34517 - [regression] XIM: Duplicate chars are committed to the edit area

Status: VERIFIED FIXED

(Core :: Internationalization, defect, P3)

Description

[ji]

Build: 2000040412-M15 Linux build

When entering JPN chars into the edit area, duplicate chars are committed.

Steps of reproduce:
1. Launch mozilla and bring up Composer.
2. Turn on Japamese IME by pressing Shift + <space>
3. Enter a Japanese character and hit Return key to commit the char
   to the edit area. You'll see two exactly same Japanese characters are 
   committed.

Comment 1

[ji]

Changed Component to i18n.

Comment 2

[Frank Tang]

reassign to toshi.

Comment 3

[Masaki Katakai]

should be closed as duplicated.

I've already filed this problem as 33847. It seems that
25452 is reproducible again. I've put a code patch.

Comment 4

[bobj]

*** This bug has been marked as a duplicate of 33847 ***

Comment 5

[ji]

Although 33847 is verified as WORKSFORME, the problem described here still exists with the
latest build (2000041109 linux commercial).
Reopened the bug.

Comment 6

[Teruko Kobayashi]

Added bobj in cc.

Comment 7

[bobj]

This is a dogfood regression from Beta1 which makes composer unusable for
Japanese.  Reassigning to jfrancis for now.

[e-mail] or [e-mail],
  Was the patch for bug 25452, supposed to fix both the Del key handling
and the duplicate committing of chars described in this bug?

Comment 8

[Masaki Katakai]

Yes, the patch of 33847 can solve both. I've verified just now with the

current tree and the patch.

Comment 9

[bobj]

I did a quick scan on lxr.mozilla.org
  http://lxr.mozilla.org/seamonkey/source/editor/base/nsTextEditRules.cpp

and it looks like katakai's patch for bug 25452 had been applied (adding 
PRInt32 aAction to WillInsertText + some code for IME).

Has something else broken since the patch?

Comment 10

[Joe Francis]

i dont believe this is editor related.  Frank, can you find the right person to 
assign linux related ime problems too?

Comment 11

[Masaki Katakai]

I checked, but the patch (of 33847) has not been applied yet.
JoeInsertText() needs to look action, but doesn't. JoeInsertText()
should reset the IMEtext when text=null & action = KInsertTextIME,
but just returns. Please try the patch, the problem can be solved.

Comment 12

[bobj]

jfrancis,
Does this make sense to bounce back to you to apply the patch from bug 33847?
    http://bugzilla.mozilla.org/showattachment.cgi?attach_id=7208

Can we still get this into the M15 branch?  Without it, Japanese editing and
mail composing is not usable for M15.

Comment 13

[Joe Francis]

looking into this...

Comment 14

[Joe Francis]

I *think* I may have fixed this, but I don't have a linux ime box.  Please check 
on tomorrows build.

Comment 15

[Hidetoshi Tajima]

Attachment: fix to prevent duplicated IME commit text.

Comment 16

[Hidetoshi Tajima]

Not yet. The bug is still there. Joe, please work
with, Bob.Jung, Frank.Tang and Teruko to double-check it. It is really easy to
reproduce the problem.

Also, please regard the patch I'm now attaching this
bug. This is exactly the same as Masaki.Katakai attached to 
http://bugzilla.mozilla.org/show_bug.cgi?id=33847 ten
days ago and since then, he has asked you to evaluate
a couple of times already. Please also note, I just 
applied the patch on top of your today's fix, and I'm
seeing the problem gets fixed.

Comment 17

[Hidetoshi Tajima]

Reopened it as I'm seeing this with the latest
check-out.

Comment 18

[Joe Francis]

I thought I had spoken with Masaki Katakai before about the patch.  I didn't want 
to apply it because I was in the middle of modifiying the same code, and thought 
that my modifications would fix the same problems.  I've examined the patch 
(again) and I don't understand yet why it fixes the problem.  It is very similar 
to changes I have recently checked in.

I would like to know the sequence of events that the editor receives when this 
happens on Linux.  Even if we can fix the symptoms by applying Masaki Katakai's 
patch, it would be good to understand the problem.  Perhaps Linux IME is doing 
something it shouldn't (much as Winows IME was sending duplicate events for every 
IME transaction).

Comment 19

[Joe Francis]

assigning to me

Comment 20

[Joe Francis]

looking at the differences between my changes and the patch, I think the only way 
this problem could be happening is if Linux IME is sending all IME commits to the 
editor twice, once as an IME commit, and then once as a normal non IME action.  
If this is so then we need to fix Linux IME.

Frank Tang is going to update his Linux IME build and we will look at what's 
going on early next week.  Meanwhile if anyone reading this has a current MLinux 
IME build and can get stack traces, try setting a breakpoint at the start of 
nsEditor::InsertTextImpl() and see how many times it gets called when you do an 
IME commit, and what the stack looks like each time.  That data would be very 
useful.

Comment 21

[Hidetoshi Tajima]

My apology. When I built and tested my buuild yesteday, I did not fairly run
cvs checkout onto editor/base repositry. I did this morning again and checked
out yesterday's joe's check-in to run the build again. Now I verified the
problem is fixed with it. Yes, it does the similar thing as Masaki's patch,
which was posted last week.

Comment 22

[Joe Francis]

ok, thanks for checking.  Marking fixed again...

Comment 23

[Teruko Kobayashi]

I verified this in 2000041808 M16 Linux build, not 2000041805 M15 Linux build.

Comment 24

[nhottanscp]

*** Bug 36887 has been marked as a duplicate of this bug. ***