2019.2: Revert "Update liblouis to version 3.10 (#9678)"

[leonardder]

This reverts commit d5ed95c.

Link to issue number:

Fixes #9982

Summary of the issue:

Large chinese braille tables are unable to load in some cases when using Liblouis 3.10. Downgrading to Liblouis 3.9 seems to fix this according to #9982 (comment)

Description of how this pull request fixes the issue:

Downgrades to liblouis 3.9 by reverting the 3.10 pr

Testing performed:

Tested in a try build that the issues for Chinese no longer occurred.

Known issues with pull request:

• I had a branch that was based on rc, and when I rebased on beta, I noticed that there are commits in rc not in beta.
• In Braille display auto-detection feature causes errors on loading large braille tables on NVDA startup #9982 (comment), @josephsl expressed concerns about a downgrade
• We probably have to be very careful when merging beta into master.

Change log entry:

• Downgraded liblouis to version 3.9 to solve issues when loading Chinese braille tables. (Braille display auto-detection feature causes errors on loading large braille tables on NVDA startup #9982)

[josephsl]

Hi, specifically, my worry that it will break Korean braille tables. Thanks. From: Leonard de Ruijter <notifications@github.com> Sent: Monday, September 16, 2019 10:44 PM To: nvaccess/nvda <nvda@noreply.github.com> Cc: Joseph Lee <joseph.lee22590@gmail.com>; Mention <mention@noreply.github.com> Subject: [nvaccess/nvda] 2018.2: Revert "Update liblouis to version 3.10 (#9678)" (#10232) This reverts commit <d5ed95c> d5ed95c. Link to issue number: Fixes #9982 <#9982> Summary of the issue: Large chinese braille tables are unable to load in some cases when using Liblouis 3.10. Downgrading to Liblouis 3.9 seems to fix this according to #9982 (comment) <#9982 (comment)> Description of how this pull request fixes the issue: Downgrades to liblouis 3.9 by reverting the 3.10 pr Testing performed: Tested in a try build that the issues for Chinese no longer occurred. Known issues with pull request: * I had a branch that was based on rc, and when I rebased on beta, I noticed that there are commits in rc not in beta. * In #9982 (comment) <#9982 (comment)> , @josephsl <https://github.com/josephsl> expressed concerns about a downgrade * We probably have to be very careful when merging beta into master. Change log entry: * Downgraded liblouis to version 3.9 to solve issues when loading Chinese braille tables. (#9982 <#9982> )

…

_____ You can view, comment on, or merge this pull request online at: #10232 Commit Summary * Revert "Update liblouis to version 3.10 (#9678)" File Changes * M include/liblouis <https://github.com/nvaccess/nvda/pull/10232/files#diff-0> (2) * M nvdaHelper/espeak/sconscript <https://github.com/nvaccess/nvda/pull/10232/files#diff-1> (2) * M nvdaHelper/liblouis/sconscript <https://github.com/nvaccess/nvda/pull/10232/files#diff-2> (3) * M readme.md <https://github.com/nvaccess/nvda/pull/10232/files#diff-3> (2) Patch Links: * https://github.com/nvaccess/nvda/pull/10232.patch * https://github.com/nvaccess/nvda/pull/10232.diff — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#10232?email_source=notifications&email_token=AB4AXEC2ZRZUEAP4FTBOZATQKBVDNA5CNFSM4IXK66W2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HLYESDA> , or mute the thread <https://github.com/notifications/unsubscribe-auth/AB4AXEHNAY7HCW3PWJ7R4KTQKBVDNANCNFSM4IXK66WQ> .

[leonardder]

Could you specify the expected breakage? Note that Korean isn't at all mentioned in the news file.

[josephsl]

Hi, specifically, grade 1 tables load, but grade 2 tables may or may not load. We shall find out once this goes into beta branch and ultimately, 2019.2.1. Thanks. From: Leonard de Ruijter <notifications@github.com> Sent: Monday, September 16, 2019 10:55 PM To: nvaccess/nvda <nvda@noreply.github.com> Cc: Joseph Lee <joseph.lee22590@gmail.com>; Mention <mention@noreply.github.com> Subject: Re: [nvaccess/nvda] 2018.2: Revert "Update liblouis to version 3.10 (#9678)" (#10232) Could you specify the expected breakage? Note that Korean isn't at all mentioned in the news file. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#10232?email_source=notifications&email_token=AB4AXEDOM2ZNNIMJTCWFOFLQKBWM3A5CNFSM4IXK66W2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD63ME3Q#issuecomment-532071022> , or mute the thread <https://github.com/notifications/unsubscribe-auth/AB4AXEFHUVH4I7XHDJG3BO3QKBWM3ANCNFSM4IXK66WQ> .

[michaelDCurran]

I'm really not sure what we should do about this yet. As Liblouis 3.10 does fix some buffer overruns. Knowingly moving back to a release that introduces them is strange. Yet it obviously fixes the Chinese crash... Perhaps I'll spend a few hours debugging the chinese issue and see if we can provide a patch. If not we'll downgrade.

[lukaszgo1]

It would be nice not to reintroduce #9486 if at all possible. Note that #9486 happens only when Liblouis dll's are optimized, so in the worth case it may be possible to use Liblouis 3.9, and if it breaks Korean disable optimization for it.

[Qchristensen]

Would upgrading to LibLouis 3.11, rather than downgrading, resolve both the large Chinese table issue and the previous buffer overrun issue? http://liblouis.org/liblouis/2019/09/02/release-3.11.0.html

Or is that too unknown a change?

[leonardder]

There is an issue about updating to liblouis 3.11 (#10161), but that one is either blocked by an update to Visual Studio 2019, or bundling a binary build produced by the liblouis folks.

[leonardder]

It would be nice not to reintroduce #9486 if at all possible. Note that #9486 happens only when Liblouis dll's are optimized, so in the worth case it may be possible to use Liblouis 3.9, and if it breaks Korean disable optimization for it.

Liblouis 3.9 is likely to reintroduce this, indeed, so this pull request is certainly not ready as it is now.

[DrSooom]

Comparing result between NVDA 2019.1.1 and 2019.2:

ko.cti – Line 64:

include chardefs.cti All character definition opcodes
» include en-chardefs.cti English character definition opcodes

ko-g2.ctb – New comment at line 45, rest is identical.

ko-chars.cti, ko-g1.ctb, ko-g1-rules.cti and ko-g2-rules.cti have the same SHA-256 checksum.

And please don't forget to read this comment here and this one here as well. There is no answer why the whole thing worked as expected with NVDA 2019.2 RC1 yet.

[dpy013]

hello
So is the problem with this pr completely solved?
thanks

This reverts commit d5ed95c.

Link to issue number:

Fixes #9982

Summary of the issue:

Large traditional Chinese braille tables are unable to load in some cases when using Liblouis 3.10. Downgrading to Liblouis 3.9 seems to fix this according to #9982 (comment)

Description of how this pull request fixes the issue:

Downgrades to liblouis 3.9 by reverting the 3.10 pr

Testing performed:

Tested in a try build that the issues for traditional Chinese no longer occurred.

Known issues with pull request:

• I had a branch that was based on rc, and when I rebased on beta, I noticed that there are commits in rc not in beta.
• In #9982 (comment), @josephsl expressed concerns about a downgrade
• We probably have to be very careful when merging beta into master.

Change log entry:

• Downgraded liblouis to version 3.9 to solve issues when loading traditional Chinese braille tables. (Braille display auto-detection feature causes errors on loading large braille tables on NVDA startup #9982)

hi Collaborator
I replaced the previous Chinese with the traditional Chinese. In the issue, the test is based on traditional Chinese braille.

[dpy013]

In Chinese, it is divided into Simplified Chinese (zh_CN) and Traditional Chinese (zh_TW)
issue 9982
Test braille is zh_TW

[Adriani90]

To be honnest, I am not really understanding what exactly the user impact will be 9in practice. We need more testing results from chinese users on this. @josephsl do you have any contacts we could use for this purpose? I would rather update to liblouis 3.11 and try to solve this issue after the user impact is assessed. Maybe it is not really a liblouis issue. In fact, there are people who cannot reproduce this issue under the same conditions. Could someone maybe post this to the NVDA users list and try to get some clearer test results?

[michaelDCurran]

Due to the inability to reproduce this, confusion around how many users are really affected, and no suitable solution being found which does not reintroduce other errors, this will not be in 2019.2.1.