don't screenshot dictation placeholders as pasted images (#10313)

This commit is contained in:
Spence Pope
2026-04-21 02:35:22 -04:00
committed by GitHub
parent 985129dd34
commit 3931b90818
2 changed files with 158 additions and 0 deletions
+22
View File
@@ -0,0 +1,22 @@
# Expo Paste Input Patch
`expo-paste-input` observes `UITextView.textDidChangeNotification` and treats any
`NSTextAttachment` in the text view's `attributedText` as a pasted image. When
it can't find a real image payload on an attachment, it falls back to
`image(forBounds:)` and, failing that, to a `drawHierarchy` screenshot of the
text view at the attachment's glyph rect.
iOS Dictation inserts its own `NSTextAttachment` (the shimmer/cursor indicator)
into the text view during dictation. Those attachments don't carry real image
data, so the fallbacks would fire — emitting a zoomed-in screenshot of the
composer as if the user had pasted an image at the end of dictation.
This patch:
- Removes the `image(forBounds:)` and `renderTextAttachment` fallbacks in
`extractMediaPayload` so the library only accepts attachments carrying a real
payload (`fileWrapper`, `contents`, or `image`).
- Only sanitizes (deletes) attachment ranges that produced a payload, and
skips the "unsupported" toast when an attachment has no payload. Unknown
system attachments like the dictation placeholder are left alone rather
than being ripped out from under iOS.