52142 - existence of XIM will crash mozilla in startup

Status: VERIFIED FIXED

(Core :: Internationalization, defect, P2)

Description

[Shanjian Li]

Login through CDE and choose a japanese locale. When system prompt for IM, select XIM or ATOK.
You will hit a bus error before mozilla bring up any window.

Comment 1

[Shanjian Li]

call stack shows the crash happens inside gdk call: gdk_im_begin.
I am wondering if this is a gtk/gdk problem on HPUX. I need to find 
a verification application to verify this.

Comment 2

[Shanjian Li]

To be more specific, this crash happen when using on-the-spot input method
style. As suggested by toshi, putting 
user_pref("xin.input_style", "over-the-spot"); 
into prefs.js will provide a temporary workaround. Lower its severity.

Comment 3

[Shanjian Li]

We have 4 kinds of input style available on X, They are "on-the-spot",
"over-the-spot", "separate", "none". Input style can be further defined
using preedit style and status style. From my testing, "over-the-spot" is
the only input style working on HP at this time. ( I did not test various
combination of preedit and status style.)

Since "on-the-spot" is the default IME style, I would suggest to check in
following fix to stop crash:

Index: nsGtkIMEHelper.cpp
===================================================================
RCS file: /cvsroot/mozilla/widget/src/gtk/nsGtkIMEHelper.cpp,v
retrieving revision 1.14
diff -r1.14 nsGtkIMEHelper.cpp
1150a1151,1167
> #ifdef HPUX
>   prefered_preedit_style = (GdkIMStyle) GDK_IM_PREEDIT_POSITION;
>   prefered_status_style = (GdkIMStyle) GDK_IM_STATUS_NOTHING;
>   style = gdk_im_decide_style((GdkIMStyle)(prefered_preedit_style |
prefered_status_style));
>   if (style) {
>     gInputStyle = style;
>   } else {
>     style = gdk_im_decide_style((GdkIMStyle) (SUPPORTED_PREEDIT |
SUPPORTED_STATUS));
>     if (style) {
>       gInputStyle = style;
>     } else {
>       gInputStyle = (GdkIMStyle)(GDK_IM_PREEDIT_NONE|GDK_IM_STATUS_NONE);
>     }
>   }
>   return gInputStyle;
> #endif
>

Comment 4

[Shanjian Li]

nominate nsbeta3

Comment 5

[Frank Tang]

add tajima san to the cc list
shanjian- please provide context diff. no one know where you want to put that 
part of code into the file.
tajima- please review the code. 
shanjian- you need to follow the new review rule. Read tinderbox for details.

Comment 6

[Hidetoshi Tajima]

Shanjian's patch seems to be okay as far as whole codes are ifdefed for hpux
only and hp agrees that on-the-spot does not work well with their XIMs. Another
comment is, GDK_IM_STATUS_AREA might be preferable to GDK_IM_STATUS_NOTHING
if that combination works.

Comment 7

[Frank Tang]

nsbeta3+ since this is low risk for other platform. Please do the proper new 
code review and approval process before check in.

Comment 8

[Teruko Kobayashi]

Just comfirm the bug.

Comment 9

[Frank Tang]

This is a P2 for HP-UX. 

Comment 10

[Shanjian Li]

Attachment: A better context diff

Comment 11

[Hidetoshi Tajima]

Shanjian, I'm reviewing the proposed patch, but it seems to me that it
should be enough to change primary inputstyle to over-the-spot
for hpux, since the rest of the codes inside ifdef HPUX is duplicated to
the original. I'm going to attach a new code diff, which I tested on linux
and Solaris.

Comment 12

[Hidetoshi Tajima]

Attachment: Use over-the-spot as default input style - hpux only.

Comment 13

[Shanjian Li]

fix has been checked in.

Comment 14

[Hidetoshi Tajima]

Shanjian, you seemed to check in your original.
I'm fine if you have checked mine and then decide to
use yours. But just for sure.

Comment 15

[Shanjian Li]

toshi, I didn't noticed your comment when I did the check in. Since it does not 
make much difference, I will stick to mine. I am still working on on-the-spot
problem. If that can be resolve, those fix will be backed out. Otherwise, I may 
try the combination as you suggested and finalize it after PR3.

Comment 16

[Hidetoshi Tajima]

shanjian, no problem.

Comment 17

[Teruko Kobayashi]

Verified.