Re: [Mingw-users] Building MinGW GCC

[Keith M. <kei...@us...>]

On 09/06/15 14:23, Keith Marshall wrote:
> On 08/06/15 17:56, Eli Zaretskii wrote:
>>> From: David Gressett <DGr...@am...>
>>> I have removed the - -with-mpc, --with-mpfr, and --with-gmp elements
>>> from CONFOPT, which means that you have to have the mpc, mpfr, and
>>> gmp libraries installed in your MinGW configuration. I did this because
>>
>> You mean, as opposed to having their sources in the GCC tree?
> 
> I normally place (symlink) these in-tree, and let them build along with
> GCC itself.
> 
>> I see you use --enable-threads -- AFAIU this means libstdc++ will be
>> devoid of <thread> implementation, is that right?  Did you do this on
>> purpose, or just because previous MinGW compilers were built like
>> that?
> 
> IIRC, Danny Smith, (a former GCC developer who also maintained our
> releases up to, and including GCC-3.4.5), always built with this; I
> guess we've just stuck with it, as a result of inertia, but there may
> well be grounds for reviewing this choice.
> 
>>> I have had no success building gcc 5.1.0. The MinGW 3.21 runtime
>>> introduced a new function which triggered a bug in the Fortran runtime
>>> library. 
>>
>> Which function is that?
> 
> I guess he's referring to the issue he reported, and I identified here:
> http://thread.gmane.org/gmane.comp.gnu.mingw.user/44500/focus=44515

FWIW, I've patched my local sources to circumvent this.  GCC-4.8.2
remains a "lost cause" (see below), but I have successfully created,
(and posted[*]) crossed-native builds of both GCC-4.8.5 and GCC-4.9.3.
GCC-5.x failed similar treatment, due to a different regression, as
noted in http://thread.gmane.org/gmane.comp.gnu.mingw.user/44776

>>> Going back to a previous V3 runtime would work around that problem,
>>> but it would probably not work around another problem that occurs when
>>> the Ada 5.1.0 runtime is built.

I've used (effectively) mingwrt-3.21.1 for my builds; (I've fixed the
hypot() issue in my local working copy).  GCC-5.x still chokes for me,
but it's a libgcc issue which blocks it, before any other.

>> Building Ada requires a previous version of Gnat to be installed,
>> doesn't it?
> 
> AFAIK, yes.  In fact, building Ada is a real PITA; when building
> crossed-native, as I do, it requires a bootstrap of a build-native,
> Ada-enabled GCC at identically the same version as the ultimate target,
> which is then used to build an Ada-enabled cross-compiler at the same
> version, which is then used to complete the crossed-native build.  When
> I built GCC-4.8.2, I got as far as successfully building the Linux
> hosted, Ada-enabled mingw32 cross-compiler, but never was able to
> successfully complete the Ada component for the crossed-native build.

GCC bug #55946 blocks building GCC-4.8.2 with ada included; hence my
designation of this as a "lost cause".  This is fixed in both GCC-4.8.5
and GCC-4.9.3, both of which build (relatively) cleanly.  I didn't apply
any ada specific patches to achieve a successful build, but I've no idea
if it will actually work, (and I've no knowledge of ada at all, to help
me to assess it); perhaps David could comment?

[*] download links:
https://sourceforge.net/projects/mingw/files/MinGW/Base/gcc/Version4/experimental/gcc-4.8.5-preview/
https://sourceforge.net/projects/mingw/files/MinGW/Base/gcc/Version4/experimental/gcc-4.9.3-preview/

-- 
Regards,
Keith.