<div class="gmail_quote">2011/9/29 Ruben Van Boxem <span dir="ltr"><<a href="mailto:vanboxem.ruben@gmail.com">vanboxem.ruben@gmail.com</a>></span><br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="gmail_quote"><div><div></div><div class="h5">2011/9/29 Howard Hinnant <span dir="ltr"><<a href="mailto:hhinnant@apple.com" target="_blank">hhinnant@apple.com</a>></span><br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div><div></div><div>On Sep 29, 2011, at 7:15 AM, Ruben Van Boxem wrote:<br>
<br>
> 2011/9/29 Ruben Van Boxem <<a href="mailto:vanboxem.ruben@gmail.com" target="_blank">vanboxem.ruben@gmail.com</a>><br>
> 2011/9/29 Ruben Van Boxem <<a href="mailto:vanboxem.ruben@gmail.com" target="_blank">vanboxem.ruben@gmail.com</a>><br>
> 2011/9/29 Ruben Van Boxem <<a href="mailto:vanboxem.ruben@gmail.com" target="_blank">vanboxem.ruben@gmail.com</a>><br>
> 2011/9/29 Howard Hinnant <<a href="mailto:hhinnant@apple.com" target="_blank">hhinnant@apple.com</a>><br>
><br>
> On Sep 28, 2011, at 6:17 PM, David Chisnall wrote:<br>
><br>
> > On 28 Sep 2011, at 22:45, Howard Hinnant wrote:<br>
> ><br>
> >> I think FreeBSD should define _LLIBCPP_STABLE_APPLE_ABI too. David? And maybe with David's latest patch to FreeBSD, should also #define _LIBCPP_GET_C_LOCALE 0.<br>
> ><br>
> > FreeBSD should now be mirroring the Apple implementation for anything xlocale related. If there are any discrepancies, they are bugs and I'll fix them in libc, rather than hack around them in libc++.<br>
> ><br>
> > I'm not entirely sure what defining _LLIBCPP_STABLE_APPLE_ABI means - perhaps it could be replaced with a more descriptive name?<br>
><br>
> See revision 140734, and let me know what needs fixing (hopefully nothing).<br>
><br>
> This improves the behavior on Windows (uses the C locale instead of returning NULL like for Linux). Attached patch also removes unreachable code (_LIBCPP_HAS_DEFAULTRUNELOCALE is defined for both __APPLE__ and __FreeBSD__, so the additional checks in locale.cpp are never reached).<br>
><br>
> I'm now also wondering about _LIBCPP_WCTYPE_IS_MASK. I've been looking at MSDN, but can't figure out if it matches. I did find a warning that if a plain "char" is passed to Windows is* functions, unexpected results are obtained (<a href="http://msdn.microsoft.com/en-us/library/4yc6feha%28v=vs.80%29.aspx" target="_blank">http://msdn.microsoft.com/en-us/library/4yc6feha%28v=vs.80%29.aspx</a>). It seems that char_type could be char in the common case, and this could yield unreliable behavior?<br>
><br>
> Ruben<br>
><br>
> Never mind this patch. It's broken. Is there any reason not to call plain "toupper" in do_toupper directly? Or at least "toupper_(c,__cloc())"?<br>
><br>
> Ruben<br>
><br>
> Okay, brace yourself...<br>
><br>
> I've tried to remove locale.cpp's dependency on glibc's toupper and tolower transformation tables, which are only available in glibc. I have coded a fallback do_tolower and do_toupper using basic ASCII subtract and add tricks if glibc or the DefaultRuneLocale is not available. I know this is slower, but it works at least. It uses islower_l and isupper_l with __cloc to check. The change in behavior should be:<br>
> Mac: none<br>
> FreeBSD: none<br>
> Linux/glibc: none<br>
> other platforms (including Windows): no __classic_upper/lower_table functions declared, defined, or used.<br>
> I also declared a symbol present in msvcrt.dll that represents the mask table and added that case to classic_table.<br>
><br>
> I would prefer having a (global) static "C" locale object that can just be given through __cloc, although I don't know if this is safe, but as long as __cloc is only used in _l functions (which I think it is), this approach of a persistent (global, not function-local) C locale object should work (is initialized before any threads could call __cloc at the same time) and fix the FIXME in __cloc.<br>
><br>
> I also tried to fix the ctype<wchar_t>::do_is to keep track of the result, and now should work if multiple flags were requested.<br>
><br>
> Comments welcome!<br>
><br>
> Today is not my day. Right patch is attached, with missing bits included this time. After this, I'm taking a break, before I spam this list with broken patches :(<br>
><br>
><br>
> Ruben<br>
><br>
</div></div>> <windows.patch><br>
<br>
This works for Apple, thanks! Committed revision 140781.<br></blockquote></div></div><div><br>Sorry about the mess, it seems the missing bits were... missing. A file was missed, and I had to rename the locale support header/source file or else the buildit script overwrites libc++'s locale.o with my support/win32/locale.o.<br>
<br>Please apply, thanks,<br></div></div></blockquote><div><br>Ugh, with patch.<br> </div><blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div class="gmail_quote">
<div><br>Ruben<br><br>
</div><blockquote class="gmail_quote" style="margin:0pt 0pt 0pt 0.8ex;border-left:1px solid rgb(204, 204, 204);padding-left:1ex">
<font color="#888888"><br>
Howard<br>
<br>
</font></blockquote></div><br>
</blockquote></div><br>