Created attachment 179129 [details] [review]
possible fix

Created attachment 179132 [details] [review]
Debug file with an oo selection

Joanie, your patch is very good, my OO-3.2 is working again, for example OO not crashing if I not saving a file and closing the OO Writer.
I see now only one problem when testing (ten minute time period), this problem is not happening if the patch is not applied):
Orca doesn't spokening selected state if I selecting a text for example with Shift+Down arrow key (selected message with end of text spokening).
If I unselecting the text, this is right spokening. Where am I and detailed where am I right presenting the text state (selected/unselected).
Reproducation steps to verify this:
1. Write two line with OO Writer.
2. Go to top of the file and press two time the Shift+Down arrow key. You will be not hear the "selected" message with end of line spokening with you selected.
If you unselecting the selected lines with Shift+Up arrow key, Orca right spokening the "unselected" text with end of text spokening.
You confirming this?

Attila

(In reply to comment #2)

> If you unselecting the selected lines with Shift+Up arrow key, Orca right
> spokening the "unselected" text with end of text spokening.
> You confirming this?

I'm not seeing that. (Though I see it in your debug.out.)

What I personally am seeing, but only with OOo 3.2 is that Orca is saying "unselected selected" sometimes. In other words, it's ultimately getting around to saying the correct thing after first saying the incorrect thing. Neither that, or what you describe, is all that surprising to me given the nature of the patch. Basically before (without the patch) we sat around queuing up events and by the time we got around to processing them, things like object state might have been updated. With my patch, we're processing things *now*, *right away* if they come from OOo. The state we expect might not have changed yet.

So what I would suggest is this: please keep testing it and see what other strange things you discover. And we'll wait and see what others in the community say. Once I have a better picture over just what all is being impacted and where, I'll be in a better position to adjust.

Make sense?

Adding myself to the list.

This patch is beautiful! I have tried two different documents which contain one or more lists (both numberred and bullet-ed) and no crashes.  Also I should observe that I have not encountered any real delays in speech either.  Maybe a slight delay when reading the first item of a list when enterring the list but on my 1.6 GHz machine, I cannot really report any performance degradations while testing with the documents I was working with.  So far, I have not encountered any crashes upon exit from Writer or Calc but that is more intermittent so not sure what the future may bring there.

Thanks Steve!

Please keep using it and look for weird things, like Attila found w.r.t. text selection. By the way, what are you seeing regarding text selection and this patch?

(In reply to comment #6)
> Thanks Steve!
> 
> Please keep using it and look for weird things, like Attila found w.r.t. text
> selection. By the way, what are you seeing regarding text selection and this
> patch?

So far, I only heard the word "Selected" spoken twice at the end of each shifted key movement.  I never heard any "Unselected" spoken during my movements except for when I would stop selecting text.  Other than that, I seemed to have clean behavior with text selection.  I had a bit of trouble getting to select the bullet or item number to be included right but that might be just how OOo does things.  I got things to work better when I forced the cursor to the left most position of the line so I didn't consider this to be a bug.  The last line I selected always seemed to be the number of the next line.  What I mean here is if I'm shift down arrowing along, I would end up hearing the last line followed by the item number of the next and then stop.  But I think that might have to do with numbered lists and cursor positioning.  Again, this may be the way the beast works; I personally wouldn't wory about it unless this is truely bad behavior.

Sorry, I forgot both times to state that I am using the 3.4 dev version of OOo.  I haven't tried LibreOffice yet.

I've been using the patch since it was attached to this report and I have not suffered one crash of openoffice writer up to now (NOTE: I only state openoffice writer because that is what I have been using, not because the other products show problems). Also I haven't really noticed any bad side affects. This is using openoffice 3.3.0.

(In reply to comment #1)
> Created an attachment (id=179129) [details] [review]
> possible fix

Before applying the patch, the crash has occured around one minute after the opening of OO. For this reason, after using it for two days without any bad event, I regard the problem is solved. I tested only Writer. I have created numberings, tables, fields etc.

the most important test was to create a table and to navigate rapidly with arrows over the table. Before the patch, this was the esiest way to provoke the crash. For two days, I have been using continuously without any problem.

As I have identified the same problem in Calc, now I will test it to confirm the patch has also corrected crashes in Calc.

I am very happy. Although there are some inaccessible features in Open Office, as a whole, the accessibility is very good. Therefore, this patch is a gift.

Comment on attachment 179129 [details] [review]
possible fix

http://git.gnome.org/browse/orca/commit/?id=627e3b4a41b8f46e88a8e534786eed4835e8bdf4

Thanks everyone for the testing!

Given the positive results reported here, the minor nature of the change with respect to the Orca code, and the fact that the change I made can be undone by anyone who wants to via their orca-customizations.py.... I've committed the patch to master. That should get it additional testing and feedback long before we hit the GNOME 3.0 code freeze.

Created attachment 180387 [details]
Debug file with show Libreoffice 3.3 result

Joanie, I am very surprising this bug is resolved, I not get prewious bugreport e-mails unfortunately, I don't no why. :-):-)

I see unfortunately my wrote problem with Libreoffice 3.3 version with my Ubuntu Lucid system. The Orca is full uptodated. Not happening crashes, but for example when I select two lines with Shift+Down arrow key, I not hear the "Selected" message. When I press Shift+Up arrow key one time, I hear one the "unselected" message with end of the text spokening. But when I press again Shift+Up arrow key and remove the first selected line selection, I hear two time the "unselected" message.

I see unfortunately this similar result with Openoffice.org 3.2 version with native awailable with Ubuntu Lucid.
Anything I doing wrong, or my system is too old for Orca 2.91?

Reproducation steps, hopefuly my results is not confirmed:
1. Launch Openoffice.org or Libreoffice.org writer.
2. Write following two line:
apple
potato
Press enter key after the second line.
3. Press two up arrow key to jump the first line.
4. Press two Shift+Down arrow combination and listening the spokened output. If you not hear the "selected" message with end of text spokening, the first problem is reproduced.
5. Press one time the Shift+Up arrow key combination. You will be hear one time the "unselected" text with end of text spokening.
6. Press again the Shift+Up arrow key combination. Hopefuly you will be hear two spokened "unselected" text with end of text spokening.
If this is happening, second problem is reproduced.

What showing my debug.out file? I switched language with english and tested Libreoffice 3.3.

Attila

I notice this as well. However in the scale of things the speaking of "selected" and unselected" is quite minor compared to the crashing, particularly as the selection change is spoken correctly (IE. the text which has either just been selected or unselected).

This is with openoffice 3.3.0 and a 3.4 dev build.

Yes the speaking of the messages should be corrected, but should it be done here or as a separate issue?
(In reply to comment #13)
> Created an attachment (id=180387) [details]
> Debug file with show Libreoffice 3.3 result
> 
> Joanie, I am very surprising this bug is resolved, I not get prewious bugreport
> e-mails unfortunately, I don't no why. :-):-)
> 
> I see unfortunately my wrote problem with Libreoffice 3.3 version with my
> Ubuntu Lucid system. The Orca is full uptodated. Not happening crashes, but for
> example when I select two lines with Shift+Down arrow key, I not hear the
> "Selected" message. When I press Shift+Up arrow key one time, I hear one the
> "unselected" message with end of the text spokening. But when I press again
> Shift+Up arrow key and remove the first selected line selection, I hear two
> time the "unselected" message.
> 
> I see unfortunately this similar result with Openoffice.org 3.2 version with
> native awailable with Ubuntu Lucid.
> Anything I doing wrong, or my system is too old for Orca 2.91?
> 
> Reproducation steps, hopefuly my results is not confirmed:
> 1. Launch Openoffice.org or Libreoffice.org writer.
> 2. Write following two line:
> apple
> potato
> Press enter key after the second line.
> 3. Press two up arrow key to jump the first line.
> 4. Press two Shift+Down arrow combination and listening the spokened output. If
> you not hear the "selected" message with end of text spokening, the first
> problem is reproduced.
> 5. Press one time the Shift+Up arrow key combination. You will be hear one time
> the "unselected" text with end of text spokening.
> 6. Press again the Shift+Up arrow key combination. Hopefuly you will be hear
> two spokened "unselected" text with end of text spokening.
> If this is happening, second problem is reproduced.
> 
> What showing my debug.out file? I switched language with english and tested
> Libreoffice 3.3.
> 
> Attila

(In reply to comment #13)
........

> I see unfortunately this similar result with Openoffice.org 3.2 version with
> native awailable with Ubuntu Lucid.
> Anything I doing wrong, or my system is too old for Orca 2.91?
> 
> Reproducation steps, hopefuly my results is not confirmed:
> 1. Launch Openoffice.org or Libreoffice.org writer.
> 2. Write following two line:
> apple
> potato
> Press enter key after the second line.
> 3. Press two up arrow key to jump the first line.
> 4. Press two Shift+Down arrow combination and listening the spokened output. If
> you not hear the "selected" message with end of text spokening, the first
> problem is reproduced.
> 5. Press one time the Shift+Up arrow key combination. You will be hear one time
> the "unselected" text with end of text spokening.
> 6. Press again the Shift+Up arrow key combination. Hopefuly you will be hear
> two spokened "unselected" text with end of text spokening.
> If this is happening, second problem is reproduced.

This is confirmed as described. However I think that it should be treated in a separated bug.

> 
> What showing my debug.out file? I switched language with english and tested
> Libreoffice 3.3.
> 
> Attila

(In reply to comment #12)
> Thanks everyone for the testing!
> 
> Given the positive results reported here, the minor nature of the change with
> respect to the Orca code, and the fact that the change I made can be undone by
> anyone who wants to via their orca-customizations.py.... I've committed the
> patch to master. That should get it additional testing and feedback long before
> we hit the GNOME 3.0 code freeze.

Thanks for the correction. I have some colleagues who are happy!
Any chance of the patch is officially implemented in the current version of Orca?

(In reply to comment #15)
> (In reply to comment #13)
> ........
> > 6. Press again the Shift+Up arrow key combination. Hopefuly you will be hear
> > two spokened "unselected" text with end of text spokening.
> > If this is happening, second problem is reproduced.
> 
> This is confirmed as described. However I think that it should be treated in a
> separated bug.

I agree. *This* bug is about the fact that OpenOffice/LibreOffice seems to like to kill itself if we process events asynchronously. We no longer process events asynchronously and OOo seems to no longer want to kill itself. Hence FIXED.

Having said that, I fully anticipate small side effects to crop up as a result of our handling things different. As Michael suggested, all of these things should be treated as new and separate bugs.

Thanks guys!

(In reply to comment #16)
> Thanks for the correction. I have some colleagues who are happy!

<smiles> We aim to please.

> Any chance of the patch is officially implemented in the current version of
> Orca?

If by current you mean 2.32, then I am not sure. I say that because GNOME officially is done with the 2.32 releases. So if there is to be a 2.32.2, we'd be doing it on our own and would have to hope downstream distros pick it up. (Does not mean we won't do one; merely that I don't know if we will.)

In the meantime, fortunately, users of older versions of Orca can do the same thing via orca-customizations.py. This should do it:

import orca.settings
orca.settings.synchronousToolkits = ['VCL']

Similarly, if you had a case where OOo was totally stable and now it's behaving strangely as a result of my fix for this bug, you can similarly undo my change by instead placing the following in your orca-customizations.py:

import orca.settings
orca.settings.synchronousToolkits = []

So:
Need I reporting I seed issue with a separate bug? This problem is need fixing with Orca level? Have a chance to fix this problem but keep this now working right setting to prewent OO crash?
The problem if I turn off this setting, I hear right the selection messages, but OO is crash. If I using this setting, I not hear the selected messages, some time I hear two time the unselected messages, but OO is not crash. So, not have another choice, need keep now committed setting.
Unfortunately I not see this bugreport e-mails, I prewious added with CC list. Sorry the long answer.

Attila

I reported I experienced selection state messages spokening problems with bug 642353 bugreport, and attached a debug.out file.

Attila