Decorations v3#459
Draft
amadeus wants to merge 131 commits into
Draft
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
amadeus
force-pushed
the
amadeus/decorations-v3
branch
from
March 31, 2026 21:55
c69f24b to
c710e4e
Compare
amadeus
force-pushed
the
amadeus/decorations-v3
branch
from
April 2, 2026 20:21
6caaa7c to
69429b4
Compare
amadeus
force-pushed
the
amadeus/decorations-v3
branch
from
April 3, 2026 01:57
69429b4 to
70ea1c9
Compare
amadeus
force-pushed
the
amadeus/decorations-v3
branch
from
April 3, 2026 18:18
70ea1c9 to
ed0b818
Compare
amadeus
force-pushed
the
amadeus/decorations-v3
branch
from
April 4, 2026 01:48
ed0b818 to
5f13123
Compare
amadeus
force-pushed
the
amadeus/decorations-v3
branch
from
April 6, 2026 21:13
5f13123 to
591a6ba
Compare
amadeus
force-pushed
the
amadeus/decorations-v3
branch
from
April 6, 2026 22:20
591a6ba to
85a1d55
Compare
amadeus
force-pushed
the
amadeus/decorations-v3
branch
from
April 6, 2026 22:36
85a1d55 to
3c0ed60
Compare
…uard usage (#914) Move the monorepo browser baseline to browsers with native Intl.Segmenter support, which the diffs editor relies on. Firefox 125 is the binding constraint (Chrome 87+, Edge 87+, Safari 14.1+ are already covered), so bump firefox >= 120 to firefox >= 125 in .browserslistrc. Make the editor's Intl.Segmenter usage safe at import time so importing the editor never throws on engines without it: - Add createSegmenter() (feature-detected) and a lazily-cached getGraphemeSegmenter() in editor/utils.ts, replacing the module top-level grapheme segmenter in selection.ts. - Guard every segmenter call site (word + grapheme in selection.ts, two grapheme in textMeasure.ts) with a minimal code-point/regex fallback for engines lacking Intl.Segmenter. Behavior is unchanged on supported browsers. Co-authored-by: Claude <noreply@anthropic.com>
* rename * fix height
To reproduce, in the diff editor on Safari: 1. Click into the additions side. 2. Select all (cmd+a), then press Backspace or Delete. Nothing is deleted — the additions can't be cleared — though it works in Chrome. The editor paints selections with an overlay and leaves the native selection a collapsed caret. After select-all the caret sits at the document start, and Safari fires no delete beforeinput when nothing precedes the caret, so the delete is dropped; Chrome fires it anyway and deletes via the editor's own selection. Give WebKit a non-collapsed native selection so it emits the delete; the edit still runs on #selections, which spans the whole document. Only rendered lines resolve to DOM nodes and the diff is virtualized, so anchor the native range to the editable lines currently on screen rather than the document bounds — otherwise it stays offscreen in a long file and Safari drops the delete again. A one-shot guard keeps the resulting selectionchange from reading that shorter range back over #selections; the next keydown or pointer press (mouse, touch, or pen) clears it.
* Misc fixes * Stable updates
* [diffs/editor] Support `moveLineUp` and `moveLineDown` commands * fix "Clamp moved selections at EOF" * fix performance: 60ms -> 0.3ms (m4 pro)
1. Open a diff editor with an annotation on an added line. 2. Delete that line, or delete the newline immediately above it. 3. Undo and redo the edit, or insert another line above it. 4. The annotation disappears or fails to return with history. The annotation mapper only used the aggregate line delta for an edit batch. That made it guess deleted ranges, drop annotations on deleted lines, and skip net-zero batches where one edit added lines and another removed them. Record per-edit line changes on TextDocumentChange and map annotations through each line edit in order. Keep annotations by moving them to the nearest surviving line, and refresh touched annotation lists so undo and redo restore the right annotation state.
A user can trigger the lag like this: 1. Focus an editable diff. 2. Hold Cmd+V or Ctrl+V with text on the clipboard. 3. The editor accepts each repeated paste keydown. 4. Each paste can add lines and make the page fall behind. The root cause is that a held paste key produces repeat keydown events. Each accepted repeat can lead to another paste and another expensive diff update. For multi-line text, that update can recompute diff state from the full edited document. Suppress repeated paste keydowns and let the first keydown keep the existing behavior. This keeps normal paste, including custom clipboard providers, while blocking the runaway repeat path.
* Phase 1: Add basic CodeView accounting for header/footer * DOM reconcilation stuff * Various math-ing stuff * Lots of verbose comments by claude * Phase 2: Resize observer/measurement handling Cleaned up some slop from earlier on the initial measure, but then also added the ResizeObserver stuff * Phase 3: Start fleshing out vanilla JS apis * Phase 4: The React APIs Took a bit of back and forth from the robot to get on the right solution that worked within the same confines of before. * Misc fixes * Prevent scroll fixing if scrolled to the top and a header component is added lazily * Scroll to should take the header into account * demo: Add header/footer methods to codeview tests * tests: add CodeView header/footer tests * type -> kind Dunno why, i just prefer that... * review: ensure height is cleaned up properly * React re-rendering fixes for react compiler * PR review * Fix bugs with missing llms.txt content * First pass at content for headers/footers
* Slop phase 1 * Slop phase 2 * Phase 1: Adding types for codeview editor support * Phase 2.1: Add recycling support to editor * Phase 2.n: The rest of Phase 2's functionality for vanilla JS etc * Phase 3: React API tweaks * Phase 4: CodeView edit tests * Phase 5, Part 1: Fix a variety of issues with the initial CodeView implemention * First pass at getting a somewhat editable flow in virtualizer/codeview * Fix WorkerPool related settings * Fix up annotations in the new views * Fix virtualizer strict mode bug Also adds a test for it * Add window/element virtualizer variants It's important to test both versions * PR review fix * PR Review Fixes: Fix some virtualization invalidation issues with Files * PR Review Fixes 2: Fix scenarios where a controlled element is removed from CodeView to ensure it properly fires onItemEditComplete * Removing some useless ai comments, & code polish * Fix expandUnchanged jumps that can occur during edit * Fix a virtualizer out of screen height issue * fix(diffs/editor): Exclude virtualized host offset from caret scroll-margin .getScrollMargin folded `fileInstance.top` into the primary caret's scroll-margin-top. For the original non-virtualized editor that value was always 0, but on virtualized hosts (VirtualizedFile/VirtualizedFileDiff, CodeView items) `top` is the item's offset in scroll content — a coordinate, not a margin. Once scrolled down a list, the caret carried a scroll-margin-top of thousands of pixels, so every caret reveal (`scrollIntoView` with block: 'nearest') treated the caret as an oversized box and mis-scrolled — typically pinning the line being edited to the bottom of the scroll view, or suppressing the reveal entirely. The margin now only reserves real viewport space above/beside the caret (search panel height, gutter width), with a comment explaining why the host offset must never be folded back in.
* AI slopped deferred updates * Mb optimize nothing... Fix tests * Use CodeViewHandle for scroll tubing * Add a regression test for stale pooled headers * Add a regression test for stale slots
* Remove unused file * Update edit mode feature grid * more copy edits and mention codeview
* New local input component, update copy and functionality, widen dropdown * Misc input polish * Disable 1password bullshits * Fix bg color in light theme * Tweak messaging to say saved only in localStorage * Access Token Tweaks * Make them read only for now (since we don't support reading/writing comments) * Improve contents/descriptions * Tweak to green border on active token --------- Co-authored-by: Amadeus Demarzi <amadeusdemarzi@gmail.com>
mostly a non-functional changes, a lot of types changes, that I had to refactor from the AI slop because he really did some ugly shit...
Some AI slop here for sure, but i think we wrangled it into a decent spot.
shit's mb about to get weird...
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
WIP