Re: [Libreoffice] [REVIEW] 3-5, 3-5-0, 3-4 corrupted Indic characters and crash under Windows fdo#44208, fdo#45107

On Wed, 2012-01-25 at 11:52 +0100, Andras Timar wrote:
> Hi,
>
> Yesterday two bugs came to my attention.
> fdo#44208 - under Windows crash occurred when user selected Tools -
> Options in selected Indian locales

Doesn't seem to be a lot of info wrt backtraces or anything like that in
that bug. I've seen crashes like this before with FreeSans and icu so my
instinct would be to blame icu :-) 

In which case it *might* be possible to figure that crash out by copying
the windows fonts to a linux box under ~/.fonts and running under
valgrind and scrolling through the font list

> fdo#45107 - under Windows garbage characters appeared at random places
> in selected Indian locales

> I would like to have these commits in libreoffice-3-5,
> libreoffice-3-5-0, and libreoffice-3-4. These are rather serious
> issues, that make Indian localizations useless under Windows.

On the basis that we should keep language tags as short as possible when
there isn't ambiguity I'm happy enough to have e.g. "hi" instead of
"hi-IN" and so on, while zh-CN has to remain as zh-CN given that there
are two scripts in use in different territories to write Chinese.

I *presume* this isn't simply some casing cockup ?, e.g. hi-IN vs hi-in
or some such. Not entirely sure where the code is for reading those and
matching them against the requested locale.

> Also, if somebody (Caolán?) knows how VCL.xcu works today, please help
> to update the documentation in wiki:
> http://wiki.documentfoundation.org/LibreOffice_Localization_Guide#Define_the_default_fonts_for_the_locale

Well, I updated it a little to show where fontconfig comes into the
picture under Unix, which is the bit I know best.

> If for example "ta-IN" entry was ignored, and it works with "ta", then
> we need to review the whole file, because many locales have country
> codes, while UI localization uses only language code. Should we add
> entries for locales which are not there? 

In the general case, stuff like hu-HU should probably be just "hu", its
not like Hungarian is some territory outside of Hungary gets written in
a different script (I presume anyway), so following the "as short as is
possible" rule, while e.g. pa-IN is a bit more problematic given the
multiple scripts used to write Punjabi in different territories and best
left alone without a bit of research to determine which script a bare
"pa" would best default to.

> Moreover, I'm not sure if the
> recently fixed fdo#43984 works at all (it uses zh-cn and zh-hk locale
> codes).

You mean
http://cgit.freedesktop.org/libreoffice/core/commit/?id=969abdc19a4d686b0288fee3d8e223df04becc7e
It better work :-), because like the pa-IN/pa-PK example zh-TW and zh-HK
are Traditional Chinese territories while zh-CN is Simplified Chinese so
need different lists of preferred fonts for the two different scripts.

Anyway, best would be to find the bit of locale matching code and see
why it doesn't do the right thing, but either way the final outcome of

82e0266b05dd53d0ce8b12bd5c76ae25872b87a9
86bf32c487852f4e466b27320e3724266cb53dba
9b55eeba78d4e6418750a12c3dad4b42e2dd3f03

looks sane to me

C.