On Mon, Jan 23, 2012 at 03:13:07PM +0100, Dimitry Andric wrote:
> On 2012-01-23 13:31, Kostik Belousov wrote:
> >On Mon, Jan 23, 2012 at 01:25:04PM +0100, Dimitry Andric wrote:
> ...
> >There is a typo in the second or condition, should it be gcc46 both times ?
>
> Ah, sorry, that was a copy/paste error.
>
>
> >Anyway, the reason to answer this message is two ask the for seemingly
> >unreasonable approach of matching compiler type/version based on the
> >driver name. This completely precludes anybody from using gcc installed
> >not from the ports tree.
>
> Yes, that is indeed the big problem with this approach. Although I
> think the number of people that hand-build gcc, and not use the ports
> will be rather low. :)
>
>
> >Could the tests performed based on the driver version information
> >instead of name ?
>
> Probably, but then you would have to run "${CC} --version" plus some
> sed/awk'ing from bsd.sys.mk. That could add quite some extra forking
> during buildworld.
>
> It may be easier to add a COMPILER_FLAVOR (just an example name) setting
> in make.conf, which can be set independently of CC, CXX and so on, to
> tell which specific variant of gcc, clang etc you want to use.
>
> The processing of that setting could happen in either sys.mk or
> bsd.sys.mk, or be separated out to a bsd.compiler.mk, for instance.
>
> In each .mk file or Makefile that needs it, the ${CC:T:M:foo} == "foo"
> comparisons can then be replaced with ${COMPILER_FLAVOR} == "foo".I agree that even such palliative measures are more useful and provides better future-proof of the build system then current name matching.
pgpU5MxlXeK5Q.pgp
Description: PGP signature
