336365 - Patch from 332579 breaks Unicode keyboard input

Status: RESOLVED FIXED

(Core Graveyard :: Widget: Mac, defect)

Description

[Mike Shaver (:shaver -- probably not reading bugmail closely)]

With U.S. Extended, I cannot type letters or numbers.  Also, keyboard shortcuts don't so much with the working.

With U.S., I am free to express myself alphanumerically.

Comment 1

[Mark Mentovai]

The TSM handlers need to be installed after the catch-all transitional handler.  Updated patch in bug 332579.

Comment 2

[Mark Mentovai]

Related-ish problems still present:

Some double-double-processing of events is back again.  Testcase: enter javascript:alert("alert"); in the URL bar and press return.  Expect the alert sheet to be dismissed.  It is, but it's replaced by another one immediately.  (Two return events were processed, possibly one by the TSM ae handler and the other by the MacPump ce handler - is the ae handler not swallowing when it should?)

-----

Press Escape to dismiss the sheet for real.  Now try to close the window with command-W (a Unicode keyboard layout must be selected like US Extended).  The window seems to have other plans.  You get this output:

###!!! ASSERTION: nsMacEventHandler::ConvertKeyEventToUnicode: UpgradeScriptInfoToTextEncoding failed.: 'err == noErr', file /Users/mark lizard/firefox-1503/mozilla/widget/src/mac/nsMacEventHandler.cpp, line 977
###!!! ASSERTION: nsMacEventHandler::InitializeKeyEvent: ConvertKeyEventToUnicode returned 0.: '0 != aKeyEvent.charCode', file /Users/mark/lizard/firefox-1503/mozilla/widget/src/mac/nsMacEventHandler.cpp, line 864

When the sheet is up, a new appshell is running the loop.  When that appshell disappears, it takes the TSM singleton out with it.  Something needs to enforce that only the root MacPump owns the TSM singleton (possibly by making the MacPump itself a singleton.)  Alternatively, we could just leak the TSM singleton - that's what the old code did.

Comment 3

[Mark Mentovai]

Comment 2 fixed in bug 332579v8.

Comment 4

[Mark Mentovai]

Comment 0 and comment 2 part 2 are fixed.  Comment 2 part 1 is worked around, but the real problem is bug 337199.