Skip to content

Note geometry cache retains the previous hunk when only the resolved target changes #1346

Open

Description

@fernandezbaptiste

Hullo folks,

I’m doing some research for Tessl Code Review and ran it against your repo. It flagged a potential issue, and I figured it made sense to share it with you:

What happened?

When a note's resolved target moves to another hunk while its annotation and thread metadata stay the same, measureDiffSectionGeometry returns the previous cached geometry. Its noteTarget.hunkIndex still points to the old hunk.

This surfaced during a retrospective review of #1202. It was identified as a pre-existing case, not a regression introduced by that PR. I reproduced it on current commit 57a90b4.

Steps to reproduce

Using the checkout's createTestDiffFile helper, make a 40-line file with changes at lines 5 and 35, giving two hunks. Build two notes with createVisibleAgentNote, keeping these fields identical:

const base = {
  id: "stored-note",
  annotation: { summary: "Same note" },
  source: "user" as const,
  thread: { noteId: "note-1", depth: 0 },
};

Supply target: { hunkIndex: 0, side: "new", line: 5 } for the first note and target: { hunkIndex: 1, side: "new", line: 35 } for the second.

Call measureDiffSectionGeometry(file, "split", true, theme, [note], 120) first with the original note, then with the moved note, retaining the same file object.

Observed:

resolved anchor owner:     0 -> 1
returned geometry object:  same object
cached note target:        hunkIndex 0
fresh-file control:        hunkIndex 1

The control passes { ...file } with the moved note, forcing a fresh WeakMap entry. All other inputs stay the same.

Expected behavior

Changing the resolved anchor should invalidate note-aware geometry so bounds and keyboard note targets refer to its current placement.

notesCacheKey includes annotation/thread fields but omits VisibleAgentNote.anchor, although row planning uses that resolved placement.

Version

Source checkout 57a90b4 (package 0.25.0), macOS, Bun 1.3.10. This was a direct geometry-module reproduction with the actual imports and repository helper, not an interactive TUI/navigation test.

Full transparency: I work at Tessl, and we've spotted that using Tessl Code Review - using repository-level context and specific review lenses that improve over time through feedback. That’s how this finding surfaced.

We’re just getting started and want to offer the tool free to open-source projects. If this is useful, happy to give you access.

Bap

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions