Re: [jamboost] Re: Re: win32 gcc should use -Wl,--enable-auto-image-base for dll building

From: William E. Kempf (wekempf_at_[hidden])
Date: 2003-01-24 01:07:54

David Abrahams said:
> "William E. Kempf" <wekempf_at_[hidden]> writes:
>
>> Beman Dawes said:
>>> At 10:24 AM 1/22/2003, David Abrahams wrote:
>>>
>>> >"Raoul Gough" <raoulgough_at_[hidden]> writes:
>>> >
>>> >>
>>> >> OK, I'll go with that (see the attached diff - it does the trick
>>> on
>>> Win2000). David, do you mind if I check this in?
>>> >
>>> >looks good to me!
>>>
>>> I haven't been following this thread, or the one about gcc-nocygwin.
>>>
>>> "gcc" is currently used for the Win32 regression tests. I do have
>>> cygwin
>>> installed on my machine, but get the gcc sources directly and
>>> compile
>>> them for the version used in the tests.
>>>
>>> If any of that should change, please let me know.
>>
>> Once the dust settles, yes, I think something should change. The gcc
>> toolset, using a Cygwin compiler, picks up the Cygwin POSIX emulation
>> layer and effects how many libraries get compiled/implemented. For
>> example, it causes the Boost.Threads library to build using pthreads
>> instead of Win32 threads. I think this is at least misleading, when
>> it's reported as a "Windows" compiler.
>
> It's not misleading if the results are labelled "Cygwin GCC"
> (whatever). And even so, the toolset docs for GCC clearly state that
> it's for Cygwin GCC only.

It's not labeled "Cygwin GCC", it's only labeled "GNU GCC". I wasn't
aware that the gcc toolset only worked with the Cygwin GCC, however. And
I think it's still at least confusing, if not misleading. Further, I
think it's important that we include a "true" Windows GCC compiler in the
Windows regression reports, regardless of whether or not we continue to
include the Cygwin results.

>> The question would be, then, whether or not the Windows regression
>> report should only be using gcc-nocygwin, or if it maybe should be
>> using both toolsets.
>
> We have a mingw toolset; it could always be done with gcc and mingw.

Yes... that would be up to Beman I guess. It would require yet another
installed compiler, though, so the gcc-nocygwin would probably be better.

William E. Kempf
wekempf_at_[hidden]