Request for Testers: gsoc_nci branch

Peter Lobsinger plobsing at gmail.com
Wed Nov 10 17:05:09 UTC 2010


On Wed, Nov 10, 2010 at 11:54 AM, Andy Dougherty <doughera at lafayette.edu> wrote:
> On Wed, 10 Nov 2010, Peter Lobsinger wrote:
>
>> On Wed, Nov 10, 2010 at 9:22 AM, Andy Dougherty <doughera at lafayette.edu> wrote:
>> > On Wed, 10 Nov 2010, Peter Lobsinger wrote:
>> >
>> >> The gsoc_nci branch is at a point where it should be mergeable. Please
>> >> test it on any platforms you happen to care about. 'make test' should
>> >> exercise the full functionality, but other variants couldn't hurt
>> >  > either.
>> >>
>> >> Test configurations of interest are (in more or less descending order
>> >> of importance):
>> >>  * libffi *not* installed
>> >
>> > Parrot built, and all tests passed for Solaris 10/x86 without libffi.
>>
>> Thank you for your efforts.
>>
>> > Since libffi assumes gcc, and I don't have gcc installed on that system,
>> > this is the only permutation I tested.
>>
>> I find that hard to believe. The README explicitly states support for
>> MSVC. Also, why would it need to use autoconf if it only supported
>> gcc?
>
> Yes, there are undoubtedly some exceptions to my sweeping
> over-generalization. I don't know anything about the Windows build. There
> are also a few other comments sprinkled in the source about AIX, HP, and
> SGI that may or may not be out of date.  I don't know.
>
> What I do know is that I have never gotten it to build with anything other
> than gcc, and the defaults seem to assume gcc.  Here are some specific
> observations:
>
> The autoconf-generated Makefile unconditionally contains gcc-specific
> options, such as:
>
>        AM_CFLAGS = -Wall -g -fexceptions
>
> the header file ffi_common.h contains gcc-specific attributes:
>
>       typedef unsigned int UINT8  __attribute__((__mode__(__QI__)));
>
> Manually fixing those issues,
>
> On Solaris/SPARC, the compilation fails on the asm() statements
> in src/sparc/ffi.c with:
>
>    "src/sparc/ffi.c", line 476: syntax error before or at: :
>
> On Solaris/x86, the compilation fails with
>
>    "./include/ffitarget.h", line 88: undefined symbol: FFI_DEFAULT_ABI
>
> I have not pursued it any further.
>
> My conclusion is that there are specific exceptions in the code, but the
> default is to assume gcc.
>
> --
>    Andy Dougherty              doughera at lafayette.edu
>

Upon closer inspection, the MSVC support appears to be effected by
wrapping the command line tools with gcc->msvc argument translators.
Cruft^WNifty.


More information about the parrot-dev mailing list