-	PSHUFL	$0, X0, X0
-
-	// Broadcast b2 into X4
-	MOVD	CX, X4
-	PUNPCKLBW	X4, X4
-	PUNPCKLBW	X4, X4
-	PSHUFL	$0, X4, X4
-
-	CMPQ	BX, $16
-	JLT	back_small
-
-	// DI = start of last 16-byte chunk
-	LEAQ	-16(SI)(BX*1), DI
-
-back_sseloop:
-	CMPQ	DI, SI
-	JBE	back_ssefirst
-
-	MOVOU	(DI), X1
-	MOVOU	X1, X2
-	PCMPEQB	X0, X1
-	PCMPEQB	X4, X2
-	POR	X2, X1
-	PMOVMSKB	X1, DX
-	BSRL	DX, DX
-	JNZ	back_ssesuccess
-	SUBQ	$16, DI
-	JMP	back_sseloop
-
-back_ssefirst:
-	// First 16 bytes (DI <= SI, load from SI)
-	MOVOU	(SI), X1
-	MOVOU	X1, X2
-	PCMPEQB	X0, X1
-	PCMPEQB	X4, X2
-	POR	X2, X1
-	PMOVMSKB	X1, DX
-	BSRL	DX, DX
-	JNZ	back_ssefirstsuccess
-
-back_failure:
-	MOVQ	$-1, (R8)
-	RET
-
-back_ssesuccess:
-	SUBQ	SI, DI
-	ADDQ	DX, DI
-	MOVQ	DI, (R8)
-	RET
-
-back_ssefirstsuccess:
-	// DX = byte offset from base
-	MOVQ	DX, (R8)
-	RET
-
-back_small:
-	// Check page boundary
-	LEAQ	16(SI), AX
-	TESTW	$0xff0, AX
-	JEQ	back_endofpage
-
-	MOVOU	(SI), X1
-	MOVOU	X1, X2
-	PCMPEQB	X0, X1
-	PCMPEQB	X4, X2
-	POR	X2, X1
-	PMOVMSKB	X1, DX
-	// Mask to first BX bytes: keep bits 0..BX-1
-	MOVL	$1, AX
-	MOVL	BX, CX
-	SHLL	CX, AX
-	DECL	AX
-	ANDL	AX, DX
-	BSRL	DX, DX
-	JZ	back_failure
-	MOVQ	DX, (R8)
-	RET
-
-back_endofpage:
-	// Load 16 bytes ending at base+n
-	MOVOU	-16(SI)(BX*1), X1
-	MOVOU	X1, X2
-	PCMPEQB	X0, X1
-	PCMPEQB	X4, X2
-	POR	X2, X1
-	PMOVMSKB	X1, DX
-	// Bits correspond to bytes [base+n-16, base+n).
-	// We want original bytes [0, n), which are bits [16-n, 16).
-	// Mask: keep bits (16-n) through 15.
-	MOVL	$16, CX
-	SUBL	BX, CX
-	SHRL	CX, DX
-	SHLL	CX, DX
-	BSRL	DX, DX
-	JZ	back_failure
-	// DX is the bit position in the loaded chunk.
-	// Original byte index = DX - (16 - n) = DX + n - 16
-	ADDL	BX, DX
-	SUBL	$16, DX
-	MOVQ	DX, (R8)
-	RET
--- a/src/algo/indexbyte2_arm64.go
+++ b/src/algo/indexbyte2_arm64.go
@@ -1,17 +0,0 @@
-//go:build arm64
-
-package algo
-
-// indexByteTwo returns the index of the first occurrence of b1 or b2 in s,
-// or -1 if neither is present. Implemented in assembly using ARM64 NEON
-// to search for both bytes in a single pass.
-//
-//go:noescape
-func IndexByteTwo(s []byte, b1, b2 byte) int
-
-// lastIndexByteTwo returns the index of the last occurrence of b1 or b2 in s,
-// or -1 if neither is present. Implemented in assembly using ARM64 NEON,
-// scanning backward.
-//
-//go:noescape
-func lastIndexByteTwo(s []byte, b1, b2 byte) int
--- a/src/algo/indexbyte2_arm64.s
+++ b/src/algo/indexbyte2_arm64.s
@@ -1,249 +0,0 @@
-#include "textflag.h"
-
-// func IndexByteTwo(s []byte, b1, b2 byte) int
-//
-// Returns the index of the first occurrence of b1 or b2 in s, or -1.
-// Uses ARM64 NEON to search for both bytes in a single pass over the data.
-// Adapted from Go's internal/bytealg/indexbyte_arm64.s (single-byte version).
-TEXT ·IndexByteTwo(SB),NOSPLIT,$0-40
-	MOVD	s_base+0(FP), R0
-	MOVD	s_len+8(FP), R2
-	MOVBU	b1+24(FP), R1
-	MOVBU	b2+25(FP), R7
-	MOVD	$ret+32(FP), R8
-
-	// Core algorithm:
-	// For each 32-byte chunk we calculate a 64-bit syndrome value,
-	// with two bits per byte. We compare against both b1 and b2,
-	// OR the results, then use the same syndrome extraction as
-	// Go's IndexByte.
-
-	CBZ	R2, fail
-	MOVD	R0, R11
-	// Magic constant 0x40100401 allows us to identify which lane matches.
-	// Each byte in the group of 4 gets a distinct bit: 1, 4, 16, 64.
-	MOVD	$0x40100401, R5
-	VMOV	R1, V0.B16    // V0 = splat(b1)
-	VMOV	R7, V7.B16    // V7 = splat(b2)
-	// Work with aligned 32-byte chunks
-	BIC	$0x1f, R0, R3
-	VMOV	R5, V5.S4
-	ANDS	$0x1f, R0, R9
-	AND	$0x1f, R2, R10
-	BEQ	loop
-
-	// Input string is not 32-byte aligned. Process the first
-	// aligned 32-byte block and mask off bytes before our start.
-	VLD1.P	(R3), [V1.B16, V2.B16]
-	SUB	$0x20, R9, R4
-	ADDS	R4, R2, R2
-	// Compare against both needles
-	VCMEQ	V0.B16, V1.B16, V3.B16  // b1 vs first 16 bytes
-	VCMEQ	V7.B16, V1.B16, V8.B16  // b2 vs first 16 bytes
-	VORR	V8.B16, V3.B16, V3.B16  // combine
-	VCMEQ	V0.B16, V2.B16, V4.B16  // b1 vs second 16 bytes
-	VCMEQ	V7.B16, V2.B16, V9.B16  // b2 vs second 16 bytes
-	VORR	V9.B16, V4.B16, V4.B16  // combine
-	// Build syndrome
-	VAND	V5.B16, V3.B16, V3.B16
-	VAND	V5.B16, V4.B16, V4.B16
-	VADDP	V4.B16, V3.B16, V6.B16
-	VADDP	V6.B16, V6.B16, V6.B16
-	VMOV	V6.D[0], R6
-	// Clear the irrelevant lower bits
-	LSL	$1, R9, R4
-	LSR	R4, R6, R6
-	LSL	R4, R6, R6
-	// The first block can also be the last
-	BLS	masklast
-	// Have we found something already?
-	CBNZ	R6, tail
-
-loop:
-	VLD1.P	(R3), [V1.B16, V2.B16]
-	SUBS	$0x20, R2, R2
-	// Compare against both needles, OR results
-	VCMEQ	V0.B16, V1.B16, V3.B16
-	VCMEQ	V7.B16, V1.B16, V8.B16
-	VORR	V8.B16, V3.B16, V3.B16
-	VCMEQ	V0.B16, V2.B16, V4.B16
-	VCMEQ	V7.B16, V2.B16, V9.B16
-	VORR	V9.B16, V4.B16, V4.B16
-	// If we're out of data we finish regardless of the result
-	BLS	end
-	// Fast check: OR both halves and check for any match
-	VORR	V4.B16, V3.B16, V6.B16
-	VADDP	V6.D2, V6.D2, V6.D2
-	VMOV	V6.D[0], R6
-	CBZ	R6, loop
-
-end:
-	// Found something or out of data — build full syndrome
-	VAND	V5.B16, V3.B16, V3.B16
-	VAND	V5.B16, V4.B16, V4.B16
-	VADDP	V4.B16, V3.B16, V6.B16
-	VADDP	V6.B16, V6.B16, V6.B16
-	VMOV	V6.D[0], R6
-	// Only mask for the last block
-	BHS	tail
-
-masklast:
-	// Clear irrelevant upper bits
-	ADD	R9, R10, R4
-	AND	$0x1f, R4, R4
-	SUB	$0x20, R4, R4
-	NEG	R4<<1, R4
-	LSL	R4, R6, R6
-	LSR	R4, R6, R6
-
-tail:
-	CBZ	R6, fail
-	RBIT	R6, R6
-	SUB	$0x20, R3, R3
-	CLZ	R6, R6
-	ADD	R6>>1, R3, R0
-	SUB	R11, R0, R0
-	MOVD	R0, (R8)
-	RET
-
-fail:
-	MOVD	$-1, R0
-	MOVD	R0, (R8)
-	RET
-
-// func lastIndexByteTwo(s []byte, b1, b2 byte) int
-//
-// Returns the index of the last occurrence of b1 or b2 in s, or -1.
-// Scans backward using ARM64 NEON.
-TEXT ·lastIndexByteTwo(SB),NOSPLIT,$0-40
-	MOVD	s_base+0(FP), R0
-	MOVD	s_len+8(FP), R2
-	MOVBU	b1+24(FP), R1
-	MOVBU	b2+25(FP), R7
-	MOVD	$ret+32(FP), R8
-
-	CBZ	R2, lfail
-	MOVD	R0, R11          // save base
-	ADD	R0, R2, R12      // R12 = end = base + len
-	MOVD	$0x40100401, R5
-	VMOV	R1, V0.B16       // V0 = splat(b1)
-	VMOV	R7, V7.B16       // V7 = splat(b2)
-	VMOV	R5, V5.S4
-
-	// Align: find the aligned block containing the last byte
-	SUB	$1, R12, R3
-	BIC	$0x1f, R3, R3    // R3 = start of aligned block containing last byte
-
-	// --- Process tail block ---
-	VLD1	(R3), [V1.B16, V2.B16]
-	VCMEQ	V0.B16, V1.B16, V3.B16
-	VCMEQ	V7.B16, V1.B16, V8.B16
-	VORR	V8.B16, V3.B16, V3.B16
-	VCMEQ	V0.B16, V2.B16, V4.B16
-	VCMEQ	V7.B16, V2.B16, V9.B16
-	VORR	V9.B16, V4.B16, V4.B16
-	VAND	V5.B16, V3.B16, V3.B16
-	VAND	V5.B16, V4.B16, V4.B16
-	VADDP	V4.B16, V3.B16, V6.B16
-	VADDP	V6.B16, V6.B16, V6.B16
-	VMOV	V6.D[0], R6
-
-	// Mask upper bits (bytes past end of slice)
-	// tail_bytes = end - R3 (1..32)
-	SUB	R3, R12, R10     // R10 = tail_bytes
-	MOVD	$64, R4
-	SUB	R10<<1, R4, R4   // R4 = 64 - 2*tail_bytes
-	LSL	R4, R6, R6
-	LSR	R4, R6, R6
-
-	// Is this also the head block?
-	CMP	R11, R3          // R3 - R11
-	BLO	lmaskfirst       // R3 < base: head+tail in same block
-	BEQ	ltailonly         // R3 == base: single aligned block
-
-	// R3 > base: more blocks before this one
-	CBNZ	R6, llast
-	B	lbacksetup
-
-ltailonly:
-	// Single block, already masked upper bits
-	CBNZ	R6, llast
-	B	lfail
-
-lmaskfirst:
-	// Mask lower bits (bytes before start of slice)
-	SUB	R3, R11, R4      // R4 = base - R3
-	LSL	$1, R4, R4
-	LSR	R4, R6, R6
-	LSL	R4, R6, R6
-	CBNZ	R6, llast
-	B	lfail
-
-lbacksetup:
-	SUB	$0x20, R3
-
-lbackloop:
-	VLD1	(R3), [V1.B16, V2.B16]
-	VCMEQ	V0.B16, V1.B16, V3.B16
-	VCMEQ	V7.B16, V1.B16, V8.B16
-	VORR	V8.B16, V3.B16, V3.B16
-	VCMEQ	V0.B16, V2.B16, V4.B16
-	VCMEQ	V7.B16, V2.B16, V9.B16
-	VORR	V9.B16, V4.B16, V4.B16
-	// Quick check: any match in this block?
-	VORR	V4.B16, V3.B16, V6.B16
-	VADDP	V6.D2, V6.D2, V6.D2
-	VMOV	V6.D[0], R6
-
-	// Is this a head block? (R3 < base)
-	CMP	R11, R3
-	BLO	lheadblock
-
-	// Full block (R3 >= base)
-	CBNZ	R6, lbackfound
-	// More blocks?
-	BEQ	lfail            // R3 == base, no more
-	SUB	$0x20, R3
-	B	lbackloop
-
-lbackfound:
-	// Build full syndrome
-	VAND	V5.B16, V3.B16, V3.B16
-	VAND	V5.B16, V4.B16, V4.B16
-	VADDP	V4.B16, V3.B16, V6.B16
-	VADDP	V6.B16, V6.B16, V6.B16
-	VMOV	V6.D[0], R6
-	B	llast
-
-lheadblock:
-	// R3 < base. Build full syndrome if quick check had a match.
-	CBZ	R6, lfail
-	VAND	V5.B16, V3.B16, V3.B16
-	VAND	V5.B16, V4.B16, V4.B16
-	VADDP	V4.B16, V3.B16, V6.B16
-	VADDP	V6.B16, V6.B16, V6.B16
-	VMOV	V6.D[0], R6
-	// Mask lower bits
-	SUB	R3, R11, R4      // R4 = base - R3
-	LSL	$1, R4, R4
-	LSR	R4, R6, R6
-	LSL	R4, R6, R6
-	CBZ	R6, lfail
-
-llast:
-	// Find last match: highest set bit in syndrome
-	// Syndrome has bit 2i set for matching byte i.
-	// CLZ gives leading zeros; byte_offset = (63 - CLZ) / 2.
-	CLZ	R6, R6
-	MOVD	$63, R4
-	SUB	R6, R4, R6       // R6 = 63 - CLZ = bit position
-	LSR	$1, R6            // R6 = byte offset within block
-	ADD	R3, R6, R0        // R0 = absolute address
-	SUB	R11, R0, R0       // R0 = slice index
-	MOVD	R0, (R8)
-	RET
-
-lfail:
-	MOVD	$-1, R0
-	MOVD	R0, (R8)
-	RET
--- a/src/algo/indexbyte2_other.go
+++ b/src/algo/indexbyte2_other.go
@@ -1,33 +0,0 @@
-//go:build !arm64 && !amd64
-
-package algo
-
-import "bytes"
-
-// indexByteTwo returns the index of the first occurrence of b1 or b2 in s,
-// or -1 if neither is present.
-func IndexByteTwo(s []byte, b1, b2 byte) int {
-	i1 := bytes.IndexByte(s, b1)
-	if i1 == 0 {
-		return 0
-	}
-	scope := s
-	if i1 > 0 {
-		scope = s[:i1]
-	}
-	if i2 := bytes.IndexByte(scope, b2); i2 >= 0 {
-		return i2
-	}
-	return i1
-}
-
-// lastIndexByteTwo returns the index of the last occurrence of b1 or b2 in s,
-// or -1 if neither is present.
-func lastIndexByteTwo(s []byte, b1, b2 byte) int {
-	for i := len(s) - 1; i >= 0; i-- {
-		if s[i] == b1 || s[i] == b2 {
-			return i
-		}
-	}
-	return -1
-}
--- a/src/algo/indexbyte2_test.go
+++ b/src/algo/indexbyte2_test.go
@@ -1,259 +0,0 @@
-package algo
-
-import (
-	"bytes"
-	"testing"
-)
-
-func TestIndexByteTwo(t *testing.T) {
-	tests := []struct {
-		name string
-		s    string
-		b1   byte
-		b2   byte
-		want int
-	}{
-		{"empty", "", 'a', 'b', -1},
-		{"single_b1", "a", 'a', 'b', 0},
-		{"single_b2", "b", 'a', 'b', 0},
-		{"single_none", "c", 'a', 'b', -1},
-		{"b1_first", "xaxb", 'a', 'b', 1},
-		{"b2_first", "xbxa", 'a', 'b', 1},
-		{"same_byte", "xxa", 'a', 'a', 2},
-		{"at_end", "xxxxa", 'a', 'b', 4},
-		{"not_found", "xxxxxxxx", 'a', 'b', -1},
-		{"long_b1_at_3000", string(make([]byte, 3000)) + "a" + string(make([]byte, 1000)), 'a', 'b', 3000},
-		{"long_b2_at_3000", string(make([]byte, 3000)) + "b" + string(make([]byte, 1000)), 'a', 'b', 3000},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			got := IndexByteTwo([]byte(tt.s), tt.b1, tt.b2)
-			if got != tt.want {
-				t.Errorf("IndexByteTwo(%q, %c, %c) = %d, want %d", tt.s[:min(len(tt.s), 40)], tt.b1, tt.b2, got, tt.want)
-			}
-		})
-	}
-
-	// Exhaustive test: compare against loop reference for various lengths,
-	// including sizes around SIMD block boundaries (16, 32, 64).
-	for n := 0; n <= 256; n++ {
-		data := make([]byte, n)
-		for i := range data {
-			data[i] = byte('c' + (i % 20))
-		}
-		// Test with match at every position
-		for pos := 0; pos < n; pos++ {
-			for _, b := range []byte{'A', 'B'} {
-				data[pos] = b
-				got := IndexByteTwo(data, 'A', 'B')
-				want := loopIndexByteTwo(data, 'A', 'B')
-				if got != want {
-					t.Fatalf("IndexByteTwo(len=%d, match=%c@%d) = %d, want %d", n, b, pos, got, want)
-				}
-				data[pos] = byte('c' + (pos % 20))
-			}
-		}
-		// Test with no match
-		got := IndexByteTwo(data, 'A', 'B')
-		if got != -1 {
-			t.Fatalf("IndexByteTwo(len=%d, no match) = %d, want -1", n, got)
-		}
-		// Test with both bytes present
-		if n >= 2 {
-			data[n/3] = 'A'
-			data[n*2/3] = 'B'
-			got := IndexByteTwo(data, 'A', 'B')
-			want := loopIndexByteTwo(data, 'A', 'B')
-			if got != want {
-				t.Fatalf("IndexByteTwo(len=%d, both@%d,%d) = %d, want %d", n, n/3, n*2/3, got, want)
-			}
-			data[n/3] = byte('c' + ((n / 3) % 20))
-			data[n*2/3] = byte('c' + ((n * 2 / 3) % 20))
-		}
-	}
-}
-
-func TestLastIndexByteTwo(t *testing.T) {
-	tests := []struct {
-		name string
-		s    string
-		b1   byte
-		b2   byte
-		want int
-	}{
-		{"empty", "", 'a', 'b', -1},
-		{"single_b1", "a", 'a', 'b', 0},
-		{"single_b2", "b", 'a', 'b', 0},
-		{"single_none", "c", 'a', 'b', -1},
-		{"b1_last", "xbxa", 'a', 'b', 3},
-		{"b2_last", "xaxb", 'a', 'b', 3},
-		{"same_byte", "axx", 'a', 'a', 0},
-		{"at_start", "axxxx", 'a', 'b', 0},
-		{"both_present", "axbx", 'a', 'b', 2},
-		{"not_found", "xxxxxxxx", 'a', 'b', -1},
-		{"long_b1_at_3000", string(make([]byte, 3000)) + "a" + string(make([]byte, 1000)), 'a', 'b', 3000},
-		{"long_b2_at_end", string(make([]byte, 4000)) + "b", 'a', 'b', 4000},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			got := lastIndexByteTwo([]byte(tt.s), tt.b1, tt.b2)
-			if got != tt.want {
-				t.Errorf("lastIndexByteTwo(%q, %c, %c) = %d, want %d", tt.s[:min(len(tt.s), 40)], tt.b1, tt.b2, got, tt.want)
-			}
-		})
-	}
-
-	// Exhaustive test against loop reference
-	for n := 0; n <= 256; n++ {
-		data := make([]byte, n)
-		for i := range data {
-			data[i] = byte('c' + (i % 20))
-		}
-		for pos := 0; pos < n; pos++ {
-			for _, b := range []byte{'A', 'B'} {
-				data[pos] = b
-				got := lastIndexByteTwo(data, 'A', 'B')
-				want := refLastIndexByteTwo(data, 'A', 'B')
-				if got != want {
-					t.Fatalf("lastIndexByteTwo(len=%d, match=%c@%d) = %d, want %d", n, b, pos, got, want)
-				}
-				data[pos] = byte('c' + (pos % 20))
-			}
-		}
-		// No match
-		got := lastIndexByteTwo(data, 'A', 'B')
-		if got != -1 {
-			t.Fatalf("lastIndexByteTwo(len=%d, no match) = %d, want -1", n, got)
-		}
-		// Both bytes present
-		if n >= 2 {
-			data[n/3] = 'A'
-			data[n*2/3] = 'B'
-			got := lastIndexByteTwo(data, 'A', 'B')
-			want := refLastIndexByteTwo(data, 'A', 'B')
-			if got != want {
-				t.Fatalf("lastIndexByteTwo(len=%d, both@%d,%d) = %d, want %d", n, n/3, n*2/3, got, want)
-			}
-			data[n/3] = byte('c' + ((n / 3) % 20))
-			data[n*2/3] = byte('c' + ((n * 2 / 3) % 20))
-		}
-	}
-}
-
-func FuzzIndexByteTwo(f *testing.F) {
-	f.Add([]byte("hello world"), byte('o'), byte('l'))
-	f.Add([]byte(""), byte('a'), byte('b'))
-	f.Add([]byte("aaa"), byte('a'), byte('a'))
-	f.Fuzz(func(t *testing.T, data []byte, b1, b2 byte) {
-		got := IndexByteTwo(data, b1, b2)
-		want := loopIndexByteTwo(data, b1, b2)
-		if got != want {
-			t.Errorf("IndexByteTwo(len=%d, b1=%d, b2=%d) = %d, want %d", len(data), b1, b2, got, want)
-		}
-	})
-}
-
-func FuzzLastIndexByteTwo(f *testing.F) {
-	f.Add([]byte("hello world"), byte('o'), byte('l'))
-	f.Add([]byte(""), byte('a'), byte('b'))
-	f.Add([]byte("aaa"), byte('a'), byte('a'))
-	f.Fuzz(func(t *testing.T, data []byte, b1, b2 byte) {
-		got := lastIndexByteTwo(data, b1, b2)
-		want := refLastIndexByteTwo(data, b1, b2)
-		if got != want {
-			t.Errorf("lastIndexByteTwo(len=%d, b1=%d, b2=%d) = %d, want %d", len(data), b1, b2, got, want)
-		}
-	})
-}
-
-// Reference implementations for correctness checking
-func refIndexByteTwo(s []byte, b1, b2 byte) int {
-	i1 := bytes.IndexByte(s, b1)
-	if i1 == 0 {
-		return 0
-	}
-	scope := s
-	if i1 > 0 {
-		scope = s[:i1]
-	}
-	if i2 := bytes.IndexByte(scope, b2); i2 >= 0 {
-		return i2
-	}
-	return i1
-}
-
-func loopIndexByteTwo(s []byte, b1, b2 byte) int {
-	for i, b := range s {
-		if b == b1 || b == b2 {
-			return i
-		}
-	}
-	return -1
-}
-
-func refLastIndexByteTwo(s []byte, b1, b2 byte) int {
-	for i := len(s) - 1; i >= 0; i-- {
-		if s[i] == b1 || s[i] == b2 {
-			return i
-		}
-	}
-	return -1
-}
-
-func benchIndexByteTwo(b *testing.B, size int, pos int) {
-	data := make([]byte, size)
-	for i := range data {
-		data[i] = byte('a' + (i % 20))
-	}
-	data[pos] = 'Z'
-
-	type impl struct {
-		name string
-		fn   func([]byte, byte, byte) int
-	}
-	impls := []impl{
-		{"asm", IndexByteTwo},
-		{"2xIndexByte", refIndexByteTwo},
-		{"loop", loopIndexByteTwo},
-	}
-	for _, im := range impls {
-		b.Run(im.name, func(b *testing.B) {
-			for i := 0; i < b.N; i++ {
-				im.fn(data, 'Z', 'z')
-			}
-		})
-	}
-}
-
-func benchLastIndexByteTwo(b *testing.B, size int, pos int) {
-	data := make([]byte, size)
-	for i := range data {
-		data[i] = byte('a' + (i % 20))
-	}
-	data[pos] = 'Z'
-
-	type impl struct {
-		name string
-		fn   func([]byte, byte, byte) int
-	}
-	impls := []impl{
-		{"asm", lastIndexByteTwo},
-		{"loop", refLastIndexByteTwo},
-	}
-	for _, im := range impls {
-		b.Run(im.name, func(b *testing.B) {
-			for i := 0; i < b.N; i++ {
-				im.fn(data, 'Z', 'z')
-			}
-		})
-	}
-}
-
-func BenchmarkIndexByteTwo_10(b *testing.B)       { benchIndexByteTwo(b, 10, 8) }
-func BenchmarkIndexByteTwo_100(b *testing.B)      { benchIndexByteTwo(b, 100, 80) }
-func BenchmarkIndexByteTwo_1000(b *testing.B)     { benchIndexByteTwo(b, 1000, 800) }
-func BenchmarkLastIndexByteTwo_10(b *testing.B)   { benchLastIndexByteTwo(b, 10, 2) }
-func BenchmarkLastIndexByteTwo_100(b *testing.B)  { benchLastIndexByteTwo(b, 100, 20) }
-func BenchmarkLastIndexByteTwo_1000(b *testing.B) { benchLastIndexByteTwo(b, 1000, 200) }
--- a/src/ansi.go
+++ b/src/ansi.go
@@ -6,7 +6,6 @@ import (
 	"strings"
 	"unicode/utf8"

-	"github.com/junegunn/fzf/src/algo"
 	"github.com/junegunn/fzf/src/tui"
 )

@@ -124,31 +123,31 @@ func toAnsiString(color tui.Color, offset int) string {
 	return ret + ";"
 }

+func isPrint(c uint8) bool {
+	return '\x20' <= c && c <= '\x7e'
+}
+
 func matchOperatingSystemCommand(s string, start int) int {
 	// `\x1b][0-9][;:][[:print:]]+(?:\x1b\\\\|\x07)`
 	//                 ^ match starting here after the first printable character
 	//
 	i := start // prefix matched in nextAnsiEscapeSequence()
-
-	// Find the terminator: BEL (\x07) or ESC (\x1b) for ST (\x1b\\)
-	idx := algo.IndexByteTwo(stringBytes(s[i:]), '\x07', '\x1b')
-	if idx < 0 {
-		return -1
+	for ; i < len(s) && isPrint(s[i]); i++ {
 	}
-	i += idx
-
+	if i < len(s) {
 		if s[i] == '\x07' {
 			return i + 1
 		}
 		// `\x1b]8;PARAMS;URI\x1b\\TITLE\x1b]8;;\x1b`
 		//                   ------
-	if i < len(s)-1 && s[i+1] == '\\' {
+		if s[i] == '\x1b' && i < len(s)-1 && s[i+1] == '\\' {
 			return i + 2
 		}
+	}

 	// `\x1b]8;PARAMS;URI\x1b\\TITLE\x1b]8;;\x1b`
 	//                              ------------
-	if s[:i+1] == "\x1b]8;;\x1b" {
+	if i < len(s) && s[:i+1] == "\x1b]8;;\x1b" {
 		return i + 1
 	}

@@ -234,7 +233,7 @@ Loop:

 				// \x1b][0-9]+[;:][[:print:]]+(?:\x1b\\\\|\x07)
 				//            ---------------
-				if j > 2 && i+j+1 < len(s) && (s[i+j] == ';' || s[i+j] == ':') && s[i+j+1] >= '\x20' {
+				if j > 2 && i+j+1 < len(s) && (s[i+j] == ';' || s[i+j] == ':') && isPrint(s[i+j+1]) {
 					if k := matchOperatingSystemCommand(s[i:], j+2); k != -1 {
 						return i, i + k
 					}
--- a/src/cache.go
+++ b/src/cache.go
@@ -2,13 +2,10 @@ package fzf

 import "sync"

-// ChunkBitmap is a bitmap with one bit per item in a chunk.
-type ChunkBitmap [chunkBitWords]uint64
+// queryCache associates strings to lists of items
+type queryCache map[string][]Result

-// queryCache associates query strings to bitmaps of matching items
-type queryCache map[string]ChunkBitmap
-
-// ChunkCache associates Chunk and query string to bitmaps
+// ChunkCache associates Chunk and query string to lists of items
 type ChunkCache struct {
 	mutex sync.Mutex
 	cache map[*Chunk]*queryCache
@@ -33,9 +30,9 @@ func (cc *ChunkCache) retire(chunk ...*Chunk) {
 	cc.mutex.Unlock()
 }

-// Add stores the bitmap for the given chunk and key
-func (cc *ChunkCache) Add(chunk *Chunk, key string, bitmap ChunkBitmap, matchCount int) {
-	if len(key) == 0 || !chunk.IsFull() || matchCount > queryCacheMax {
+// Add adds the list to the cache
+func (cc *ChunkCache) Add(chunk *Chunk, key string, list []Result) {
+	if len(key) == 0 || !chunk.IsFull() || len(list) > queryCacheMax {
 		return
 	}

@@ -47,11 +44,11 @@ func (cc *ChunkCache) Add(chunk *Chunk, key string, bitmap ChunkBitmap, matchCou
 		cc.cache[chunk] = &queryCache{}
 		qc = cc.cache[chunk]
 	}
-	(*qc)[key] = bitmap
+	(*qc)[key] = list
 }

-// Lookup returns the bitmap for the exact key
-func (cc *ChunkCache) Lookup(chunk *Chunk, key string) *ChunkBitmap {
+// Lookup is called to lookup ChunkCache
+func (cc *ChunkCache) Lookup(chunk *Chunk, key string) []Result {
 	if len(key) == 0 || !chunk.IsFull() {
 		return nil
 	}
@@ -61,15 +58,15 @@ func (cc *ChunkCache) Lookup(chunk *Chunk, key string) *ChunkBitmap {

 	qc, ok := cc.cache[chunk]
 	if ok {
-		if bm, ok := (*qc)[key]; ok {
-			return &bm
+		list, ok := (*qc)[key]
+		if ok {
+			return list
 		}
 	}
 	return nil
 }

-// Search finds the bitmap for the longest prefix or suffix of the key
-func (cc *ChunkCache) Search(chunk *Chunk, key string) *ChunkBitmap {
+func (cc *ChunkCache) Search(chunk *Chunk, key string) []Result {
 	if len(key) == 0 || !chunk.IsFull() {
 		return nil
 	}
@@ -89,8 +86,8 @@ func (cc *ChunkCache) Search(chunk *Chunk, key string) *ChunkBitmap {
 		prefix := key[:len(key)-idx]
 		suffix := key[idx:]
 		for _, substr := range [2]string{prefix, suffix} {
-			if bm, found := (*qc)[substr]; found {
-				return &bm
+			if cached, found := (*qc)[substr]; found {
+				return cached
 			}
 		}
 	}
--- a/src/cache_test.go
+++ b/src/cache_test.go
@@ -6,34 +6,34 @@ func TestChunkCache(t *testing.T) {
 	cache := NewChunkCache()
 	chunk1p := &Chunk{}
 	chunk2p := &Chunk{count: chunkSize}
-	bm1 := ChunkBitmap{1}
-	bm2 := ChunkBitmap{1, 2}
-	cache.Add(chunk1p, "foo", bm1, 1)
-	cache.Add(chunk2p, "foo", bm1, 1)
-	cache.Add(chunk2p, "bar", bm2, 2)
+	items1 := []Result{{}}
+	items2 := []Result{{}, {}}
+	cache.Add(chunk1p, "foo", items1)
+	cache.Add(chunk2p, "foo", items1)
+	cache.Add(chunk2p, "bar", items2)

 	{ // chunk1 is not full
 		cached := cache.Lookup(chunk1p, "foo")
 		if cached != nil {
-			t.Error("Cached disabled for non-full chunks", cached)
+			t.Error("Cached disabled for non-empty chunks", cached)
 		}
 	}
 	{
 		cached := cache.Lookup(chunk2p, "foo")
-		if cached == nil || cached[0] != 1 {
-			t.Error("Expected bitmap cached", cached)
+		if cached == nil || len(cached) != 1 {
+			t.Error("Expected 1 item cached", cached)
 		}
 	}
 	{
 		cached := cache.Lookup(chunk2p, "bar")
-		if cached == nil || cached[1] != 2 {
-			t.Error("Expected bitmap cached", cached)
+		if cached == nil || len(cached) != 2 {
+			t.Error("Expected 2 items cached", cached)
 		}
 	}
 	{
 		cached := cache.Lookup(chunk1p, "foobar")
 		if cached != nil {
-			t.Error("Expected nil cached", cached)
+			t.Error("Expected 0 item cached", cached)
 		}
 	}
 }
--- a/src/chunklist.go
+++ b/src/chunklist.go
@@ -99,21 +99,6 @@ func (cl *ChunkList) Clear() {
 	cl.mutex.Unlock()
 }

-// ForEachItem iterates all items and applies fn to each one.
-// The done callback runs under the lock to safely update shared state.
-func (cl *ChunkList) ForEachItem(fn func(*Item), done func()) {
-	cl.mutex.Lock()
-	for _, chunk := range cl.chunks {
-		for i := 0; i < chunk.count; i++ {
-			fn(&chunk.items[i])
-		}
-	}
-	if done != nil {
-		done()
-	}
-	cl.mutex.Unlock()
-}
-
 // Snapshot returns immutable snapshot of the ChunkList
 func (cl *ChunkList) Snapshot(tail int) ([]*Chunk, int, bool) {
 	cl.mutex.Lock()
--- a/src/constants.go
+++ b/src/constants.go
@@ -34,18 +34,19 @@ const (
 	maxBgProcessesPerAction = 3

 	// Matcher
+	numPartitionsMultiplier = 8
+	maxPartitions           = 32
 	progressMinDuration     = 200 * time.Millisecond

 	// Capacity of each chunk
-	chunkSize     int = 1024
-	chunkBitWords     = (chunkSize + 63) / 64
+	chunkSize int = 1000

 	// Pre-allocated memory slices to minimize GC
 	slab16Size int = 100 * 1024 // 200KB * 32 = 12.8MB
 	slab32Size int = 2048       // 8KB * 32 = 256KB

 	// Do not cache results of low selectivity queries
-	queryCacheMax int = chunkSize / 2
+	queryCacheMax int = chunkSize / 5

 	// Not to cache mergers with large lists
 	mergerCacheMax int = 100000
--- a/src/core.go
+++ b/src/core.go
@@ -113,9 +113,16 @@ func Run(opts *Options) (int, error) {
 	cache := NewChunkCache()
 	var chunkList *ChunkList
 	var itemIndex int32
-	// transformItem applies with-nth transformation to an item's raw data.
-	// It handles ANSI token propagation using prevLineAnsiState for cross-line continuity.
-	transformItem := func(item *Item, data []byte, transformer func([]Token, int32) string, index int32) {
+	if opts.WithNth == nil {
+		chunkList = NewChunkList(cache, func(item *Item, data []byte) bool {
+			item.text, item.colors = ansiProcessor(data)
+			item.text.Index = itemIndex
+			itemIndex++
+			return true
+		})
+	} else {
+		nthTransformer := opts.WithNth(opts.Delimiter)
+		chunkList = NewChunkList(cache, func(item *Item, data []byte) bool {
 			tokens := Tokenize(byteString(data), opts.Delimiter)
 			if opts.Ansi && len(tokens) > 1 {
 				var ansiState *ansiState
@@ -133,7 +140,7 @@ func Run(opts *Options) (int, error) {
 					}
 				}
 			}
-		transformed := transformer(tokens, index)
+			transformed := nthTransformer(tokens, itemIndex)
 			item.text, item.colors = ansiProcessor(stringBytes(transformed))

 			// We should not trim trailing whitespaces with background colors
@@ -146,24 +153,6 @@ func Run(opts *Options) (int, error) {
 				}
 			}
 			item.text.TrimTrailingWhitespaces(int(maxColorOffset))
-	}
-
-	var nthTransformer func([]Token, int32) string
-	if opts.WithNth == nil {
-		chunkList = NewChunkList(cache, func(item *Item, data []byte) bool {
-			item.text, item.colors = ansiProcessor(data)
-			item.text.Index = itemIndex
-			itemIndex++
-			return true
-		})
-	} else {
-		nthTransformer = opts.WithNth(opts.Delimiter)
-		chunkList = NewChunkList(cache, func(item *Item, data []byte) bool {
-			if nthTransformer == nil {
-				item.text, item.colors = ansiProcessor(data)
-			} else {
-				transformItem(item, data, nthTransformer, itemIndex)
-			}
 			item.text.Index = itemIndex
 			item.origText = &data
 			itemIndex++
@@ -195,13 +184,11 @@ func Run(opts *Options) (int, error) {
 	// Reader
 	streamingFilter := opts.Filter != nil && !sort && !opts.Tac && !opts.Sync && opts.Bench == 0
 	var reader *Reader
-	var ingestionStart time.Time
 	if !streamingFilter {
 		reader = NewReader(func(data []byte) bool {
 			return chunkList.Push(data)
 		}, eventBox, executor, opts.ReadZero, opts.Filter == nil)

-		ingestionStart = time.Now()
 		readyChan := make(chan bool)
 		go reader.ReadSource(opts.Input, opts.WalkerRoot, opts.WalkerOpts, opts.WalkerSkip, initialReload, initialEnv, readyChan)
 		<-readyChan
@@ -285,7 +272,6 @@ func Run(opts *Options) (int, error) {
 		} else {
 			eventBox.Unwatch(EvtReadNew)
 			eventBox.WaitFor(EvtReadFin)
-			ingestionTime := time.Since(ingestionStart)

 			// NOTE: Streaming filter is inherently not compatible with --tail
 			snapshot, _, _ := chunkList.Snapshot(opts.Tail)
@@ -319,14 +305,13 @@ func Run(opts *Options) (int, error) {
 				}
 				avg := total / time.Duration(len(times))
 				selectivity := float64(matchCount) / float64(totalItems) * 100
-				fmt.Printf("  %d iterations  avg: %.2fms  min: %.2fms  max: %.2fms  total: %.2fs  items: %d  matches: %d (%.2f%%)  ingestion: %.2fms\n",
+				fmt.Printf("  %d iterations  avg: %.2fms  min: %.2fms  max: %.2fms  total: %.2fs  items: %d  matches: %d (%.2f%%)\n",
 					len(times),
 					float64(avg.Microseconds())/1000,
 					float64(minD.Microseconds())/1000,
 					float64(maxD.Microseconds())/1000,
 					total.Seconds(),
-					totalItems, matchCount, selectivity,
-					float64(ingestionTime.Microseconds())/1000)
+					totalItems, matchCount, selectivity)
 				return ExitOk, nil
 			}

@@ -476,7 +461,6 @@ func Run(opts *Options) (int, error) {
 					var environ []string
 					var changed bool
 					headerLinesChanged := false
-					withNthChanged := false
 					switch val := value.(type) {
 					case searchRequest:
 						sort = val.sort
@@ -503,34 +487,6 @@ func Run(opts *Options) (int, error) {
 							headerLinesChanged = true
 							bump = true
 						}
-						if val.withNth != nil {
-							newTransformer := val.withNth.fn
-							// Cancel any in-flight scan and block the terminal from reading
-							// items before mutating them in-place. Snapshot shares middle
-							// chunk pointers, so the matcher and terminal can race with us.
-							matcher.CancelScan()
-							terminal.PauseRendering()
-							// Reset cross-line ANSI state before re-processing all items
-							lineAnsiState = nil
-							prevLineAnsiState = nil
-							chunkList.ForEachItem(func(item *Item) {
-								origBytes := *item.origText
-								savedIndex := item.Index()
-								if newTransformer != nil {
-									transformItem(item, origBytes, newTransformer, savedIndex)
-								} else {
-									item.text, item.colors = ansiProcessor(origBytes)
-								}
-								item.text.Index = savedIndex
-								item.transformed = nil
-							}, func() {
-								nthTransformer = newTransformer
-							})
-							terminal.ResumeRendering()
-							matcher.ResumeScan()
-							withNthChanged = true
-							bump = true
-						}
 						if bump {
 							patternCache = make(map[string]*Pattern)
 							cache.Clear()
@@ -574,8 +530,6 @@ func Run(opts *Options) (int, error) {
 						} else {
 							terminal.UpdateHeader(nil)
 						}
-					} else if withNthChanged && headerLines > 0 {
-						terminal.UpdateHeader(GetItems(snapshot, int(headerLines)))
 					}
 					matcher.Reset(snapshot, input(), true, !reading, sort, snapshotRevision)
 					delay = false
--- a/src/matcher.go
+++ b/src/matcher.go
@@ -4,7 +4,6 @@ import (
 	"fmt"
 	"runtime"
 	"sync"
-	"sync/atomic"
 	"time"

 	"github.com/junegunn/fzf/src/util"
@@ -46,8 +45,6 @@ type Matcher struct {
 	sortBuf        [][]Result
 	mergerCache    map[string]MatchResult
 	revision       revision
-	scanMutex      sync.Mutex
-	cancelScan     *util.AtomicBool
 }

 const (
@@ -58,7 +55,7 @@ const (
 // NewMatcher returns a new Matcher
 func NewMatcher(cache *ChunkCache, patternBuilder func([]rune) *Pattern,
 	sort bool, tac bool, eventBox *util.EventBox, revision revision, threads int) *Matcher {
-	partitions := runtime.NumCPU()
+	partitions := min(numPartitionsMultiplier*runtime.NumCPU(), maxPartitions)
 	if threads > 0 {
 		partitions = threads
 	}
@@ -73,8 +70,7 @@ func NewMatcher(cache *ChunkCache, patternBuilder func([]rune) *Pattern,
 		slab:           make([]*util.Slab, partitions),
 		sortBuf:        make([][]Result, partitions),
 		mergerCache:    make(map[string]MatchResult),
-		revision:       revision,
-		cancelScan:     util.NewAtomicBool(false)}
+		revision:       revision}
 }

 // Loop puts Matcher in action
@@ -134,9 +130,7 @@ func (m *Matcher) Loop() {
 		}

 		if result.merger == nil {
-			m.scanMutex.Lock()
 			result = m.scan(request)
-			m.scanMutex.Unlock()
 		}

 		if !result.cancelled {
@@ -149,6 +143,27 @@ func (m *Matcher) Loop() {
 	}
 }

+func (m *Matcher) sliceChunks(chunks []*Chunk) [][]*Chunk {
+	partitions := m.partitions
+	perSlice := len(chunks) / partitions
+
+	if perSlice == 0 {
+		partitions = len(chunks)
+		perSlice = 1
+	}
+
+	slices := make([][]*Chunk, partitions)
+	for i := 0; i < partitions; i++ {
+		start := i * perSlice
+		end := start + perSlice
+		if i == partitions-1 {
+			end = len(chunks)
+		}
+		slices[i] = chunks[start:end]
+	}
+	return slices
+}
+
 type partialResult struct {
 	index   int
 	matches []Result
@@ -172,37 +187,39 @@ func (m *Matcher) scan(request MatchRequest) MatchResult {
 	maxIndex := request.chunks[numChunks-1].lastIndex(minIndex)
 	cancelled := util.NewAtomicBool(false)

-	numWorkers := min(m.partitions, numChunks)
-	var nextChunk atomic.Int32
-	resultChan := make(chan partialResult, numWorkers)
+	slices := m.sliceChunks(request.chunks)
+	numSlices := len(slices)
+	resultChan := make(chan partialResult, numSlices)
 	countChan := make(chan int, numChunks)
 	waitGroup := sync.WaitGroup{}

-	for idx := range numWorkers {
+	for idx, chunks := range slices {
 		waitGroup.Add(1)
 		if m.slab[idx] == nil {
 			m.slab[idx] = util.MakeSlab(slab16Size, slab32Size)
 		}
-		go func(idx int, slab *util.Slab) {
-			defer waitGroup.Done()
-			var matches []Result
-			for {
-				ci := int(nextChunk.Add(1)) - 1
-				if ci >= numChunks {
-					break
-				}
-				chunkMatches := request.pattern.Match(request.chunks[ci], slab)
-				matches = append(matches, chunkMatches...)
+		go func(idx int, slab *util.Slab, chunks []*Chunk) {
+			defer func() { waitGroup.Done() }()
+			count := 0
+			allMatches := make([][]Result, len(chunks))
+			for idx, chunk := range chunks {
+				matches := request.pattern.Match(chunk, slab)
+				allMatches[idx] = matches
+				count += len(matches)
 				if cancelled.Get() {
 					return
 				}
-				countChan <- len(chunkMatches)
+				countChan <- len(matches)
+			}
+			sliceMatches := make([]Result, 0, count)
+			for _, matches := range allMatches {
+				sliceMatches = append(sliceMatches, matches...)
 			}
 			if m.sort && request.pattern.sortable {
-				m.sortBuf[idx] = radixSortResults(matches, m.tac, m.sortBuf[idx])
+				m.sortBuf[idx] = radixSortResults(sliceMatches, m.tac, m.sortBuf[idx])
 			}
-			resultChan <- partialResult{idx, matches}
-		}(idx, m.slab[idx])
+			resultChan <- partialResult{idx, sliceMatches}
+		}(idx, m.slab[idx], chunks)
 	}

 	wait := func() bool {
@@ -221,7 +238,7 @@ func (m *Matcher) scan(request MatchRequest) MatchResult {
 			break
 		}

-		if m.cancelScan.Get() || m.reqBox.Peek(reqReset) {
+		if m.reqBox.Peek(reqReset) {
 			return MatchResult{nil, nil, wait()}
 		}

@@ -230,8 +247,8 @@ func (m *Matcher) scan(request MatchRequest) MatchResult {
 		}
 	}

-	partialResults := make([][]Result, numWorkers)
-	for range numWorkers {
+	partialResults := make([][]Result, numSlices)
+	for range slices {
 		partialResult := <-resultChan
 		partialResults[partialResult.index] = partialResult.matches
 	}
@@ -252,20 +269,6 @@ func (m *Matcher) Reset(chunks []*Chunk, patternRunes []rune, cancel bool, final
 	m.reqBox.Set(event, MatchRequest{chunks, pattern, final, sort, revision})
 }

-// CancelScan cancels any in-flight scan, waits for it to finish,
-// and prevents new scans from starting until ResumeScan is called.
-// This is used to safely mutate shared items (e.g., during with-nth changes).
-func (m *Matcher) CancelScan() {
-	m.cancelScan.Set(true)
-	m.scanMutex.Lock()
-	m.cancelScan.Set(false)
-}
-
-// ResumeScan allows scans to proceed again after CancelScan.
-func (m *Matcher) ResumeScan() {
-	m.scanMutex.Unlock()
-}
-
 func (m *Matcher) Stop() {
 	m.reqBox.Set(reqQuit, nil)
 }
--- a/src/options.go
+++ b/src/options.go
@@ -588,7 +588,6 @@ type Options struct {
 	FreezeLeft        int
 	FreezeRight       int
 	WithNth           func(Delimiter) func([]Token, int32) string
-	WithNthExpr       string
 	AcceptNth         func(Delimiter) func([]Token, int32) string
 	Delimiter         Delimiter
 	Sort              int
@@ -1630,7 +1629,7 @@ const (

 func init() {
 	executeRegexp = regexp.MustCompile(
-		`(?si)[:+](become|execute(?:-multi|-silent)?|reload(?:-sync)?|preview|(?:change|bg-transform|transform)-(?:query|prompt|(?:border|list|preview|input|header|footer)-label|header-lines|header|footer|search|with-nth|nth|pointer|ghost)|bg-transform|transform|change-(?:preview-window|preview|multi)|(?:re|un|toggle-)bind|pos|put|print|search|trigger)`)
+		`(?si)[:+](become|execute(?:-multi|-silent)?|reload(?:-sync)?|preview|(?:change|bg-transform|transform)-(?:query|prompt|(?:border|list|preview|input|header|footer)-label|header-lines|header|footer|search|nth|pointer|ghost)|bg-transform|transform|change-(?:preview-window|preview|multi)|(?:re|un|toggle-)bind|pos|put|print|search|trigger)`)
 	splitRegexp = regexp.MustCompile("[,:]+")
 	actionNameRegexp = regexp.MustCompile("(?i)^[a-z-]+")
 }
@@ -2073,8 +2072,6 @@ func isExecuteAction(str string) actionType {
 		return actChangeMulti
 	case "change-nth":
 		return actChangeNth
-	case "change-with-nth":
-		return actChangeWithNth
 	case "pos":
 		return actPosition
 	case "execute":
@@ -2111,8 +2108,6 @@ func isExecuteAction(str string) actionType {
 		return actTransformGhost
 	case "transform-nth":
 		return actTransformNth
-	case "transform-with-nth":
-		return actTransformWithNth
 	case "transform-pointer":
 		return actTransformPointer
 	case "transform-prompt":
@@ -2145,8 +2140,6 @@ func isExecuteAction(str string) actionType {
 		return actBgTransformGhost
 	case "bg-transform-nth":
 		return actBgTransformNth
-	case "bg-transform-with-nth":
-		return actBgTransformWithNth
 	case "bg-transform-pointer":
 		return actBgTransformPointer
 	case "bg-transform-prompt":
@@ -2788,7 +2781,6 @@ func parseOptions(index *int, opts *Options, allArgs []string) error {
 			if opts.WithNth, err = nthTransformer(str); err != nil {
 				return err
 			}
-			opts.WithNthExpr = str
 		case "--accept-nth":
 			str, err := nextString("nth expression required")
 			if err != nil {
--- a/src/pattern.go
+++ b/src/pattern.go
@@ -61,7 +61,7 @@ type Pattern struct {
 	delimiter     Delimiter
 	nth           []Range
 	revision      revision
-	procFun       [6]algo.Algo
+	procFun       map[termType]algo.Algo
 	cache         *ChunkCache
 	denylist      map[int32]struct{}
 	startIndex    int32
@@ -150,7 +150,7 @@ func BuildPattern(cache *ChunkCache, patternCache map[string]*Pattern, fuzzy boo
 		cache:         cache,
 		denylist:      denylist,
 		startIndex:    startIndex,
-	}
+		procFun:       make(map[termType]algo.Algo)}

 	ptr.cacheKey = ptr.buildCacheKey()
 	ptr.directAlgo, ptr.directTerm = ptr.buildDirectAlgo(fuzzyAlgo)
@@ -300,87 +300,104 @@ func (p *Pattern) CacheKey() string {

 // Match returns the list of matches Items in the given Chunk
 func (p *Pattern) Match(chunk *Chunk, slab *util.Slab) []Result {
+	// ChunkCache: Exact match
 	cacheKey := p.CacheKey()
-
-	// Bitmap cache: exact match or prefix/suffix
-	var cachedBitmap *ChunkBitmap
 	if p.cacheable {
-		cachedBitmap = p.cache.Lookup(chunk, cacheKey)
+		if cached := p.cache.Lookup(chunk, cacheKey); cached != nil {
+			return cached
 		}
-	if cachedBitmap == nil {
-		cachedBitmap = p.cache.Search(chunk, cacheKey)
 	}

-	matches, bitmap := p.matchChunk(chunk, cachedBitmap, slab)
+	// Prefix/suffix cache
+	space := p.cache.Search(chunk, cacheKey)
+
+	matches := p.matchChunk(chunk, space, slab)

 	if p.cacheable {
-		p.cache.Add(chunk, cacheKey, bitmap, len(matches))
+		p.cache.Add(chunk, cacheKey, matches)
 	}
 	return matches
 }

-func (p *Pattern) matchChunk(chunk *Chunk, cachedBitmap *ChunkBitmap, slab *util.Slab) ([]Result, ChunkBitmap) {
+func (p *Pattern) matchChunk(chunk *Chunk, space []Result, slab *util.Slab) []Result {
 	matches := []Result{}
-	var bitmap ChunkBitmap

 	// Skip header items in chunks that contain them
 	startIdx := 0
 	if p.startIndex > 0 && chunk.count > 0 && chunk.items[0].Index() < p.startIndex {
 		startIdx = int(p.startIndex - chunk.items[0].Index())
 		if startIdx >= chunk.count {
-			return matches, bitmap
+			return matches
 		}
 	}

-	hasCachedBitmap := cachedBitmap != nil
-
 	// Fast path: single fuzzy term, no nth, no denylist.
 	// Calls the algo function directly, bypassing MatchItem/extendedMatch/iter
 	// and avoiding per-match []Offset heap allocation.
 	if p.directAlgo != nil && len(p.denylist) == 0 {
 		t := p.directTerm
+		if space == nil {
 			for idx := startIdx; idx < chunk.count; idx++ {
-			if hasCachedBitmap && cachedBitmap[idx/64]&(uint64(1)<<(idx%64)) == 0 {
-				continue
-			}
 				res, _ := p.directAlgo(t.caseSensitive, t.normalize, p.forward,
 					&chunk.items[idx].text, t.text, p.withPos, slab)
 				if res.Start >= 0 {
-				bitmap[idx/64] |= uint64(1) << (idx % 64)
 					matches = append(matches, buildResultFromBounds(
 						&chunk.items[idx], res.Score,
 						int(res.Start), int(res.End), int(res.End), true))
 				}
 			}
-		return matches, bitmap
+		} else {
+			for _, result := range space {
+				res, _ := p.directAlgo(t.caseSensitive, t.normalize, p.forward,
+					&result.item.text, t.text, p.withPos, slab)
+				if res.Start >= 0 {
+					matches = append(matches, buildResultFromBounds(
+						result.item, res.Score,
+						int(res.Start), int(res.End), int(res.End), true))
+				}
+			}
+		}
+		return matches
 	}

 	if len(p.denylist) == 0 {
+		// Huge code duplication for minimizing unnecessary map lookups
+		if space == nil {
 			for idx := startIdx; idx < chunk.count; idx++ {
-			if hasCachedBitmap && cachedBitmap[idx/64]&(uint64(1)<<(idx%64)) == 0 {
-				continue
-			}
 				if match, _, _ := p.MatchItem(&chunk.items[idx], p.withPos, slab); match.item != nil {
-				bitmap[idx/64] |= uint64(1) << (idx % 64)
 					matches = append(matches, match)
 				}
 			}
-		return matches, bitmap
+		} else {
+			for _, result := range space {
+				if match, _, _ := p.MatchItem(result.item, p.withPos, slab); match.item != nil {
+					matches = append(matches, match)
+				}
+			}
+		}
+		return matches
 	}

+	if space == nil {
 		for idx := startIdx; idx < chunk.count; idx++ {
-		if hasCachedBitmap && cachedBitmap[idx/64]&(uint64(1)<<(idx%64)) == 0 {
-			continue
-		}
 			if _, prs := p.denylist[chunk.items[idx].Index()]; prs {
 				continue
 			}
 			if match, _, _ := p.MatchItem(&chunk.items[idx], p.withPos, slab); match.item != nil {
-			bitmap[idx/64] |= uint64(1) << (idx % 64)
 				matches = append(matches, match)
 			}
 		}
-	return matches, bitmap
+	} else {
+		for _, result := range space {
+			if _, prs := p.denylist[result.item.Index()]; prs {
+				continue
+			}
+			if match, _, _ := p.MatchItem(result.item, p.withPos, slab); match.item != nil {
+				matches = append(matches, match)
+			}
+		}
+	}
+	return matches
 }

 // MatchItem returns the match result if the Item is a match.
--- a/src/pattern_test.go
+++ b/src/pattern_test.go
@@ -2,7 +2,6 @@ package fzf

 import (
 	"reflect"
-	"runtime"
 	"testing"

 	"github.com/junegunn/fzf/src/algo"
@@ -138,7 +137,7 @@ func TestOrigTextAndTransformed(t *testing.T) {
 			origText:    &origBytes,
 			transformed: &transformed{pattern.revision, trans}}
 		pattern.extended = extended
-		matches, _ := pattern.matchChunk(&chunk, nil, slab) // No cache
+		matches := pattern.matchChunk(&chunk, nil, slab) // No cache
 		if !(matches[0].item.text.ToString() == "junegunn" &&
 			string(*matches[0].item.origText) == "junegunn.choi" &&
 			reflect.DeepEqual((*matches[0].item.transformed).tokens, trans)) {
@@ -200,119 +199,3 @@ func TestCacheable(t *testing.T) {
 	test(false, "foo 'bar", "foo", false)
 	test(false, "foo !bar", "foo", false)
 }
-
-func buildChunks(numChunks int) []*Chunk {
-	chunks := make([]*Chunk, numChunks)
-	words := []string{
-		"src/main/java/com/example/service/UserService.java",
-		"src/test/java/com/example/service/UserServiceTest.java",
-		"docs/api/reference/endpoints.md",
-		"lib/internal/utils/string_helper.go",
-		"pkg/server/http/handler/auth.go",
-		"build/output/release/app.exe",
-		"config/production/database.yml",
-		"scripts/deploy/kubernetes/setup.sh",
-		"vendor/github.com/junegunn/fzf/src/core.go",
-		"node_modules/.cache/babel/transform.js",
-	}
-	for ci := range numChunks {
-		chunks[ci] = &Chunk{count: chunkSize}
-		for i := range chunkSize {
-			text := words[(ci*chunkSize+i)%len(words)]
-			chunks[ci].items[i] = Item{text: util.ToChars([]byte(text))}
-			chunks[ci].items[i].text.Index = int32(ci*chunkSize + i)
-		}
-	}
-	return chunks
-}
-
-func buildPatternWith(cache *ChunkCache, runes []rune) *Pattern {
-	return BuildPattern(cache, make(map[string]*Pattern),
-		true, algo.FuzzyMatchV2, true, CaseSmart, false, true,
-		false, true, []Range{}, Delimiter{}, revision{}, runes, nil, 0)
-}
-
-func TestBitmapCacheBenefit(t *testing.T) {
-	numChunks := 100
-	chunks := buildChunks(numChunks)
-	queries := []string{"s", "se", "ser", "serv", "servi"}
-
-	// 1. Run all queries with shared cache (simulates incremental typing)
-	cache := NewChunkCache()
-	for _, q := range queries {
-		pat := buildPatternWith(cache, []rune(q))
-		for _, chunk := range chunks {
-			pat.Match(chunk, slab)
-		}
-	}
-
-	// 2. GC and measure memory with cache populated
-	runtime.GC()
-	runtime.GC()
-	var memWith runtime.MemStats
-	runtime.ReadMemStats(&memWith)
-
-	// 3. Clear cache, GC, measure again
-	cache.Clear()
-	runtime.GC()
-	runtime.GC()
-	var memWithout runtime.MemStats
-	runtime.ReadMemStats(&memWithout)
-
-	cacheMem := int64(memWith.Alloc) - int64(memWithout.Alloc)
-	t.Logf("Chunks: %d, Queries: %d", numChunks, len(queries))
-	t.Logf("Cache memory: %d bytes (%.1f KB)", cacheMem, float64(cacheMem)/1024)
-	t.Logf("Per-chunk-per-query: %.0f bytes", float64(cacheMem)/float64(numChunks*len(queries)))
-
-	// 4. Verify correctness: cached vs uncached produce same results
-	cache2 := NewChunkCache()
-	for _, q := range queries {
-		pat := buildPatternWith(cache2, []rune(q))
-		for _, chunk := range chunks {
-			pat.Match(chunk, slab)
-		}
-	}
-	for _, q := range queries {
-		patCached := buildPatternWith(cache2, []rune(q))
-		patFresh := buildPatternWith(NewChunkCache(), []rune(q))
-		var countCached, countFresh int
-		for _, chunk := range chunks {
-			countCached += len(patCached.Match(chunk, slab))
-			countFresh += len(patFresh.Match(chunk, slab))
-		}
-		if countCached != countFresh {
-			t.Errorf("query=%q: cached=%d, fresh=%d", q, countCached, countFresh)
-		}
-		t.Logf("query=%q: matches=%d", q, countCached)
-	}
-}
-
-func BenchmarkWithCache(b *testing.B) {
-	numChunks := 100
-	chunks := buildChunks(numChunks)
-	queries := []string{"s", "se", "ser", "serv", "servi"}
-
-	b.Run("cached", func(b *testing.B) {
-		for range b.N {
-			cache := NewChunkCache()
-			for _, q := range queries {
-				pat := buildPatternWith(cache, []rune(q))
-				for _, chunk := range chunks {
-					pat.Match(chunk, slab)
-				}
-			}
-		}
-	})
-
-	b.Run("uncached", func(b *testing.B) {
-		for range b.N {
-			for _, q := range queries {
-				cache := NewChunkCache()
-				pat := buildPatternWith(cache, []rune(q))
-				for _, chunk := range chunks {
-					pat.Match(chunk, slab)
-				}
-			}
-		}
-	})
-}
--- a/src/protector/protector_openbsd.go
+++ b/src/protector/protector_openbsd.go
@@ -6,5 +6,5 @@ import "golang.org/x/sys/unix"

 // Protect calls OS specific protections like pledge on OpenBSD
 func Protect() {
-	unix.PledgePromises("stdio cpath dpath wpath rpath tty proc exec inet")
+	unix.PledgePromises("stdio dpath wpath rpath tty proc exec inet tmppath")
 }
--- a/src/result.go
+++ b/src/result.go
@@ -128,7 +128,7 @@ func minRank() Result {
 	return Result{item: &minItem, points: [4]uint16{math.MaxUint16, 0, 0, 0}}
 }

-func (result *Result) colorOffsets(matchOffsets []Offset, nthOffsets []Offset, theme *tui.ColorTheme, colBase tui.ColorPair, colMatch tui.ColorPair, attrNth tui.Attr, nthOverlay tui.Attr, hidden bool) []colorOffset {
+func (result *Result) colorOffsets(matchOffsets []Offset, nthOffsets []Offset, theme *tui.ColorTheme, colBase tui.ColorPair, colMatch tui.ColorPair, attrNth tui.Attr, hidden bool) []colorOffset {
 	itemColors := result.item.Colors()

 	// No ANSI codes
@@ -213,10 +213,6 @@ func (result *Result) colorOffsets(matchOffsets []Offset, nthOffsets []Offset, t
 		}
 		return tui.NewColorPair(fg, bg, ansi.color.attr).WithUl(ansi.color.ul).MergeAttr(base)
 	}
-	fgAttr := tui.ColNormal.Attr()
-	nthAttrFinal := fgAttr.Merge(attrNth).Merge(nthOverlay)
-	nthBase := colBase.WithNewAttr(nthAttrFinal)
-
 	var colors []colorOffset
 	add := func(idx int) {
 		if curr.fbg >= 0 {
@@ -230,7 +226,7 @@ func (result *Result) colorOffsets(matchOffsets []Offset, nthOffsets []Offset, t
 			if curr.match {
 				var color tui.ColorPair
 				if curr.nth {
-					color = nthBase.Merge(colMatch)
+					color = colBase.WithAttr(attrNth).Merge(colMatch)
 				} else {
 					color = colBase.Merge(colMatch)
 				}
@@ -250,7 +246,7 @@ func (result *Result) colorOffsets(matchOffsets []Offset, nthOffsets []Offset, t
 					if color.Fg().IsDefault() && origColor.HasBg() {
 						color = origColor
 						if curr.nth {
-							color = color.WithAttr((attrNth &^ tui.AttrRegular).Merge(nthOverlay))
+							color = color.WithAttr(attrNth &^ tui.AttrRegular)
 						}
 					} else {
 						color = origColor.MergeNonDefault(color)
@@ -262,7 +258,7 @@ func (result *Result) colorOffsets(matchOffsets []Offset, nthOffsets []Offset, t
 				ansi := itemColors[curr.index]
 				base := colBase
 				if curr.nth {
-					base = nthBase
+					base = base.WithAttr(attrNth)
 				}
 				if hidden {
 					base = base.WithFg(theme.Nomatch)
@@ -274,7 +270,7 @@ func (result *Result) colorOffsets(matchOffsets []Offset, nthOffsets []Offset, t
 					match:  false,
 					url:    ansi.color.url})
 			} else {
-				color := nthBase
+				color := colBase.WithAttr(attrNth)
 				if hidden {
 					color = color.WithFg(theme.Nomatch)
 				}
--- a/src/result_test.go
+++ b/src/result_test.go
@@ -132,7 +132,7 @@ func TestColorOffset(t *testing.T) {

 	colBase := tui.NewColorPair(89, 189, tui.AttrUndefined)
 	colMatch := tui.NewColorPair(99, 199, tui.AttrUndefined)
-	colors := item.colorOffsets(offsets, nil, tui.Dark256, colBase, colMatch, tui.AttrUndefined, 0, false)
+	colors := item.colorOffsets(offsets, nil, tui.Dark256, colBase, colMatch, tui.AttrUndefined, false)
 	assert := func(idx int, b int32, e int32, c tui.ColorPair) {
 		o := colors[idx]
 		if o.offset[0] != b || o.offset[1] != e || o.color != c {
@@ -159,7 +159,7 @@ func TestColorOffset(t *testing.T) {

 	nthOffsets := []Offset{{37, 39}, {42, 45}}
 	for _, attr := range []tui.Attr{tui.AttrRegular, tui.StrikeThrough} {
-		colors = item.colorOffsets(offsets, nthOffsets, tui.Dark256, colRegular, colUnderline, attr, 0, false)
+		colors = item.colorOffsets(offsets, nthOffsets, tui.Dark256, colRegular, colUnderline, attr, false)

 		// [{[0 5] {1 5 0}} {[5 15] {1 5 8}} {[15 20] {1 5 0}}
 		//  {[22 25] {2 6 1}} {[25 27] {2 6 9}} {[27 30] {-1 -1 8}}
@@ -182,37 +182,6 @@ func TestColorOffset(t *testing.T) {
 		assert(10, 37, 39, tui.NewColorPair(4, 8, expected))
 		assert(11, 39, 40, tui.NewColorPair(4, 8, tui.Bold))
 	}
-
-	// Test nthOverlay: simulates nth:regular with current-fg:underline
-	// The overlay (underline) should survive even though nth:regular clears attrs.
-	// Precedence: fg < nth < current-fg
-	colors = item.colorOffsets(offsets, nthOffsets, tui.Dark256, colRegular, colUnderline, tui.AttrRegular, tui.Underline, false)
-
-	// nth regions should have Underline (from overlay), not cleared by AttrRegular
-	// Non-nth regions keep colBase attrs (AttrUndefined)
-	assert(0, 0, 5, tui.NewColorPair(1, 5, tui.AttrUndefined))
-	assert(1, 5, 15, tui.NewColorPair(1, 5, tui.Underline))
-	assert(2, 15, 20, tui.NewColorPair(1, 5, tui.AttrUndefined))
-	assert(3, 22, 25, tui.NewColorPair(2, 6, tui.Bold))
-	assert(4, 25, 27, tui.NewColorPair(2, 6, tui.Bold|tui.Underline))
-	assert(5, 27, 30, colUnderline)
-	assert(6, 30, 32, tui.NewColorPair(3, 7, tui.Underline))
-	assert(7, 32, 33, colUnderline)
-	assert(8, 33, 35, tui.NewColorPair(4, 8, tui.Bold|tui.Underline))
-	assert(9, 35, 37, tui.NewColorPair(4, 8, tui.Bold))
-	// nth region within ANSI bold: AttrRegular clears, overlay adds Underline back
-	assert(10, 37, 39, tui.NewColorPair(4, 8, tui.Bold|tui.Underline))
-	assert(11, 39, 40, tui.NewColorPair(4, 8, tui.Bold))
-
-	// Test nthOverlay with additive attrs: nth:strikethrough with selected-fg:bold
-	colors = item.colorOffsets(offsets, nthOffsets, tui.Dark256, colRegular, colUnderline, tui.StrikeThrough, tui.Bold, false)
-
-	// Non-nth entries unchanged from overlay=0 case
-	assert(0, 0, 5, tui.NewColorPair(1, 5, tui.AttrUndefined))
-	assert(5, 27, 30, colUnderline) // match only, no nth
-	assert(7, 32, 33, colUnderline) // match only, no nth
-	// nth region within ANSI bold: StrikeThrough|Bold merged with ANSI Bold
-	assert(10, 37, 39, tui.NewColorPair(4, 8, tui.Bold|tui.StrikeThrough))
 }

 func TestRadixSortResults(t *testing.T) {
--- a/src/terminal.go
+++ b/src/terminal.go
@@ -340,9 +340,6 @@ type Terminal struct {
 	nthAttr              tui.Attr
 	nth                  []Range
 	nthCurrent           []Range
-	withNthDefault       string
-	withNthExpr          string
-	withNthEnabled       bool
 	acceptNth            func([]Token, int32) string
 	tabstop              int
 	margin               [4]sizeSpec
@@ -387,7 +384,6 @@ type Terminal struct {
 	hasLoadActions       bool
 	hasResizeActions     bool
 	triggerLoad          bool
-	filterSelection      bool
 	reading              bool
 	running              *util.AtomicBool
 	failed               *string
@@ -555,7 +551,6 @@ const (
 	actChangeListLabel
 	actChangeMulti
 	actChangeNth
-	actChangeWithNth
 	actChangePointer
 	actChangePreview
 	actChangePreviewLabel
@@ -641,7 +636,6 @@ const (
 	actTransformInputLabel
 	actTransformListLabel
 	actTransformNth
-	actTransformWithNth
 	actTransformPointer
 	actTransformPreviewLabel
 	actTransformPrompt
@@ -661,7 +655,6 @@ const (
 	actBgTransformInputLabel
 	actBgTransformListLabel
 	actBgTransformNth
-	actBgTransformWithNth
 	actBgTransformPointer
 	actBgTransformPreviewLabel
 	actBgTransformPrompt
@@ -728,7 +721,6 @@ func processExecution(action actionType) bool {
 		actTransformInputLabel,
 		actTransformListLabel,
 		actTransformNth,
-		actTransformWithNth,
 		actTransformPointer,
 		actTransformPreviewLabel,
 		actTransformPrompt,
@@ -745,7 +737,6 @@ func processExecution(action actionType) bool {
 		actBgTransformInputLabel,
 		actBgTransformListLabel,
 		actBgTransformNth,
-		actBgTransformWithNth,
 		actBgTransformPointer,
 		actBgTransformPreviewLabel,
 		actBgTransformPrompt,
@@ -775,15 +766,10 @@ type placeholderFlags struct {
 	raw           bool
 }

-type withNthSpec struct {
-	fn func([]Token, int32) string // nil = clear (restore original)
-}
-
 type searchRequest struct {
 	sort        bool
 	sync        bool
 	nth         *[]Range
-	withNth     *withNthSpec
 	headerLines *int
 	command     *commandSpec
 	environ     []string
@@ -1094,9 +1080,6 @@ func NewTerminal(opts *Options, eventBox *util.EventBox, executor *util.Executor
 		nthAttr:            opts.Theme.Nth.Attr,
 		nth:                opts.Nth,
 		nthCurrent:         opts.Nth,
-		withNthDefault:     opts.WithNthExpr,
-		withNthExpr:        opts.WithNthExpr,
-		withNthEnabled:     opts.WithNth != nil,
 		tabstop:            opts.Tabstop,
 		raw:                opts.Raw,
 		hasStartActions:    false,
@@ -1368,9 +1351,6 @@ func (t *Terminal) environImpl(forPreview bool) []string {
 	if len(t.nthCurrent) > 0 {
 		env = append(env, "FZF_NTH="+RangesToString(t.nthCurrent))
 	}
-	if len(t.withNthExpr) > 0 {
-		env = append(env, "FZF_WITH_NTH="+t.withNthExpr)
-	}
 	if t.raw {
 		val := "0"
 		if t.isCurrentItemMatch() {
@@ -1554,7 +1534,7 @@ func (t *Terminal) ansiLabelPrinter(str string, color *tui.ColorPair, fill bool)
 	printFn := func(window tui.Window, limit int) {
 		if offsets == nil {
 			// tui.Col* are not initialized until renderer.Init()
-			offsets = result.colorOffsets(nil, nil, t.theme, *color, *color, t.nthAttr, 0, false)
+			offsets = result.colorOffsets(nil, nil, t.theme, *color, *color, t.nthAttr, false)
 		}
 		for limit > 0 {
 			if length > limit {
@@ -1617,7 +1597,7 @@ func (t *Terminal) parsePrompt(prompt string) (func(), int) {
 				return 1
 			}
 			t.printHighlighted(
-				Result{item: item}, tui.ColPrompt, tui.ColPrompt, false, false, false, line, line, true, preTask, nil, 0)
+				Result{item: item}, tui.ColPrompt, tui.ColPrompt, false, false, false, line, line, true, preTask, nil)
 		})
 		t.wrap = wrap
 	}
@@ -1740,17 +1720,6 @@ func (t *Terminal) Input() (bool, []rune) {
 	return paused, copySlice(src)
 }

-// PauseRendering blocks the terminal from reading items.
-// Must be paired with ResumeRendering.
-func (t *Terminal) PauseRendering() {
-	t.mutex.Lock()
-}
-
-// ResumeRendering releases the lock acquired by PauseRendering.
-func (t *Terminal) ResumeRendering() {
-	t.mutex.Unlock()
-}
-
 // UpdateCount updates the count information
 func (t *Terminal) UpdateCount(cnt int, final bool, failedCommand *string) {
 	t.mutex.Lock()
@@ -1873,21 +1842,6 @@ func (t *Terminal) UpdateList(result MatchResult) {
 		}
 		t.revision = newRevision
 		t.version++
-
-		// Filter out selections that no longer match after with-nth change.
-		// Must be inside the revision check so we don't consume the flag
-		// on a stale EvtSearchFin from a previous search.
-		if t.filterSelection && t.multi > 0 && len(t.selected) > 0 {
-			matchMap := t.resultMerger.ToMap()
-			filtered := make(map[int32]selectedItem)
-			for k, v := range t.selected {
-				if _, matched := matchMap[k]; matched {
-					filtered[k] = v
-				}
-			}
-			t.selected = filtered
-		}
-		t.filterSelection = false
 	}
 	if t.triggerLoad {
 		t.triggerLoad = false
@@ -3185,7 +3139,7 @@ func (t *Terminal) printFooter() {
 				func(markerClass) int {
 					t.footerWindow.Print(indent)
 					return indentSize
-				}, nil, 0)
+				}, nil)
 		}
 	})
 	t.wrap = wrap
@@ -3269,7 +3223,7 @@ func (t *Terminal) printHeaderImpl(window tui.Window, borderShape tui.BorderShap
 			func(markerClass) int {
 				t.window.Print(indent)
 				return indentSize
-			}, nil, 0)
+			}, nil)
 	}
 	t.wrap = wrap
 }
@@ -3507,14 +3461,7 @@ func (t *Terminal) printItem(result Result, line int, maxLine int, index int, cu
 			}
 			return indentSize
 		}
-		colCurrent := tui.ColCurrent
-		nthOverlay := t.theme.NthCurrentAttr
-		if selected {
-			nthOverlay = t.theme.NthSelectedAttr.Merge(t.theme.NthCurrentAttr)
-			baseAttr := tui.ColNormal.Attr().Merge(t.theme.NthSelectedAttr).Merge(t.theme.NthCurrentAttr)
-			colCurrent = colCurrent.WithNewAttr(baseAttr)
-		}
-		finalLineNum = t.printHighlighted(result, colCurrent, tui.ColCurrentMatch, true, true, !matched, line, maxLine, forceRedraw, preTask, postTask, nthOverlay)
+		finalLineNum = t.printHighlighted(result, tui.ColCurrent, tui.ColCurrentMatch, true, true, !matched, line, maxLine, forceRedraw, preTask, postTask)
 	} else {
 		preTask := func(marker markerClass) int {
 			w := t.window.Width() - t.pointerLen
@@ -3548,11 +3495,7 @@ func (t *Terminal) printItem(result Result, line int, maxLine int, index int, cu
 			base = base.WithBg(altBg)
 			match = match.WithBg(altBg)
 		}
-		var nthOverlay tui.Attr
-		if selected {
-			nthOverlay = t.theme.NthSelectedAttr
-		}
-		finalLineNum = t.printHighlighted(result, base, match, false, true, !matched, line, maxLine, forceRedraw, preTask, postTask, nthOverlay)
+		finalLineNum = t.printHighlighted(result, base, match, false, true, !matched, line, maxLine, forceRedraw, preTask, postTask)
 	}
 	for i := 0; i < t.gap && finalLineNum < maxLine; i++ {
 		finalLineNum++
@@ -3653,7 +3596,7 @@ func (t *Terminal) overflow(runes []rune, max int) bool {
 	return t.displayWidthWithLimit(runes, 0, max) > max
 }

-func (t *Terminal) printHighlighted(result Result, colBase tui.ColorPair, colMatch tui.ColorPair, current bool, match bool, hidden bool, lineNum int, maxLineNum int, forceRedraw bool, preTask func(markerClass) int, postTask func(int, int, bool, bool, tui.ColorPair), nthOverlay tui.Attr) int {
+func (t *Terminal) printHighlighted(result Result, colBase tui.ColorPair, colMatch tui.ColorPair, current bool, match bool, hidden bool, lineNum int, maxLineNum int, forceRedraw bool, preTask func(markerClass) int, postTask func(int, int, bool, bool, tui.ColorPair)) int {
 	var displayWidth int
 	item := result.item
 	matchOffsets := []Offset{}
@@ -3694,9 +3637,7 @@ func (t *Terminal) printHighlighted(result Result, colBase tui.ColorPair, colMat
 			// But if 'nth' is set to 'regular', it's a sign that you're applying
 			// a different style to the rest of the string. e.g. 'nth:regular,fg:dim'
 			// In this case, we still need to apply it to clear the style.
-			fgAttr := tui.ColNormal.Attr()
-			nthAttrFinal := fgAttr.Merge(t.nthAttr).Merge(nthOverlay)
-			colBase = colBase.WithNewAttr(nthAttrFinal)
+			colBase = colBase.WithAttr(t.nthAttr)
 		}
 		if !wholeCovered && t.nthAttr > 0 {
 			var tokens []Token
@@ -3715,7 +3656,7 @@ func (t *Terminal) printHighlighted(result Result, colBase tui.ColorPair, colMat
 			sort.Sort(ByOrder(nthOffsets))
 		}
 	}
-	allOffsets := result.colorOffsets(charOffsets, nthOffsets, t.theme, colBase, colMatch, t.nthAttr, nthOverlay, hidden)
+	allOffsets := result.colorOffsets(charOffsets, nthOffsets, t.theme, colBase, colMatch, t.nthAttr, hidden)

 	// Determine split offset for horizontal scrolling with freeze
 	splitOffset1 := -1
@@ -3905,7 +3846,6 @@ func (t *Terminal) printHighlighted(result Result, colBase tui.ColorPair, colMat
 			frozenRight = line[splitOffsetRight:]
 		}
 		displayWidthSum := 0
-		displayWidthLeft := 0
 		todo := [3]func(){}
 		for fidx, runes := range [][]rune{frozenLeft, frozenRight, middle} {
 			if len(runes) == 0 {
@@ -3931,11 +3871,7 @@ func (t *Terminal) printHighlighted(result Result, colBase tui.ColorPair, colMat
 				// For frozen parts, reserve space for the ellipsis in the middle part
 				adjustedMaxWidth -= ellipsisWidth
 			}
-			var prefixWidth int
-			if fidx == 2 {
-				prefixWidth = displayWidthLeft
-			}
-			displayWidth = t.displayWidthWithLimit(runes, prefixWidth, adjustedMaxWidth)
+			displayWidth = t.displayWidthWithLimit(runes, 0, adjustedMaxWidth)
 			if !t.wrap && displayWidth > adjustedMaxWidth {
 				maxe = util.Constrain(maxe+min(maxWidth/2-ellipsisWidth, t.hscrollOff), 0, len(runes))
 				transformOffsets := func(diff int32) {
@@ -3973,9 +3909,6 @@ func (t *Terminal) printHighlighted(result Result, colBase tui.ColorPair, colMat
 				displayWidth = t.displayWidthWithLimit(runes, 0, maxWidth)
 			}
 			displayWidthSum += displayWidth
-			if fidx == 0 {
-				displayWidthLeft = displayWidth
-			}

 			if maxWidth > 0 {
 				color := colBase
@@ -3983,7 +3916,7 @@ func (t *Terminal) printHighlighted(result Result, colBase tui.ColorPair, colMat
 					color = color.WithFg(t.theme.Nomatch)
 				}
 				todo[fidx] = func() {
-					t.printColoredString(t.window, runes, offs, color, prefixWidth)
+					t.printColoredString(t.window, runes, offs, color)
 				}
 			} else {
 				break
@@ -4010,13 +3943,10 @@ func (t *Terminal) printHighlighted(result Result, colBase tui.ColorPair, colMat
 	return finalLineNum
 }

-func (t *Terminal) printColoredString(window tui.Window, text []rune, offsets []colorOffset, colBase tui.ColorPair, initialPrefixWidth ...int) {
+func (t *Terminal) printColoredString(window tui.Window, text []rune, offsets []colorOffset, colBase tui.ColorPair) {
 	var index int32
 	var substr string
 	var prefixWidth int
-	if len(initialPrefixWidth) > 0 {
-		prefixWidth = initialPrefixWidth[0]
-	}
 	maxOffset := int32(len(text))
 	var url *url
 	for _, offset := range offsets {
@@ -4223,7 +4153,7 @@ func (t *Terminal) followOffset() int {
 	for i := len(body) - 1; i >= 0; i-- {
 		h := t.previewLineHeight(body[i], maxWidth)
 		if visualLines+h > height {
-			return min(len(lines)-1, headerLines+i+1)
+			return headerLines + i + 1
 		}
 		visualLines += h
 	}
@@ -4521,7 +4451,7 @@ Loop:
 				}
 			}

-			t.previewer.scrollable = t.previewer.scrollable || t.pwindow.Y() == height-1 && t.pwindow.X() == t.pwindow.Width() || t.previewed.filled
+			t.previewer.scrollable = t.previewer.scrollable || t.pwindow.Y() == height-1 && t.pwindow.X() == t.pwindow.Width()
 			if fillRet == tui.FillNextLine {
 				continue
 			} else if fillRet == tui.FillSuspend {
@@ -4544,7 +4474,7 @@ Loop:
 		}
 		lineNo++
 	}
-	t.previewer.scrollable = t.previewer.scrollable || t.previewed.filled || index < len(lines)-1
+	t.previewer.scrollable = t.previewer.scrollable || index < len(lines)-1
 	t.previewed.image = image
 	t.previewed.wireframe = wireframe
 }
@@ -5992,7 +5922,6 @@ func (t *Terminal) Loop() error {
 	events := []util.EventType{}
 	changed := false
 	var newNth *[]Range
-	var newWithNth *withNthSpec
 	var newHeaderLines *int
 	req := func(evts ...util.EventType) {
 		for _, event := range evts {
@@ -6010,7 +5939,6 @@ func (t *Terminal) Loop() error {
 		events = []util.EventType{}
 		changed = false
 		newNth = nil
-		newWithNth = nil
 		newHeaderLines = nil
 		beof := false
 		queryChanged := false
@@ -6402,33 +6330,6 @@ func (t *Terminal) Loop() error {
 						t.forceRerenderList()
 					}
 				})
-			case actChangeWithNth, actTransformWithNth, actBgTransformWithNth:
-				if !t.withNthEnabled {
-					break Action
-				}
-				capture(true, func(expr string) {
-					tokens := strings.Split(expr, "|")
-					withNthExpr := tokens[0]
-					if len(tokens) > 1 {
-						a.a = strings.Join(append(tokens[1:], tokens[0]), "|")
-					}
-					// Empty value restores the default --with-nth
-					if len(withNthExpr) == 0 {
-						withNthExpr = t.withNthDefault
-					}
-					if withNthExpr != t.withNthExpr {
-						if factory, err := nthTransformer(withNthExpr); err == nil {
-							newWithNth = &withNthSpec{fn: factory(t.delimiter)}
-						} else {
-							return
-						}
-						t.withNthExpr = withNthExpr
-						t.filterSelection = true
-						changed = true
-						t.clearNumLinesCache()
-						t.forceRerenderList()
-					}
-				})
 			case actChangeQuery:
 				t.input = []rune(a.a)
 				t.cx = len(t.input)
@@ -7576,7 +7477,7 @@ func (t *Terminal) Loop() error {
 		reload := changed || newCommand != nil
 		var reloadRequest *searchRequest
 		if reload {
-			reloadRequest = &searchRequest{sort: t.sort, sync: reloadSync, nth: newNth, withNth: newWithNth, headerLines: newHeaderLines, command: newCommand, environ: t.environ(), changed: changed, denylist: denylist, revision: t.resultMerger.Revision()}
+			reloadRequest = &searchRequest{sort: t.sort, sync: reloadSync, nth: newNth, headerLines: newHeaderLines, command: newCommand, environ: t.environ(), changed: changed, denylist: denylist, revision: t.resultMerger.Revision()}
 		}

 		// Dispatch queued background requests
--- a/src/tokenizer.go
+++ b/src/tokenizer.go
@@ -161,7 +161,7 @@ func awkTokenizer(input string) ([]string, int) {
 	end := 0
 	for idx := 0; idx < len(input); idx++ {
 		r := input[idx]
-		white := r == 9 || r == 32 || r == 10
+		white := r == 9 || r == 32
 		switch state {
 		case awkNil:
 			if white {
@@ -218,12 +218,11 @@ func Tokenize(text string, delimiter Delimiter) []Token {
 	return withPrefixLengths(tokens, 0)
 }

-// StripLastDelimiter removes the trailing delimiter
+// StripLastDelimiter removes the trailing delimiter and whitespaces
 func StripLastDelimiter(str string, delimiter Delimiter) string {
 	if delimiter.str != nil {
-		return strings.TrimSuffix(str, *delimiter.str)
-	}
-	if delimiter.regex != nil {
+		str = strings.TrimSuffix(str, *delimiter.str)
+	} else if delimiter.regex != nil {
 		locs := delimiter.regex.FindAllStringIndex(str, -1)
 		if len(locs) > 0 {
 			lastLoc := locs[len(locs)-1]
@@ -231,7 +230,6 @@ func StripLastDelimiter(str string, delimiter Delimiter) string {
 				str = str[:lastLoc[0]]
 			}
 		}
-		return str
 	}
 	return strings.TrimRightFunc(str, unicode.IsSpace)
 }
--- a/src/tokenizer_test.go
+++ b/src/tokenizer_test.go
@@ -56,9 +56,9 @@ func TestParseRange(t *testing.T) {

 func TestTokenize(t *testing.T) {
 	// AWK-style
-	input := "  abc: \n\t def:  ghi  "
+	input := "  abc:  def:  ghi  "
 	tokens := Tokenize(input, Delimiter{})
-	if tokens[0].text.ToString() != "abc: \n\t " || tokens[0].prefixLength != 2 {
+	if tokens[0].text.ToString() != "abc:  " || tokens[0].prefixLength != 2 {
 		t.Errorf("%s", tokens)
 	}

@@ -71,9 +71,9 @@ func TestTokenize(t *testing.T) {
 	// With delimiter regex
 	tokens = Tokenize(input, delimiterRegexp("\\s+"))
 	if tokens[0].text.ToString() != "  " || tokens[0].prefixLength != 0 ||
-		tokens[1].text.ToString() != "abc: \n\t " || tokens[1].prefixLength != 2 ||
-		tokens[2].text.ToString() != "def:  " || tokens[2].prefixLength != 10 ||
-		tokens[3].text.ToString() != "ghi  " || tokens[3].prefixLength != 16 {
+		tokens[1].text.ToString() != "abc:  " || tokens[1].prefixLength != 2 ||
+		tokens[2].text.ToString() != "def:  " || tokens[2].prefixLength != 8 ||
+		tokens[3].text.ToString() != "ghi  " || tokens[3].prefixLength != 14 {
 		t.Errorf("%s", tokens)
 	}
 }
--- a/src/tui/tui.go
+++ b/src/tui/tui.go
@@ -447,12 +447,6 @@ func (p ColorPair) WithAttr(attr Attr) ColorPair {
 	return dup
 }

-func (p ColorPair) WithNewAttr(attr Attr) ColorPair {
-	dup := p
-	dup.attr = attr
-	return dup
-}
-
 func (p ColorPair) WithFg(fg ColorAttr) ColorPair {
 	dup := p
 	fgPair := ColorPair{fg.Color, colUndefined, colUndefined, fg.Attr}
@@ -526,8 +520,6 @@ type ColorTheme struct {
 	ListLabel        ColorAttr
 	ListBorder       ColorAttr
 	GapLine          ColorAttr
-	NthCurrentAttr   Attr // raw current-fg attr (before fg merge) for nth overlay
-	NthSelectedAttr  Attr // raw selected-fg attr (before ListFg inherit) for nth overlay
 }

 type Event struct {
@@ -1207,19 +1199,13 @@ func InitTheme(theme *ColorTheme, baseTheme *ColorTheme, boldify bool, forceBlac
 		match.Attr = Underline
 	}
 	theme.Match = o(baseTheme.Match, match)
-	// These colors are not defined in the base themes.
-	// Resolve ListFg/ListBg early so Current and Selected can inherit from them.
-	theme.ListFg = o(theme.Fg, theme.ListFg)
-	theme.ListBg = o(theme.Bg, theme.ListBg)
-	// Inherit from 'list-fg', so that we don't have to write 'current-fg:dim'
+	// Inherit from 'fg', so that we don't have to write 'current-fg:dim'
 	// e.g. fzf --delimiter / --nth -1 --color fg:dim,nth:regular
 	current := theme.Current
 	if !baseTheme.Colored && current.IsUndefined() {
 		current.Attr |= Reverse
 	}
-	resolvedCurrent := o(baseTheme.Current, current)
-	theme.NthCurrentAttr = resolvedCurrent.Attr
-	theme.Current = theme.ListFg.Merge(resolvedCurrent)
+	theme.Current = theme.Fg.Merge(o(baseTheme.Current, current))
 	currentMatch := theme.CurrentMatch
 	if !baseTheme.Colored && currentMatch.IsUndefined() {
 		currentMatch.Attr |= Reverse | Underline
@@ -1244,8 +1230,10 @@ func InitTheme(theme *ColorTheme, baseTheme *ColorTheme, boldify bool, forceBlac
 	scrollbarDefined := theme.Scrollbar != undefined
 	previewBorderDefined := theme.PreviewBorder != undefined

-	theme.NthSelectedAttr = theme.SelectedFg.Attr
-	theme.SelectedFg = theme.ListFg.Merge(theme.SelectedFg)
+	// These colors are not defined in the base themes
+	theme.ListFg = o(theme.Fg, theme.ListFg)
+	theme.ListBg = o(theme.Bg, theme.ListBg)
+	theme.SelectedFg = o(theme.ListFg, theme.SelectedFg)
 	theme.SelectedBg = o(theme.ListBg, theme.SelectedBg)
 	theme.SelectedMatch = o(theme.Match, theme.SelectedMatch)

--- a/src/util/util_unix.go
+++ b/src/util/util_unix.go
@@ -9,7 +9,6 @@ import (
 	"strings"
 	"syscall"

-	"github.com/junegunn/go-shellwords"
 	"golang.org/x/sys/unix"
 )

@@ -21,8 +20,8 @@ type Executor struct {

 func NewExecutor(withShell string) *Executor {
 	shell := os.Getenv("SHELL")
-	args, err := shellwords.Parse(withShell)
-	if err == nil && len(args) > 0 {
+	args := strings.Fields(withShell)
+	if len(args) > 0 {
 		shell = args[0]
 		args = args[1:]
 	} else {
--- a/test/test_core.rb
+++ b/test/test_core.rb
@@ -1190,16 +1190,6 @@ class TestCore < TestInteractive
    tmux.until { |lines| assert lines.any_include?('9999␊10000') }
  end

-  def test_freeze_left_tabstop
-    writelines(%W[1\t2\t3])
-    # With --freeze-left 1 and --tabstop=2:
-    # Frozen left: "1" (width 1)
-    # Middle starts with "\t" at prefix width 1, tabstop 2 → 1 space
-    # Then "2" at column 2, next "\t" at column 3 → 1 space, then "3"
-    tmux.send_keys %(cat #{tempname} | #{FZF} --tabstop=2 --freeze-left 1), :Enter
-    tmux.until { |lines| assert_equal '> 1 2 3', lines[-3] }
-  end
-
  def test_freeze_left_keep_right
    tmux.send_keys %(seq 10000 | #{FZF} --read0 --delimiter "\n" --freeze-left 3 --keep-right --ellipsis XX --no-multi-line --bind space:toggle-multi-line), :Enter
    tmux.until { |lines| assert_match(/^> 1␊2␊3XX.*10000␊$/, lines[-3]) }
@@ -1755,191 +1745,6 @@ class TestCore < TestInteractive
    end
  end

-  def test_change_with_nth
-    input = [
-      'foo bar baz',
-      'aaa bbb ccc',
-      'xxx yyy zzz'
-    ]
-    writelines(input)
-    # Start with field 1 only, cycle through fields, verify $FZF_WITH_NTH via prompt
-    tmux.send_keys %(#{FZF} --with-nth 1 --bind 'space:change-with-nth(2|3|1),result:transform-prompt:echo "[$FZF_WITH_NTH]> "' < #{tempname}), :Enter
-    tmux.until do |lines|
-      assert_equal 3, lines.item_count
-      assert lines.any_include?('[1]>')
-      assert lines.any_include?('foo')
-      refute lines.any_include?('bar')
-    end
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert lines.any_include?('[2]>')
-      assert lines.any_include?('bar')
-      refute lines.any_include?('foo')
-    end
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert lines.any_include?('[3]>')
-      assert lines.any_include?('baz')
-      refute lines.any_include?('bar')
-    end
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert lines.any_include?('[1]>')
-      assert lines.any_include?('foo')
-      refute lines.any_include?('bar')
-    end
-  end
-
-  def test_change_with_nth_default
-    # Empty value restores the default --with-nth
-    tmux.send_keys %(echo -e 'a b c\nd e f' | #{FZF} --with-nth 1 --bind 'space:change-with-nth(2|)'), :Enter
-    tmux.until do |lines|
-      assert_equal 2, lines.item_count
-      assert lines.any_include?('a')
-      refute lines.any_include?('b')
-    end
-    # Switch to field 2
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert lines.any_include?('b')
-      refute lines.any_include?('a')
-    end
-    # Empty restores default (field 1)
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert lines.any_include?('a')
-      refute lines.any_include?('b')
-    end
-  end
-
-  def test_transform_with_nth_search
-    input = [
-      'alpha bravo charlie',
-      'delta echo foxtrot',
-      'golf hotel india'
-    ]
-    writelines(input)
-    tmux.send_keys %(#{FZF} --with-nth 1 --bind 'space:transform-with-nth(echo 2)' -q '^bravo$' < #{tempname}), :Enter
-    tmux.until do |lines|
-      assert_equal 0, lines.match_count
-    end
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert_equal 1, lines.match_count
-    end
-  end
-
-  def test_bg_transform_with_nth_output
-    tmux.send_keys %(echo -e 'a b c\nd e f' | #{FZF} --with-nth 2 --bind 'space:bg-transform-with-nth(echo 3)'), :Enter
-    tmux.until do |lines|
-      assert_equal 2, lines.item_count
-      assert lines.any_include?('b')
-    end
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert lines.any_include?('c')
-      refute lines.any_include?('b')
-    end
-    tmux.send_keys :Enter
-    tmux.until { |lines| assert lines.any_include?('a b c') || lines.any_include?('d e f') }
-  end
-
-  def test_change_with_nth_search
-    input = [
-      'alpha bravo charlie',
-      'delta echo foxtrot',
-      'golf hotel india'
-    ]
-    writelines(input)
-    tmux.send_keys %(#{FZF} --with-nth 1 --bind 'space:change-with-nth(2)' -q '^bravo$' < #{tempname}), :Enter
-    tmux.until do |lines|
-      assert_equal 0, lines.match_count
-    end
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert_equal 1, lines.match_count
-    end
-  end
-
-  def test_change_with_nth_output
-    tmux.send_keys %(echo -e 'a b c\nd e f' | #{FZF} --with-nth 2 --bind 'space:change-with-nth(3)'), :Enter
-    tmux.until do |lines|
-      assert_equal 2, lines.item_count
-      assert lines.any_include?('b')
-    end
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert lines.any_include?('c')
-      refute lines.any_include?('b')
-    end
-    tmux.send_keys :Enter
-    tmux.until { |lines| assert lines.any_include?('a b c') || lines.any_include?('d e f') }
-  end
-
-  def test_change_with_nth_selection
-    # Items: field1 has unique values, field2 has 'match' or 'miss'
-    input = [
-      'one match x',
-      'two miss y',
-      'three match z'
-    ]
-    writelines(input)
-    # Start showing field 2 (match/miss), query 'match', select all matches, then switch to field 3
-    tmux.send_keys %(#{FZF} --with-nth 2 --multi --bind 'ctrl-a:select-all,space:change-with-nth(3)' -q match < #{tempname}), :Enter
-    tmux.until do |lines|
-      assert_equal 2, lines.match_count
-    end
-    # Select all matching items
-    tmux.send_keys 'C-a'
-    tmux.until do |lines|
-      assert lines.any_include?('(2)')
-    end
-    # Now change with-nth to field 3; 'x' and 'z' don't contain 'match'
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert_equal 0, lines.match_count
-      # Selections of non-matching items should be cleared
-      assert lines.any_include?('(0)')
-    end
-  end
-
-  def test_change_with_nth_multiline
-    # Each item has 3 lines: "N-a\nN-b\nN-c"
-    # --with-nth 1 shows 1 line per item, --with-nth 1..3 shows 3 lines per item
-    tmux.send_keys %(seq 20 | xargs -I{} printf '{}-a\\n{}-b\\n{}-c\\0' | #{FZF} --read0 --delimiter "\n" --with-nth 1 --bind 'space:change-with-nth(1..3|1)' --no-sort), :Enter
-    tmux.until do |lines|
-      assert_equal 20, lines.item_count
-      assert lines.any_include?('1-a')
-      refute lines.any_include?('1-b')
-    end
-    # Expand to 3 lines per item
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert lines.any_include?('1-a')
-      assert lines.any_include?('1-b')
-      assert lines.any_include?('1-c')
-    end
-    # Scroll down a few items
-    5.times { tmux.send_keys :Down }
-    tmux.until do |lines|
-      assert lines.any_include?('6-a')
-      assert lines.any_include?('6-b')
-      assert lines.any_include?('6-c')
-    end
-    # Collapse back to 1 line per item
-    tmux.send_keys :Space
-    tmux.until do |lines|
-      assert lines.any_include?('6-a')
-      refute lines.any_include?('6-b')
-    end
-    # Scroll down more after collapse
-    5.times { tmux.send_keys :Down }
-    tmux.until do |lines|
-      assert lines.any_include?('11-a')
-      refute lines.any_include?('11-b')
-    end
-  end
-
  def test_env_vars
    def env_vars
      return {} unless File.exist?(tempname)
@@ -2095,13 +1900,13 @@ class TestCore < TestInteractive
    tmux.send_keys %(echo "foo  ,bar,baz" | #{FZF} -d, --accept-nth 2,2,1,3,1 --sync --bind start:accept > #{tempname}), :Enter
    wait do
      assert_path_exists tempname
-      # Last delimiter is removed
+      # Last delimiter and the whitespaces are removed
      assert_equal ['bar,bar,foo  ,bazfoo'], File.readlines(tempname, chomp: true)
    end
  end

  def test_accept_nth_regex_delimiter
-    tmux.send_keys %(echo "foo  :,:bar,baz" | #{FZF} --delimiter=' *[:,]+ *' --accept-nth 2,2,1,3,1 --sync --bind start:accept > #{tempname}), :Enter
+    tmux.send_keys %(echo "foo  :,:bar,baz" | #{FZF} --delimiter='[:,]+' --accept-nth 2,2,1,3,1 --sync --bind start:accept > #{tempname}), :Enter
    wait do
      assert_path_exists tempname
      # Last delimiter and the whitespaces are removed
@@ -2119,7 +1924,7 @@ class TestCore < TestInteractive
  end

  def test_accept_nth_template
-    tmux.send_keys %(echo "foo  ,bar,baz" | #{FZF} -d " *, *" --accept-nth '[{n}] 1st: {1}, 3rd: {3}, 2nd: {2}' --sync --bind start:accept > #{tempname}), :Enter
+    tmux.send_keys %(echo "foo  ,bar,baz" | #{FZF} -d, --accept-nth '[{n}] 1st: {1}, 3rd: {3}, 2nd: {2}' --sync --bind start:accept > #{tempname}), :Enter
    wait do
      assert_path_exists tempname
      # Last delimiter and the whitespaces are removed
--- a/test/test_preview.rb
+++ b/test/test_preview.rb
@@ -393,20 +393,6 @@ class TestPreview < TestInteractive
    end
  end

-  def test_preview_follow_wrap_long_line
-    tmux.send_keys %(seq 1 | #{FZF} --preview "seq 2; yes yes | head -10000 | tr '\n' ' '" --preview-window follow,wrap --bind up:preview-up,down:preview-down), :Enter
-    tmux.until do |lines|
-      assert_equal 1, lines.match_count
-      assert lines.any_include?('3/3 │')
-    end
-    tmux.send_keys :Up
-    tmux.until { |lines| assert lines.any_include?('2/3 │') }
-    tmux.send_keys :Up
-    tmux.until { |lines| assert lines.any_include?('1/3 │') }
-    tmux.send_keys :Down
-    tmux.until { |lines| assert lines.any_include?('2/3 │') }
-  end
-
  def test_close
    tmux.send_keys "seq 100 | #{FZF} --preview 'echo foo' --bind ctrl-c:close", :Enter
    tmux.until { |lines| assert_equal 100, lines.match_count }
@@ -607,7 +593,7 @@ class TestPreview < TestInteractive
  end

  def test_preview_wrap_sign_between_ansi_fragments_overflow
-    tmux.send_keys %(seq 1 | #{FZF} --preview 'echo -e "\\x1b[33m123 \\x1b[mhi"; echo -e "\\x1b[33m123 \\x1b[mhi"' --preview-window 2,wrap-word,noinfo), :Enter
+    tmux.send_keys %(seq 1 | #{FZF} --preview 'echo -e "\\x1b[33m1234567890 \\x1b[mhello"; echo -e "\\x1b[33m1234567890 \\x1b[mhello"' --preview-window 2,wrap-word), :Enter
    tmux.until do |lines|
      assert_equal 1, lines.match_count
      assert_equal(2, lines.count { |line| line.include?('│ 12 │') })
@@ -616,7 +602,7 @@ class TestPreview < TestInteractive
  end

  def test_preview_wrap_sign_between_ansi_fragments_overflow2
-    tmux.send_keys %(seq 1 | #{FZF} --preview 'echo -e "\\x1b[33m123 \\x1b[mhi"; echo -e "\\x1b[33m123 \\x1b[mhi"' --preview-window 1,wrap-word,noinfo), :Enter
+    tmux.send_keys %(seq 1 | #{FZF} --preview 'echo -e "\\x1b[33m1234567890 \\x1b[mhello"; echo -e "\\x1b[33m1234567890 \\x1b[mhello"' --preview-window 1,wrap-word), :Enter
    tmux.until do |lines|
      assert_equal 1, lines.match_count
      assert_equal(2, lines.count { |line| line.include?('│ 1 │') })
--- a/typos.toml
+++ b/typos.toml
@@ -7,4 +7,4 @@ tabe = "tabe"
 Iterm = "Iterm"

 [files]
-extend-exclude = ["README.md", "*.s"]
+extend-exclude = ["README.md"]