This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: [Patch] Fix ac_c99_complex configury
Mark Mitchell wrote:
> Benjamin Kosnik wrote:
>
>>> the use of the new __builtin_clog exposed a serious weakness in our
>>> configury: the check for ac_c99_complex (and then for
>>> _GLIBCXX_USE_C99_COMPLEX), uses the C compiler together with
>>> AC_TRY_COMPILE: we are basically checking for prototypes but C (vs C++)
>>> doesn't care and this means that the test basically *always* passes. I
>>> propose to simply use AC_TRY_LINK, here and I'm finishing regtesting
>>> the
>>> below.
>>
>> You can't use any link tests in this macro, it's used by crosses and
>> native toolchains.
>
> I thought you just couldn't use run tests. Why can't you use link tests?
At least this is not *totally* obvious, fiuu! Thanks: often I feel a
little stupid when I have to do with the configury bits and now I can
fool myself that I'm not more naive than our release manager, great! ;) ;)
Another, more serious, tought I entertained during the last weeks,
actually, is whether we could *finally* remove the include/backward
stuff! I believe that for 4.1 would be absolutely *correct* wrt our
deprecation/removal policy.
Opinions?
Paolo.