Re: [PATCH] handle casing of i/I in Turkic languages

On Apr 23 15:44, Christophe Lyon wrote:
> Hi!
> 
> I'm afraid this patch breaks users of towctrans().
> Linaro CI reported this on February 25th on arm targets, and I didn't
> notice any feedback?
> 
> To be more specific, it introduces regressions when running the GCC
> libstdc++ testsuite, for instance when you reach
> towupper, which then calls towctrans,
> The comments at the beginning of towctrans.c (and other files) say:
> ----------------------------------------------
> If <[locale]> is
> LC_GLOBAL_LOCALE or not a valid locale object, the behaviour is undefined.
> -------------------------------------------------
> and after this patch we call:
> -    return towctrans_l (c, w, 0);
> +    return towctrans_l (c, w, LC_GLOBAL_LOCALE);
> 
> Then, towctrans_l calls __jp2uc_l (c, locale)
> and _jp2uc_l does
>   const char * cs = l ? __locale_charset(l) : __current_locale_charset();
> where 'l' is the locale (so LC_GLOBAL_LOCALE == -1) in this case,
> and __locale_charset does:
> #ifdef __HAVE_LOCALE_INFO__
>   return __get_ctype_locale (locale)->codeset;
> #else
>   return locale->ctype_codeset;
> #endif
> 
> so it dereferences a pointer with value -1 / 0xFFFFFFFF
> leading to a crash.
> 
> 
> Am I missing something, is this supposed to work?

It's supposed to work.  I pushed a patch
https://sourceware.org/cgit/newlib-cygwin/commit/?id=4ba7eff4d15a


Thanks,
Corinna