Re: [boost] Using clang on Windows with mingw-64

Subject: Re: [boost] Using clang on Windows with mingw-64
From: Paul A. Bristow (pbristow_at_[hidden])
Date: 2015-07-11 05:34:40

> -----Original Message-----
> From: Boost [mailto:boost-bounces_at_[hidden]] On Behalf Of Edward Diener
> Sent: 10 July 2015 17:06
> To: boost_at_[hidden]
> Subject: [boost] Using clang on Windows with mingw-64
> 
> This is just a heads up about testing with clang on Windows.
> 
> The latest clang, when built from the latest source, is now capable on Windows of using the mingw-
> 64/gcc RTL. Furthermore it is no longer necessary to have the mingw/mingw-64 implementation
> hardcoded or linked from c:\mingw. With the latest clang the executable will look for a
> mingw/mingw-64 implementation on the PATH, meaning that the if the implementation's 'bin'
> directory is on the PATH clang will find it and use its RTL.
> 
> The ability of the latest clang to use mingw-64 means that clang can use the latest gcc-5.1 RTL on
> Windows. The mingw-64 implementation offers more up to date and better implementations of gcc
> than mingw.
> 
> There is one "gotcha" I have found with this setup on Windows when using clang on Windows with
> gcc-4.9 on up. The gcc intrin.h implementation of
> mingw-64 for gcc-4.9 on up has a bug where if the gcc version is less than 4.9 a compiler error
will
> occur. I have reported this to mingw-64 on their developer's mailing list but it has been ignored.
> 
> Unfortunately clang still sets itself to gcc version 4.2.1. But a workaround solves the problem
and that
> workaround is that when using clang you define '__MINGW_FORCE_SYS_INTRINS', which causes the
> bug in the mingw-64 intrin.h header file to not trigger.
> 
> Previous versions of clang on Windows ( 3.4, 3.5, 3.6 ) of course still have the limitation of
only
> working with mingw and needing the mingw implementation hardcoded at c:\mingw. But at least with
> the latest clang built from source this has changed as explained above.
> 
> The Boost build clang toolset, which on Windows is really clang-linux, should continue to work
with
> this latest change of clang.

Thanks for banging your head against this issue!

I look forward to being able to use this in future, preferably without all your complicated
workarounds.

Paul

---
Paul A. Bristow
Prizet Farmhouse
Kendal UK LA8 8AB
[phone]