Right now, wine claims to use DUCET data for lingustic sorting, but by section 1.9.2 of that document (as of version 6.0.0), uses it in a wrong way. The result of it are bugs such as #10767 and #9583.
A possible way around it would beby using ICU to get language specific tailoring and applying some of wine-specific for the parts addressed by #10767. In a way, wine already *indirectly* depends on icu on many distros (a potential libxml2 dep) - they were quite a few patches already addressing build failures caused by this in recent past.
I've cooked up a patch, that's unfortunately isn't working properly yet, it's more of a draft on where to go, if such road were to be taken.
Main issue with this patch ain't it's not ignoring the proper list of symbols, as this would be easy to fix, if not for one problem: the part that I put inside 'if (0)' block.
For some reason (threading, perhaps ?), if I pass a non-zero lenght string to ucol_openRules, wineboot hangs. Oddly, if the string lenght is 0, the hang doesn't happen. I can't really tell, what going wrong, as the almost same code works in native linux testcase.
Also, while in this patch UCOL_ALTERNATE_HANDLING would have been used for NORM_IGNORESYMBOLS, it would be much better to manipulate some of the settings by a custom rule and VariableTop value, but to do that, first such rule would have to be passed and that odd hang prevents it at the moment.
It would also be nice, if it was possible to initialize the collator only once for an app instance - after all, in Windows locale change requires a reboot, so while (AFAIK) one wineserver instance can have apps in different locales running, locale can't change for an already running app.
So if you have any idea of why exactly the hang happens and how to get around it or know any *technical* reasons of why ICU couldn't be used in wine, CC me with comments.