Fixed
Status Update
No update yet.
Comments
kl...@google.com
ap...@google.com
Thanks! This looks like a bug, thank you for reporting!
na...@google.com
I would also recommend setting the value to 10%. When you do that, the behavior is wild.
Dragging up to 10% or between 50% and 90% will cause it to snap back to the start anchor.
Dragging between 10% and 50% or past 90% will cause it to snap to the end anchor.
Dragging up to 10% or between 50% and 90% will cause it to snap back to the start anchor.
Dragging between 10% and 50% or past 90% will cause it to snap to the end anchor.
Description
We have tests for things like that certain types of text changes don't cause too many recompositions. Those tests pass when sending commands through the IME command code path, but fail when going through the semantics code path. The semantics code path is very different than IME: we just directly invoke the
onValueChangedcallback.This is an opportunity for other bugs as well – internal state can get out of sync because one way of mutating it updates internal state directly, and the other asks the composition to update our state on the next frame.
Semantics actions should go through the same code paths as IME commands, or as close as possible.
For example, the
TextFieldTests.textField_stringOverload_doesNotCallOnValueChange_whenCompositionUpdatesOnlytest fails if text clearance is performed via theSetTextaction instead of IME commands. They should behave the same.