Repository navigation
Conversation
The overlay covers every display from a band above the normal one, so a Chat window was behind the sprite layer for its whole life. Settings has taken this raise since #803; Chat now takes the same one. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
Closing. The change in here does not fix #891, and the branch stays if any of it is wanted back. Two arms were captured with identical binaries differing only in the Leaving this open would park a pull request whose title claims a fix it does not make, next to #900, which is the live candidate. What is in here, for whoever wants it later:
#891 carries the corrections, including the two hypotheses this work ruled out and the one bullet in the issue body that wrongly discouraged the — Claude Code agent, on @omesser's behalf. |
Why
The Chat composer drew no text caret, so typing and editing were blind.
The overlay covers every display from a window band above the normal one, and a
Chat window opened into the normal band — so it sat behind the sprite layer for
its whole life. Settings hit that and was raised out of it in #803; Chat never
was. That reading of the source is still true. It is not what causes the
missing caret.
What changed
raise_settings_windowkeeps its body and loses its Settings-only name:raise_above_overlay.open_chatcalls it afterset_focus, asshow_settingsdoes.
AI_BUDDY_OPEN_CHATjoins theAI_BUDDY_OPEN_SETTINGShook beside it, soa sitting can reach the composer without a synthetic tray click.
How to verify
scripts/verify-chat-caret-macos.sh, twice: once on this branch, once on a buildwith only the
raise_above_overlaycall removed. Same script, same rects, sameburst timing, same scratch
HOME.differing=12/15 columns=4 rows=31 span=28..31caret=truediffering=12/15 columns=4 rows=31 span=28..31caret=trueIdentical. The caret is there without the raise, so the raise is not what lights
it. The two binaries did differ — the baseline carries no
chat: … raise:formatstring and this one does.
cargo test -p ai-buddy --all-targets521 passed,cargo clippy --all-targets -- -D warningsclean,node --test tests/*.test.js291 passed.Notes
before capturing, and an activated window paints a caret whatever else is
wrong. So it would pass on the broken build too. It is not yet a regression
guard, and the two rows above are the evidence for that, not against it.
byte-identical had the Settings window overlapping the composer rect, so it was
a still picture of Settings, not of a caretless composer. It proved nothing and
should not be cited.
AXSelectedTextRangewhetheror not one is painted, which is why this was filed against a window every AX
assertion called healthy. Note
key=falsein the script's output: that pairreads the same in the healthy state, so it diagnoses nothing. An earlier
revision of this description leaned on it; that was wrong.
activation from a script. That wants a burst taken without activating, which is
less intrusive than what ran here.
Refs #891
— Claude Code agent, on @omesser's behalf.
🤖 Generated with Claude Code