On 2016-12-14 18:36, Henri Verbeet wrote:
On 14 December 2016 at 17:27, Lauri Kenttä lauri.kentta@gmail.com wrote:
Unfortunately MAP_FOLDDIGITS seems to map too many things (e.g. Tamil, which is already tested and shouldn't work).
Does it on Windows as well? That is, our unicode tables aren't necessarily quite the same as the Microsoft ones, and the Microsoft ones aren't necessarily the same across Windows versions. Likewise, is that consistent across different versions of msvcr? I could imagine different msvcr versions matching different unicode versions.
Testing on Windows Server 2003, I get 84 code points which wcstoi64 doesn't recognize but MAP_FOLDDIGITS manages to convert. These include superscripts, subscripts, circled numbers etc., and Tamil.
The most interesting question is, what's wrong with Tamil? :(
Maybe we could use MAP_FOLDDIGITS anyway, but bug for bug, right?
And as for different Windows versions, my wctoint seems to be fine: https://testbot.winehq.org/JobDetails.pl?Key=27358