Label thumbnails by the keys that pick them; never diff binary #8

Merged

cyplo merged 1 commit from image-labels-and-binary-diff into main 2026-11-09 14:37:15 +00:00

Conversation 0 Commits 1 Files changed 19 +202 -66

cyplo commented 2026-11-09 14:34:16 +00:00

Owner

• Copy link
• Copy Markdown

Two problems you spotted on the real screen, both fixed:

1. Thumbnail labels contradicted the options line

A single copy was labelled [a] - but for one conflict the options line says 2: Keep conflict and no letter key does anything. The labels now name the keys that pick each copy:

• one copy: original | conflict (matching 1: Keep original 2: Keep conflict)
• several copies: original | [a] | [b] (matching ab: Pick conflict)

And copies with identical bytes share one thumbnail with their labels combined - original, [a] - since they are the same picture and need no second column. The original always appears; a copy identical to it joins its column rather than duplicating it.

2. Binary bytes were diffed as text

Expanding a BMP conflict rendered escape soup across the screen: BMP bytes are valid UTF-8 (with NULs), so the text diff tried anyway. Binary content - NUL bytes or non-UTF-8 - now gets no text diff at all, and stays metadata-only like the file-type table says.

Pinned by fixtures

• image/identical - identical pair expands to one thumbnail labelled original, conflict
• image/identical_among_three (new) - [a] byte-identical to the original shares its column (original, [a]), [b] stays alone
• image/corrupt_png, image/dimensions, image/narrow_six_cols - single-copy labels now conflict (with the narrow one cut to con)
• image/bmp_stays_plain - expanded BMP shows no diff garbage and no image treatment; the two identical fixture JPEGs also got their 8-bit SOF precision fix so their dimensions parse (they had the same 12-bit header as the earlier ones)

Checked

• 71/71 tests pass, clippy clean with -D warnings.
• Mutating this PR's whole diff against main, as CI will: 12 mutants tested, 11 caught, 1 unviable, 0 missed.
• Rendered the screens as PNGs and looked at them: original, [a] shares one red thumbnail, [b] sits alone in its column.

Two problems you spotted on the real screen, both fixed: ### 1. Thumbnail labels contradicted the options line A single copy was labelled `[a]` - but for one conflict the options line says `2: Keep conflict` and no letter key does anything. The labels now name **the keys that pick each copy**: - one copy: `original` | `conflict` (matching `1: Keep original 2: Keep conflict`) - several copies: `original` | `[a]` | `[b]` (matching `ab: Pick conflict`) And copies with **identical bytes share one thumbnail** with their labels combined - `original, [a]` - since they are the same picture and need no second column. The original always appears; a copy identical to it joins its column rather than duplicating it. ### 2. Binary bytes were diffed as text Expanding a BMP conflict rendered escape soup across the screen: BMP bytes are valid UTF-8 (with NULs), so the text diff tried anyway. Binary content - NUL bytes or non-UTF-8 - now gets no text diff at all, and stays metadata-only like the file-type table says. ### Pinned by fixtures - `image/identical` - identical pair expands to one thumbnail labelled `original, conflict` - `image/identical_among_three` (new) - `[a]` byte-identical to the original shares its column (`original, [a]`), `[b]` stays alone - `image/corrupt_png`, `image/dimensions`, `image/narrow_six_cols` - single-copy labels now `conflict` (with the narrow one cut to `con`) - `image/bmp_stays_plain` - expanded BMP shows no diff garbage and no image treatment; the two identical fixture JPEGs also got their 8-bit SOF precision fix so their dimensions parse (they had the same 12-bit header as the earlier ones) ### Checked - 71/71 tests pass, clippy clean with `-D warnings`. - Mutating this PR's whole diff against `main`, as CI will: **12 mutants tested, 11 caught, 1 unviable, 0 missed**. - Rendered the screens as PNGs and looked at them: `original, [a]` shares one red thumbnail, `[b]` sits alone in its column.

cyplo added 1 commit 2026-11-09 14:34:17 +00:00

Label thumbnails by the keys that pick them; never diff binary ... 6432b76b8a

Two problems from looking at the real screen. A single copy was
labelled [a] while the options say '2: Keep conflict' and no letter key
does anything there: thumbnails now name the keys that pick each copy -
original, conflict for one copy, [a], [b] where letters really pick -
and copies with identical bytes share one thumbnail with the labels
combined, like 'original, [a]', since they are the same picture to
compare.

An expanded binary conflict showed its bytes diffed as text: BMP bytes
are valid UTF-8 with NULs in them, so the diff rendered escape soup
across the screen. Binary content - NUL bytes or non-UTF-8 - now gets
no text diff at all, and stays metadata-only as the file-type table
says.

Co-Authored-By: GLM 5.3 <noreply@anthropic.com>

cyplo merged commit 6432b76b8a into main 2026-11-09 14:37:15 +00:00

cyplo deleted branch image-labels-and-binary-diff 2026-11-09 14:37:15 +00:00

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

cyplo/deescalate!8

No description provided.