fix: error on ambiguous matches in edit_files tool (#23125)

## Problem

The `edit_files` tool used `strings.ReplaceAll` for exact substring
matches, silently replacing **every** occurrence. When an LLM's search
string wasn't unique in the file, this caused unintended edits. Fuzzy
matches (passes 2 and 3) only replaced the first occurrence, creating
inconsistent behavior. Zero matches were also silently ignored.

## Investigation

Investigated how **coder/mux** and **openai/codex** handle this:

| Tool | Multiple matches | No match | Flag |
|---|---|---|---|
| **coder/mux** `file_edit_replace_string` | Error (default
`replace_count=1`) | Error | `replace_count` (int, default 1, -1=all) |
| **openai/codex** `apply_patch` | Uses first match after cursor
(structural disambiguation via context lines + `@@` markers) | Error |
None (different paradigm) |
| **coder/coder** `edit_files` (before) | Exact: replaces all. Fuzzy:
replaces first. | Silent success | None |

## Solution

Adopted the mux approach (error on ambiguity) with a simpler
`replace_all: bool` instead of `replace_count: int`:

- **Default (`replace_all: false`)**: search string must match exactly
once. Multiple matches → error with guidance: *"search string matches N
occurrences. Include more surrounding context to make the match unique,
or set replace_all to true"*
- **`replace_all: true`**: replaces all occurrences (opt-in for
intentional bulk operations like variable renames)
- **Zero matches**: now returns an error instead of silently succeeding

Chose `bool` over `int` count because:
1. LLMs are bad at counting occurrences
2. The real intent is binary (one specific spot vs. all occurrences)
3. Simpler error recovery loop for the LLM

## Changes

| File | Change |
|---|---|
| `codersdk/workspacesdk/agentconn.go` | Add `ReplaceAll bool` to
`FileEdit` struct |
| `agent/agentfiles/files.go` | Count matches before replacing; error if
>1 and not opted in; error on zero matches; add `countLineMatches`
helper |
| `codersdk/toolsdk/toolsdk.go` | Expose `replace_all` in tool schema
with description |
| `agent/agentfiles/files_test.go` | Update existing tests, add
`EditEditAmbiguous`, `EditEditReplaceAll`, `NoMatchErrors`,
`AmbiguousExactMatch`, `ReplaceAllExact` |
This commit is contained in:
Kyle Carberry
2026-03-16 16:17:33 +00:00
committed by GitHub
parent 4fdd48b3f5
commit 32a894d4a7
4 changed files with 165 additions and 44 deletions
+89 -38
View File
@@ -447,13 +447,10 @@ func (api *API) editFile(ctx context.Context, path string, edits []workspacesdk.
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)),
)
var err error
content, err = fuzzyReplace(content, edit)
if err != nil {
return http.StatusBadRequest, xerrors.Errorf("edit %s: %w", path, err)
}
}
@@ -480,51 +477,92 @@ 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
// fuzzyReplace attempts to find `search` inside `content` and replace it
// 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).
// 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.
// When edit.ReplaceAll is false (the default), the search string must
// match exactly one location. If multiple matches are found, an error
// is returned asking the caller to include more context or set
// replace_all.
//
// 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).
// 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.
func fuzzyReplace(content string, edit workspacesdk.FileEdit) (string, error) {
search := edit.Search
replace := edit.Replace
// Pass 1 – exact substring match.
if strings.Contains(content, search) {
return strings.ReplaceAll(content, search, replace), true
if edit.ReplaceAll {
return strings.ReplaceAll(content, search, replace), nil
}
count := strings.Count(content, search)
if count > 1 {
return "", xerrors.Errorf("search string matches %d occurrences "+
"(expected exactly 1). Include more surrounding "+
"context to make the match unique, or set "+
"replace_all to true", count)
}
// Exactly one match.
return strings.Replace(content, search, replace, 1), nil
}
// For line-level fuzzy matching we split both content and search into lines.
// 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.
// 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 {
trimRight := 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 {
trimAll := func(a, b string) bool {
return strings.TrimSpace(a) == strings.TrimSpace(b)
}); ok {
return spliceLines(contentLines, start, end, replace), true
}
return content, false
// Pass 2 – trim trailing whitespace on each line.
if start, end, ok := seekLines(contentLines, searchLines, trimRight); ok {
if !edit.ReplaceAll {
if count := countLineMatches(contentLines, searchLines, trimRight); count > 1 {
return "", xerrors.Errorf("search string matches %d occurrences "+
"(expected exactly 1). Include more surrounding "+
"context to make the match unique, or set "+
"replace_all to true", count)
}
}
return spliceLines(contentLines, start, end, replace), nil
}
// Pass 3 – trim all leading and trailing whitespace
// (indentation-tolerant).
if start, end, ok := seekLines(contentLines, searchLines, trimAll); ok {
if !edit.ReplaceAll {
if count := countLineMatches(contentLines, searchLines, trimAll); count > 1 {
return "", xerrors.Errorf("search string matches %d occurrences "+
"(expected exactly 1). Include more surrounding "+
"context to make the match unique, or set "+
"replace_all to true", count)
}
}
return spliceLines(contentLines, start, end, replace), nil
}
return "", xerrors.New("search string not found in file. Verify the search " +
"string matches the file content exactly, including whitespace " +
"and indentation")
}
// seekLines scans contentLines looking for a contiguous subsequence that matches
@@ -549,6 +587,26 @@ outer:
return 0, 0, false
}
// countLineMatches counts how many non-overlapping contiguous
// subsequences of contentLines match searchLines according to eq.
func countLineMatches(contentLines, searchLines []string, eq func(a, b string) bool) int {
count := 0
if len(searchLines) == 0 || len(searchLines) > len(contentLines) {
return count
}
outer:
for i := 0; i <= len(contentLines)-len(searchLines); i++ {
for j, sLine := range searchLines {
if !eq(contentLines[i+j], sLine) {
continue outer
}
}
count++
i += len(searchLines) - 1 // skip past this match
}
return count
}
// 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 {
@@ -562,10 +620,3 @@ func spliceLines(contentLines []string, start, end int, replacement string) stri
}
return b.String()
}
func truncate(s string, n int) string {
if len(s) <= n {
return s
}
return s[:n] + "..."
}
+68 -3
View File
@@ -576,7 +576,9 @@ func TestEditFiles(t *testing.T) {
expected: map[string]string{filepath.Join(tmpdir, "edit1"): "bar bar"},
},
{
name: "EditEdit", // Edits affect previous edits.
// When the second edit creates ambiguity (two "bar"
// occurrences), it should fail.
name: "EditEditAmbiguous",
contents: map[string]string{filepath.Join(tmpdir, "edit-edit"): "foo bar"},
edits: []workspacesdk.FileEdits{
{
@@ -593,7 +595,33 @@ func TestEditFiles(t *testing.T) {
},
},
},
expected: map[string]string{filepath.Join(tmpdir, "edit-edit"): "qux qux"},
errCode: http.StatusBadRequest,
errors: []string{"matches 2 occurrences"},
// File should not be modified on error.
expected: map[string]string{filepath.Join(tmpdir, "edit-edit"): "foo bar"},
},
{
// With replace_all the cascading edit replaces
// both occurrences.
name: "EditEditReplaceAll",
contents: map[string]string{filepath.Join(tmpdir, "edit-edit-ra"): "foo bar"},
edits: []workspacesdk.FileEdits{
{
Path: filepath.Join(tmpdir, "edit-edit-ra"),
Edits: []workspacesdk.FileEdit{
{
Search: "foo",
Replace: "bar",
},
{
Search: "bar",
Replace: "qux",
ReplaceAll: true,
},
},
},
},
expected: map[string]string{filepath.Join(tmpdir, "edit-edit-ra"): "qux qux"},
},
{
name: "Multiline",
@@ -720,7 +748,7 @@ func TestEditFiles(t *testing.T) {
expected: map[string]string{filepath.Join(tmpdir, "exact-preferred"): "goodbye world"},
},
{
name: "NoMatchStillSucceeds",
name: "NoMatchErrors",
contents: map[string]string{filepath.Join(tmpdir, "no-match"): "original content"},
edits: []workspacesdk.FileEdits{
{
@@ -733,9 +761,46 @@ func TestEditFiles(t *testing.T) {
},
},
},
errCode: http.StatusBadRequest,
errors: []string{"search string not found in file"},
// File should remain unchanged.
expected: map[string]string{filepath.Join(tmpdir, "no-match"): "original content"},
},
{
name: "AmbiguousExactMatch",
contents: map[string]string{filepath.Join(tmpdir, "ambig-exact"): "foo bar foo baz foo"},
edits: []workspacesdk.FileEdits{
{
Path: filepath.Join(tmpdir, "ambig-exact"),
Edits: []workspacesdk.FileEdit{
{
Search: "foo",
Replace: "qux",
},
},
},
},
errCode: http.StatusBadRequest,
errors: []string{"matches 3 occurrences"},
expected: map[string]string{filepath.Join(tmpdir, "ambig-exact"): "foo bar foo baz foo"},
},
{
name: "ReplaceAllExact",
contents: map[string]string{filepath.Join(tmpdir, "ra-exact"): "foo bar foo baz foo"},
edits: []workspacesdk.FileEdits{
{
Path: filepath.Join(tmpdir, "ra-exact"),
Edits: []workspacesdk.FileEdit{
{
Search: "foo",
Replace: "qux",
ReplaceAll: true,
},
},
},
},
expected: map[string]string{filepath.Join(tmpdir, "ra-exact"): "qux bar qux baz qux"},
},
{
name: "MixedWhitespaceMultiline",
contents: map[string]string{filepath.Join(tmpdir, "mixed-ws"): "func main() {\n\tresult := compute()\n\tfmt.Println(result)\n}"},
+5 -1
View File
@@ -1702,12 +1702,16 @@ var WorkspaceEditFiles = Tool[WorkspaceEditFilesArgs, codersdk.Response]{
"properties": map[string]any{
"search": map[string]any{
"type": "string",
"description": "The old string to replace.",
"description": "The old string to replace. Must uniquely match exactly one location in the file unless replace_all is true. Include enough surrounding context to make the match unique.",
},
"replace": map[string]any{
"type": "string",
"description": "The new string that replaces the old string.",
},
"replace_all": map[string]any{
"type": "boolean",
"description": "When true, replaces all occurrences of the search string. Defaults to false, which requires the search string to match exactly once.",
},
},
"required": []string{"search", "replace"},
},
+3 -2
View File
@@ -899,8 +899,9 @@ func DefaultReadFileLinesLimits() ReadFileLinesLimits {
}
type FileEdit struct {
Search string `json:"search"`
Replace string `json:"replace"`
Search string `json:"search"`
Replace string `json:"replace"`
ReplaceAll bool `json:"replace_all,omitempty"`
}
type FileEdits struct {