Write cannot handle Composed characters

[sayamindu]

Steps to reproduce

1. Install Joyride 2330 in a laptop 2. Edit /etc/sysconfig/keyboard to look like

KEYTABLE="us"
XKB_MODEL="olpc"
XKB_LAYOUT="us,et"
XKB_VARIANT="olpc2,basic"

3. Set /home/olpc/.i18n to am_ET.UTF-8 4. Restart sugar 5. Start Write 6. Press the switch layout key (if it is a G1G1 variety, the division/multiplication key below the enter key is a substitute) 7. Type the sequence u h (no spaces in between)

Actual result: An unicode box character, followed by ሀ comes up

Expected result: The character ሁ should appear

To see the expected result, repeat this in the mesh search bar in the sugar screen.

[gregorio]

Hi Sayamindu,

Is this something which will affect users?

I set the milestone to 8.2 but no blockers status yet. We can call it a blocker if you show that it affects real users in nnnn language.

Thanks,

Greg S

[sayamindu]

This affects real users of the Amharic language, who will not be able to type in certain characters using Write. It may be the result of some interaction problems with Rainbow and Write, since removing /etc/olpc-security and restarting sugar solves the issue.

m_stone: any ideas on what may be wrong ?

[sayamindu]

Note:

I created a simple VTE based terminal from Pippy (running under isolation), and started Write from it with the command sugar-launch write, and it seemed to work fine.

[sayamindu]

It seems to be a regression from Update.1, and seems to be a rainbow specific issue.

However, since Ethiopia has chosen to go with Update.1, I'm rescheduling this for 8.2.1.

[kimquirk]

in build 8.2-760, by typing u followed by h in the search bar I get a character like the one on the 'h' key (sort of like mu), with an added a small hyphen.

In Write, when I type u I get the small box with FE on top and 75 below, then i add the 'h' and half the box goes away and is replaced by the mu character.

This testing was done on an actual Amharic keyboard from quanta.

[sayamindu]

Setting environRAINBOW_STRACE_LOG? = '1' in activityfactory.py makes the problem goes away. Also did a os.system('sugar-launch write') from pippy, which worked.

[sayamindu]

This definitely looks like some kind of race condition to me. I attached to the rainbow process with strace (used the -f option to follow forks), and noticed that under normal circumstances, the iso8859-1 Compose file was getting loaded. This indicates that X is getting the locale as 'C', and not as am_ET.UTF-8. Maybe the time overhead with $RAINBOW_STRACE_LOG = '1' is causing the normal loser to win the race ?

[sayamindu]

Patch which makes _get_lang() return xx_XX.YYY instead of "xx_XX.YYY"

[sayamindu]

Attached the fix. It turns out that _get_lang() in Rainbow class in service.py was returning "nn_NN-XXX" where it should have returned nn_NN-XXX (no quotes). This was causing setlocale() (I think which is called from somewhere deep within gtk_init()) to bomb silently.

This also seems to fix the issue where the Ethiopian keyboard was producing garbage in Write unless the font was being set explicitly. This was also being seen in #2207

Please review this patch, and I think it makes sense if this goes into 8.2.0

[mstone]

Merged and built as rainbow-0.7.23-1.fc9.

[mstone]

I found a second bug in the function in _get_lang() function which I have corrected and tested in rainbow-0.7.24-1.fc9. Please include this NVR instead of -23.

[mstone]

We need to be able to enter composed characters in Write. Upgrading this to 'blocks:'

[cscott]

Test case, please.

[sayamindu]

|Test Case|

1. Edit /etc/sysconfig/keyboard to look like

KEYTABLE="us"
XKB_MODEL="olpc"
XKB_LAYOUT="us,et"
XKB_VARIANT="olpc2,basic"

3. Set /home/olpc/.i18n to am_ET.UTF-8

4. Restart sugar

5. Start Write

6. Press the switch layout key (if it is a G1G1 variety, the division/multiplication key below the enter key is a substitute)

7. Choose the Abyssinica SIL font (this is not required, but just to be safe)

8. Type the sequence u h (no spaces in between) in a text object

Expected result: The character ሁ should appear

[sayamindu]

Re my previous comment, replace the phrase "in a text object" by "in write". I should really go to sleep :-)

[cscott]

Please test build 761 resulting from build-repo commit 9cbb2d4a.

[mchua]

assigning i18n bugs to kimquirk to test as per kim's request

[sayamindu]

Works in release 8.2-761 and above.

[gregorio]

Closed with no doc needed. GS