In order to reproduce just select any cell in any table in VisualEditor and start typing within IME. Input should appear inside the cell but instead it does not appear at all.
I investigated it a little bit: normally (not IME) when user starts typing method handleInsertion is called (via keypress handler) and that method change table cell selection to text selection but for IME input it does not happen (because there is no keypress).
| Subject | Repo | Branch | Lines +/- | |
|---|---|---|---|---|
| Delete cross-branchNode selections etc. on compositionstart | VisualEditor/VisualEditor | master | +26 -1 |
This is a quick and dirty solution: http://pastebin.com/PQr3uZ5S. At this moment it copies code from handleInsertion but it could be just extracted to a method.
@dchan, @Esanders - does it seem like a right solution?
Inez, now that we've fixed some timing issues I think this solution may be safe - I'll experiment.
Hah, contrary to my prediction, with this change I can consistently *segfault* Firefox on Ubuntu by clicking a table cell and typing a backtick (which in the UK extended keyboard layout is produced by pressing a deadkey twice, the second press of which causes a compositionstart event). I'll investigate further but I thought this was too good not to share :-)
The crashing behaviour seems similar to this minimal test case http://jsfiddle.net/s1ryfmp5/ .
That jsfiddle uses an alert, but something in our code is producing the same effect. It seems IME-sensitive: it's crashing for me on Ubuntu 15.10 but not for @Esanders on Ubuntu 14.04 . Continuing to investigate (and will fire upstream bugs as appropriate when done).
It seems there's a problem specific to IME listeners in Firefox:
I still don't know what VE is doing that's analogous to the minimal test case's alert() calls, but I suspect it's to do with changing the selection.
The problem as it affects VE amounts to doing blur();focus() in a compositionstart listener. See https://bugzilla.mozilla.org/show_bug.cgi?id=1230473 and http://jsfiddle.net/6tsyptoh/2/ .
See also https://bugzilla.mozilla.org/show_bug.cgi?id=1230520 and http://jsfiddle.net/rkprkmbs/1/ , which show further issues with compositionend listeners that might bite us here if the segfault bug is fixed.
Change 264273 had a related patch set uploaded (by Divec):
Delete cross-branchNode selections etc. on compositionstart
Change 264273 has a fix for this bug, but with the fix disabled for Firefox because of https://bugzilla.mozilla.org/show_bug.cgi?id=1230473 .
Change 264273 merged by jenkins-bot:
Delete cross-branchNode selections etc. on compositionstart