buffer_diff: Canonicalize ambiguous hunk placement to fix staging corruption by abendrothj · Pull Request #60584 · zed-industries/zed

abendrothj · 2026-07-08T09:50:25Z

Problem

With the repro from #60424, staging the first deletion hunk marked the second deletion as staged in the UI even though git still had it unstaged, and clicking the (incorrectly shown) Unstage button inserted a duplicate copy of the hunk's contents into the git index on every click.

The root cause is ambiguous hunk placement. The committed text contains repeated end\n\n line runs, so the deletion hunks can "slide": more than one placement produces a minimal diff. The uncommitted diff (HEAD vs worktree) anchored the remaining deletion at one row while the unstaged diff (index vs worktree), recomputed after the partial stage, anchored the same logical deletion at a different row. Everything that correlates hunks across those two diffs assumes they agree on positions:

the secondary-status matching in hunks_intersecting_range_impl found no unstaged hunk at the uncommitted hunk's rows and reported it as staged (NoSecondaryHunk);
the worktree→index projection in compute_uncommitted_index_edits treated the hunk's position as unchanged text and, on unstage, inserted the hunk's HEAD content at an index position that already contained it, duplicating it on every request.

Fix

Upgrade imara-diff from 0.1.8 to 0.2.0 and run Diff::postprocess_lines (imara-diff's port of git's xdiff slider/indent heuristic) after computing hunks in buffer_diff. This canonicalizes the placement of ambiguous hunks based only on their local content, so diffs of the same buffer against different base texts anchor the same logical change at the same rows. A bonus is that Zed's hunk placement now matches git diff's output for such cases (git has used the indent heuristic by default since 2.11).

As defense in depth, compute_uncommitted_index_edits now drops a pure-insertion index edit whose content is already present at the target position, so a stale secondary status can no longer duplicate index content.

The imara-diff 0.2 API removed the Sink trait and the top-level diff() function, so the other call sites (language/text_diff.rs, zeta_prompt/udiff.rs, edit_prediction_metrics/reversal.rs) are migrated mechanically to Diff::compute + hunks() with unchanged behavior (0.2 also renamed lines_with_terminator to lines and changed the default &str tokenization to include terminators; the unified-diff builders keep terminator-less tokens via str::lines()).

Testing

New regression test test_staging_hunks_with_ambiguous_placement replays the exact repro from Stage button stages part of the hunk in git but marks whole hunk as staged in UI #60424 (same file contents): stages the first deletion, asserts the remaining hunks keep their unstaged status and the index matches exactly, then issues repeated unstage requests for the unstaged hunk and asserts the index is unchanged. Before the fix this test showed the second hunk flipping to staged and the index growing by one copy of the deleted block per unstage request.
buffer_diff, language, zeta_prompt, edit_prediction_metrics, multi_buffer, git_ui, editor, and the project integration suite pass.
One test expectation updated: editor::test_fold_function_bodies asserted the old placement of an ambiguous deletion (blank line before comment); the canonicalized placement (comment before blank line) matches what git diff produces for the same texts.

Release Notes:

Fixed staging a hunk sometimes marking a different hunk as staged (and subsequent unstaging corrupting the git index) when the diff contained repeated lines (#60424).

cla-bot · 2026-07-08T09:50:29Z

We require contributors to sign our Contributor License Agreement, and we don't have @abendrothj on file. You can sign our CLA at https://zed.dev/cla. Once you've signed, post a comment here that says '@cla-bot check'.

abendrothj · 2026-07-08T09:52:14Z

@cla-bot check

cla-bot · 2026-07-08T09:52:18Z

The cla-bot has been summoned, and re-checked this pull request!

Upgrade imara-diff from 0.1.8 to 0.2.0 and run its port of git's xdiff slider/indent heuristic (Diff::postprocess_lines) when computing buffer diff hunks. Without this, diffs of the same buffer against different base texts (e.g. HEAD vs index) could anchor the same logical change at different rows when the surrounding lines repeat, and code correlating hunks across the uncommitted and unstaged diffs misbehaved: after a partial stage, an unstaged hunk rendered as staged, and unstaging it inserted a duplicate copy of its contents into the git index on every request. Also drop pure-insertion index edits whose content is already present at the target position, so a stale secondary status can no longer duplicate index content, and migrate the remaining imara-diff call sites to the 0.2 API with unchanged behavior. Closes zed-industries#60424

dinocosta · 2026-07-08T10:34:58Z

Hey @abendrothj ! Thank you for taking a stab at fixing the reported issue 🙂

Before someone chimes in with a review, would you be able to share an updated screenshot using the same example used by Kunall in this comment? Makes it easier for us to confirm the fix is working as intended and we can dive into the code. Thanks!

abendrothj · 2026-07-09T00:35:47Z

Hey @abendrothj ! Thank you for taking a stab at fixing the reported issue 🙂

Before someone chimes in with a review, would you be able to share an updated screenshot using the same example used by Kunall in this comment? Makes it easier for us to confirm the fix is working as intended and we can dive into the code. Thanks!

Sure, here's the updated screenshot, as well as a LazyGit one for consistency.

After staging the first deletion block in Zed, only that block is staged; the table:contains deletion and has_class edit remain unstaged.

  $ git status --short
  MM test.txt

  $ git diff --staged -- test.txt
  diff --git a/test.txt b/test.txt
  index a5c7ac0..372ece4 100644
  --- a/test.txt
  +++ b/test.txt
  @@ -1,16 +1,3 @@
  -function is_empty(str)
  -    assert(type(str) == string)
  -    return str == nil or str == ""
  -end
  -
  -function string:starts_with(str)
  -    return self:sub(1, #str) == str
  -end
  -
  -function string:ends_with(str)
  -    return self:sub(-#str) == str
  -end
  -
   function table:append(other)
       for _, val in ipairs(other) do table.insert(self, val) end
   end

  $ git diff -- test.txt
  diff --git a/test.txt b/test.txt
  index 372ece4..0f91d3d 100644
  --- a/test.txt
  +++ b/test.txt
  @@ -2,15 +2,6 @@ function table:append(other)
       for _, val in ipairs(other) do table.insert(self, val) end
   end

  -function table:contains(elem)
  -    assert(type(self) == "table")
  -
  -    for _, val in ipairs(self) do
  -        if val == elem then return true end
  -    end
  -    return false
  -end
  -
   function table:keys()
       local keys = {}
       for key, _ in pairs(self) do
  @@ -19,4 +10,4 @@ function table:keys()
       return keys
   end

  -function has_class(el, class) return el.classes[0] == class end
  +function has_class(el, class) return el.classes:includes(class) end

…placement # Conflicts: # Cargo.toml

cole-miller · 2026-07-18T18:39:02Z

Thanks! I removed the content-based safeguard because it seems like that would reject valid index edits in some cases, so it seems best not to have it. Otherwise this fix seems great.

…ruption (zed-industries#60584) Closes zed-industries#60424 ## Problem With the repro from zed-industries#60424, staging the first deletion hunk marked the second deletion as staged in the UI even though git still had it unstaged, and clicking the (incorrectly shown) Unstage button inserted a duplicate copy of the hunk's contents into the git index on every click. The root cause is ambiguous hunk placement. The committed text contains repeated `end\n\n` line runs, so the deletion hunks can "slide": more than one placement produces a minimal diff. The uncommitted diff (HEAD vs worktree) anchored the remaining deletion at one row while the unstaged diff (index vs worktree), recomputed after the partial stage, anchored the same logical deletion at a different row. Everything that correlates hunks across those two diffs assumes they agree on positions: - the secondary-status matching in `hunks_intersecting_range_impl` found no unstaged hunk at the uncommitted hunk's rows and reported it as staged (`NoSecondaryHunk`); - the worktree→index projection in `compute_uncommitted_index_edits` treated the hunk's position as unchanged text and, on unstage, inserted the hunk's HEAD content at an index position that already contained it, duplicating it on every request. ## Fix Upgrade `imara-diff` from 0.1.8 to 0.2.0 and run `Diff::postprocess_lines` (imara-diff's port of git's xdiff slider/indent heuristic) after computing hunks in `buffer_diff`. This canonicalizes the placement of ambiguous hunks based only on their local content, so diffs of the same buffer against different base texts anchor the same logical change at the same rows. A bonus is that Zed's hunk placement now matches `git diff`'s output for such cases (git has used the indent heuristic by default since 2.11). As defense in depth, `compute_uncommitted_index_edits` now drops a pure-insertion index edit whose content is already present at the target position, so a stale secondary status can no longer duplicate index content. The imara-diff 0.2 API removed the `Sink` trait and the top-level `diff()` function, so the other call sites (`language/text_diff.rs`, `zeta_prompt/udiff.rs`, `edit_prediction_metrics/reversal.rs`) are migrated mechanically to `Diff::compute` + `hunks()` with unchanged behavior (0.2 also renamed `lines_with_terminator` to `lines` and changed the default `&str` tokenization to include terminators; the unified-diff builders keep terminator-less tokens via `str::lines()`). ## Testing - New regression test `test_staging_hunks_with_ambiguous_placement` replays the exact repro from zed-industries#60424 (same file contents): stages the first deletion, asserts the remaining hunks keep their unstaged status and the index matches exactly, then issues repeated unstage requests for the unstaged hunk and asserts the index is unchanged. Before the fix this test showed the second hunk flipping to staged and the index growing by one copy of the deleted block per unstage request. - `buffer_diff`, `language`, `zeta_prompt`, `edit_prediction_metrics`, `multi_buffer`, `git_ui`, `editor`, and the `project` integration suite pass. - One test expectation updated: `editor::test_fold_function_bodies` asserted the old placement of an ambiguous deletion (blank line before comment); the canonicalized placement (comment before blank line) matches what `git diff` produces for the same texts. Release Notes: - Fixed staging a hunk sometimes marking a different hunk as staged (and subsequent unstaging corrupting the git index) when the diff contained repeated lines ([zed-industries#60424](zed-industries#60424)). --------- Co-authored-by: Cole Miller <[email protected]>

zed-community-bot Bot added the first contribution the author's first pull request to Zed. NOTE: the label application is automated via github actions label Jul 8, 2026

cla-bot Bot added the cla-signed The user has signed the Contributor License Agreement label Jul 8, 2026

abendrothj force-pushed the fix-ambiguous-hunk-placement branch from 548fa20 to f67e3fc Compare July 8, 2026 10:01

dinocosta added the area:integrations/git Git integration feedback label Jul 8, 2026

github-project-automation Bot added this to Git board Jul 8, 2026

abendrothj and others added 3 commits July 13, 2026 15:42

Merge remote-tracking branch 'upstream/main' into fix-ambiguous-hunk-…

8981d85

…placement # Conflicts: # Cargo.toml

Remove content-based safeguard

5cf5075

Format

4823f28

cole-miller approved these changes Jul 18, 2026

View reviewed changes

cole-miller enabled auto-merge July 18, 2026 18:39

cole-miller added this pull request to the merge queue Jul 18, 2026

Merged via the queue into zed-industries:main with commit eb96279 Jul 18, 2026
34 checks passed

shenghsi mentioned this pull request Jul 27, 2026

editor: Sync fold test with canonicalized diff hunks shenghsi/flint#114

Merged

This was referenced Jul 31, 2026

🎨 Zed GPUI Release Tracker gpui-archipelago/gpui-release-notes#1

Closed

🎨 Zed GPUI Release Tracker gpui-archipelago/gpui-release-notes#3

Open

Navigation Menu

Repository navigation

buffer_diff: Canonicalize ambiguous hunk placement to fix staging corruption - #60584

buffer_diff: Canonicalize ambiguous hunk placement to fix staging corruption#60584
cole-miller merged 4 commits into
zed-industries:mainfrom
abendrothj:fix-ambiguous-hunk-placement

abendrothj commented Jul 8, 2026 •

edited

Loading

Uh oh!

cla-bot Bot commented Jul 8, 2026

Uh oh!

abendrothj commented Jul 8, 2026

Uh oh!

cla-bot Bot commented Jul 8, 2026

Uh oh!

dinocosta commented Jul 8, 2026

Uh oh!

abendrothj commented Jul 9, 2026 •

edited

Loading

Uh oh!

cole-miller commented Jul 18, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

abendrothj commented Jul 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problem

Fix

Testing

Uh oh!

cla-bot Bot commented Jul 8, 2026

Uh oh!

abendrothj commented Jul 8, 2026

Uh oh!

cla-bot Bot commented Jul 8, 2026

Uh oh!

dinocosta commented Jul 8, 2026

Uh oh!

abendrothj commented Jul 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cole-miller commented Jul 18, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

abendrothj commented Jul 8, 2026 •

edited

Loading

abendrothj commented Jul 9, 2026 •

edited

Loading