feat(agent): add fuzzy whitespace matching to edit_files tool (#22446)

Inspired by openai/codex's `apply_patch` implementation, this changes
the `edit_files` search-and-replace to use a cascading match strategy
when the exact search string isn't found:

1. **Exact substring match** (byte-for-byte) — existing behavior,
unchanged
2. **Line-by-line match ignoring trailing whitespace** — handles
trailing spaces/tabs the LLM omits
3. **Line-by-line match ignoring all leading/trailing whitespace** —
handles tabs-vs-spaces and wrong indentation depth

## Problem

When the chat agent uses `edit_files`, it generates a search string that
must match the file content exactly. LLMs frequently get whitespace
wrong:
- Emitting spaces when the file uses tabs (or vice versa)
- Getting the indentation depth wrong by one or more levels
- Omitting trailing whitespace that exists in the file

When this happens, the edit silently does nothing, and the agent falls
into a retry loop using `cat -A` to diagnose the exact whitespace
characters.

## Solution

Adopted the same cascading fuzzy match strategy that [openai/codex uses
in
`seek_sequence.rs`](https://github.com/openai/codex/blob/main/codex-rs/apply-patch/src/seek_sequence.rs):

- Pass 1: exact match (existing behavior)
- Pass 2: `TrimRight` each line before comparing (trailing whitespace
tolerance)
- Pass 3: `TrimSpace` each line before comparing (full indentation
tolerance)

When a fuzzy match is found, the matched lines in the original file are
replaced with the replacement text. This preserves surrounding content
exactly.

## Changes

- `agent/agentfiles/files.go`: Replaced `icholy/replace` streaming
transformer with in-memory `fuzzyReplace` + helper functions
(`seekLines`, `spliceLines`)
- `agent/agentfiles/files_test.go`: Added 6 new test cases covering
trailing whitespace, tabs-vs-spaces, different indent depths, exact
match preference, no-match behavior, and mixed whitespace multiline
edits
- Removed `icholy/replace` dependency from go.mod/go.sum

---------

Co-authored-by: Kyle Carberry <kylecarbs@users.noreply.github.com>
This commit is contained in:
Kyle Carberry
2026-02-28 17:02:57 -05:00
committed by GitHub
co-authored by Kyle Carberry
parent 22d4539a7a
commit 5945febf06
4 changed files with 206 additions and 15 deletions
+106 -7
View File
@@ -13,9 +13,7 @@ import (
"strings"
"syscall"
"github.com/icholy/replace"
"github.com/spf13/afero"
"golang.org/x/text/transform"
"golang.org/x/xerrors"
"cdr.dev/slog/v3"
@@ -422,9 +420,21 @@ func (api *API) editFile(ctx context.Context, path string, edits []workspacesdk.
return http.StatusBadRequest, xerrors.Errorf("open %s: not a file", path)
}
transforms := make([]transform.Transformer, len(edits))
for i, edit := range edits {
transforms[i] = replace.String(edit.Search, edit.Replace)
data, err := io.ReadAll(f)
if err != nil {
return http.StatusInternalServerError, xerrors.Errorf("read %s: %w", path, err)
}
content := string(data)
for _, edit := range edits {
var ok bool
content, ok = fuzzyReplace(content, edit.Search, edit.Replace)
if !ok {
api.logger.Warn(ctx, "edit search string not found, skipping",
slog.F("path", path),
slog.F("search_preview", truncate(edit.Search, 64)),
)
}
}
// Create an adjacent file to ensure it will be on the same device and can be
@@ -435,8 +445,7 @@ func (api *API) editFile(ctx context.Context, path string, edits []workspacesdk.
}
defer tmpfile.Close()
_, err = io.Copy(tmpfile, replace.Chain(f, transforms...))
if err != nil {
if _, err := tmpfile.Write([]byte(content)); err != nil {
if rerr := api.filesystem.Remove(tmpfile.Name()); rerr != nil {
api.logger.Warn(ctx, "unable to clean up temp file", slog.Error(rerr))
}
@@ -450,3 +459,93 @@ func (api *API) editFile(ctx context.Context, path string, edits []workspacesdk.
return 0, nil
}
// fuzzyReplace attempts to find `search` inside `content` and replace its first
// occurrence with `replace`. It uses a cascading match strategy inspired by
// openai/codex's apply_patch:
//
// 1. Exact substring match (byte-for-byte).
// 2. Line-by-line match ignoring trailing whitespace on each line.
// 3. Line-by-line match ignoring all leading/trailing whitespace (indentation-tolerant).
//
// When a fuzzy match is found (passes 2 or 3), the replacement is still applied
// at the byte offsets of the original content so that surrounding text (including
// indentation of untouched lines) is preserved.
//
// Returns the (possibly modified) content and a bool indicating whether a match
// was found.
func fuzzyReplace(content, search, replace string) (string, bool) {
// Pass 1 – exact substring (replace all occurrences).
if strings.Contains(content, search) {
return strings.ReplaceAll(content, search, replace), true
}
// For line-level fuzzy matching we split both content and search into lines.
contentLines := strings.SplitAfter(content, "\n")
searchLines := strings.SplitAfter(search, "\n")
// A trailing newline in the search produces an empty final element from
// SplitAfter. Drop it so it doesn't interfere with line matching.
if len(searchLines) > 0 && searchLines[len(searchLines)-1] == "" {
searchLines = searchLines[:len(searchLines)-1]
}
// Pass 2 – trim trailing whitespace on each line.
if start, end, ok := seekLines(contentLines, searchLines, func(a, b string) bool {
return strings.TrimRight(a, " \t\r\n") == strings.TrimRight(b, " \t\r\n")
}); ok {
return spliceLines(contentLines, start, end, replace), true
}
// Pass 3 – trim all leading and trailing whitespace (indentation-tolerant).
if start, end, ok := seekLines(contentLines, searchLines, func(a, b string) bool {
return strings.TrimSpace(a) == strings.TrimSpace(b)
}); ok {
return spliceLines(contentLines, start, end, replace), true
}
return content, false
}
// seekLines scans contentLines looking for a contiguous subsequence that matches
// searchLines according to the provided `eq` function. It returns the start and
// end (exclusive) indices into contentLines of the match.
func seekLines(contentLines, searchLines []string, eq func(a, b string) bool) (start, end int, ok bool) {
if len(searchLines) == 0 {
return 0, 0, true
}
if len(searchLines) > len(contentLines) {
return 0, 0, false
}
outer:
for i := 0; i <= len(contentLines)-len(searchLines); i++ {
for j, sLine := range searchLines {
if !eq(contentLines[i+j], sLine) {
continue outer
}
}
return i, i + len(searchLines), true
}
return 0, 0, false
}
// spliceLines replaces contentLines[start:end] with replacement text, returning
// the full content as a single string.
func spliceLines(contentLines []string, start, end int, replacement string) string {
var b strings.Builder
for _, l := range contentLines[:start] {
_, _ = b.WriteString(l)
}
_, _ = b.WriteString(replacement)
for _, l := range contentLines[end:] {
_, _ = b.WriteString(l)
}
return b.String()
}
func truncate(s string, n int) string {
if len(s) <= n {
return s
}
return s[:n] + "..."
}
+100
View File
@@ -649,6 +649,106 @@ func TestEditFiles(t *testing.T) {
filepath.Join(tmpdir, "file3"): "edited3 3",
},
},
{
name: "TrailingWhitespace",
contents: map[string]string{filepath.Join(tmpdir, "trailing-ws"): "foo \nbar\t\t\nbaz"},
edits: []workspacesdk.FileEdits{
{
Path: filepath.Join(tmpdir, "trailing-ws"),
Edits: []workspacesdk.FileEdit{
{
Search: "foo\nbar\nbaz",
Replace: "replaced",
},
},
},
},
expected: map[string]string{filepath.Join(tmpdir, "trailing-ws"): "replaced"},
},
{
name: "TabsVsSpaces",
contents: map[string]string{filepath.Join(tmpdir, "tabs-vs-spaces"): "\tif true {\n\t\tfoo()\n\t}"},
edits: []workspacesdk.FileEdits{
{
Path: filepath.Join(tmpdir, "tabs-vs-spaces"),
Edits: []workspacesdk.FileEdit{
{
// Search uses spaces but file uses tabs.
Search: " if true {\n foo()\n }",
Replace: "\tif true {\n\t\tbar()\n\t}",
},
},
},
},
expected: map[string]string{filepath.Join(tmpdir, "tabs-vs-spaces"): "\tif true {\n\t\tbar()\n\t}"},
},
{
name: "DifferentIndentDepth",
contents: map[string]string{filepath.Join(tmpdir, "indent-depth"): "\t\t\tdeep()\n\t\t\tnested()"},
edits: []workspacesdk.FileEdits{
{
Path: filepath.Join(tmpdir, "indent-depth"),
Edits: []workspacesdk.FileEdit{
{
// Search has wrong indent depth (1 tab instead of 3).
Search: "\tdeep()\n\tnested()",
Replace: "\t\t\tdeep()\n\t\t\tchanged()",
},
},
},
},
expected: map[string]string{filepath.Join(tmpdir, "indent-depth"): "\t\t\tdeep()\n\t\t\tchanged()"},
},
{
name: "ExactMatchPreferred",
contents: map[string]string{filepath.Join(tmpdir, "exact-preferred"): "hello world"},
edits: []workspacesdk.FileEdits{
{
Path: filepath.Join(tmpdir, "exact-preferred"),
Edits: []workspacesdk.FileEdit{
{
Search: "hello world",
Replace: "goodbye world",
},
},
},
},
expected: map[string]string{filepath.Join(tmpdir, "exact-preferred"): "goodbye world"},
},
{
name: "NoMatchStillSucceeds",
contents: map[string]string{filepath.Join(tmpdir, "no-match"): "original content"},
edits: []workspacesdk.FileEdits{
{
Path: filepath.Join(tmpdir, "no-match"),
Edits: []workspacesdk.FileEdit{
{
Search: "this does not exist in the file",
Replace: "whatever",
},
},
},
},
// File should remain unchanged.
expected: map[string]string{filepath.Join(tmpdir, "no-match"): "original content"},
},
{
name: "MixedWhitespaceMultiline",
contents: map[string]string{filepath.Join(tmpdir, "mixed-ws"): "func main() {\n\tresult := compute()\n\tfmt.Println(result)\n}"},
edits: []workspacesdk.FileEdits{
{
Path: filepath.Join(tmpdir, "mixed-ws"),
Edits: []workspacesdk.FileEdit{
{
// Search uses spaces, file uses tabs.
Search: " result := compute()\n fmt.Println(result)\n",
Replace: "\tresult := compute()\n\tlog.Println(result)\n",
},
},
},
},
expected: map[string]string{filepath.Join(tmpdir, "mixed-ws"): "func main() {\n\tresult := compute()\n\tlog.Println(result)\n}"},
},
{
name: "MultiError",
contents: map[string]string{