[Boost-build] gcc version/flavour detection (Was: [Boost-cvs] CVS: boost/tools/build/v2/tools gcc.jam , 1.82, 1.83)

From: Vladimir Prus (ghost_at_[hidden])
Date: 2006-10-25 12:47:15

On Wednesday 25 October 2006 18:34, Rene Rivera wrote:
> Vladimir Prus wrote:
> > On Wednesday 25 October 2006 16:56, Rene Rivera wrote:
> >> Jürgen Hunold wrote:
> >>> I propose to call "SHELL $(command) --version" for support of
> >>> non-standard gcc names. Please find the patch attached.
> >>
> >> I figured I would forget something :-) Applied, thanks.
> >
> > And my nightly build is broken too:
> >
> > /tmp/boost-build/build/boost-build/build/feature.jam:268: in
> > implied-subfeature from module feature
> > error: "ubuntu________"
>
> That doesn't look right. It should be "ubutun_4_0_3_1ubuntu5" (what an
> idiotically redundant tag Ubuntu uses).

When I add the following debug prints:

        ECHO "XXXXX" $(command-info) ;
        ECHO "YYYYY" $(command-info[2]:L) ;
        ECHO "ZZZZZ" [ regex.replace $(command-info[2]:L) "[ .-:]" "_" ] ;

I get:

      XXXXX 4.0.3 Ubuntu 4.0.3-1ubuntu5
      YYYYY ubuntu 4.0.3-1ubuntu5
      ZZZZZ ubuntu______-_ubuntu_

Which probably means that regex.transform misbehaves. But aren't we trying too 
hard here? I think detecting, and adding to target path, the version of the 
compiler, is a good thing. Though I'm not sure if rebuild when minor version 
changes is a good thing.

At the same time, I think encoding full "flavour" starts to look unnecessary. 
The most important use case for that is mingw/cygwin -- which we need to 
support. But for Linux, adding the name of the distro to target path looks 
unncessary for most users. Can we just MATCH for mingw/cygwin?

- Volodya