Re: CC can be a program name *with arguments*

[Ralf Wildenhues]

Hi Gary,

* Gary V. Vaughan wrote on Tue, Mar 01, 2005 at 02:04:51PM CET:
> 
> Welcome back :-)

Thanks!  (freezing)

> > * Gary V. Vaughan wrote on Fri, Feb 25, 2005 at 01:21:14PM CET:
> >>Bob Friesenhahn wrote:
> >>
> >>My thoughts exactly.  cc_basename is only ever used in case statements,
> >>so lets set it with:
> >>
> >>  cc_basename=`$echo X"$compiler" | $Xsed -e 's%^.*/%%;s%[  ].*$%%'`
> > 
> > 
> > You guys ever encountered a system with compilers named
> >   gcc-3.4.1
> >   gcc-3.4.2
> >   icc-8.1
> >   ...
> 
> Yep, but you would need to build a different libtool for each compiler.
> Relying on the compiler checks performed on one working for another is
> likely to cause trouble.

Sure.  But we are talking about m4/libtool.m4, right?
This happens at configure time.

I want, in a libtoolized project,
  ../configure CC='pgcc-78.9 -foo-option'
to succeed in choosing Portland compiler options.  No multiple compilers
involved here.

> > I think the original patch was useful, and have yet to see an instance
> > where it causes trouble (e.g., two compilers starting with the same
> > prefix;  that could be disambiguated by leaving out the * at the end of
> > the case pattern on a case-by-case basis).
> 
> >From a maintenance point of view, keeping the fix isolated in 1 line of
> code will save us from forgetting the * in case statements in future
> patches.

Well, yes.  Maybe it could be worth doing s#-.*## to kill version
appendices.  But that would be more limiting than necessary IMVHO
(nobody specified version appendices had to look like this).

> > Independently still, cc_basename should be calculated like above
> > (not like is done currently, and also not like Gary suggested,
> > 
> > |  cc_basename=`$echo X"$compiler" | $Xsed -e 's%^.*/%%;s%[         
> > ]*.*$%%'`
> > 
> > which looks quite bogus to me, as it deletes everything.)
> 
> D'oh!  Good catch.  :")
> 
>   cc_basename=`$echo X"$compiler" | $Xsed -e 's%^[    ]*\([^  ]*\).*$%\1%'`

Why?  People writing
  CC=' gcc'
by accident are people that get run over by accident.  :->

> > Maybe we need to adjust for things like ccache and distcc here, so the
> > thingy takes the second argument (I still have not looked at how they
> > are typically used together with libtool, so this is just speculation)
> > for better matching.  Dunno if anybody has ever used those together with
> > non-GCC compilers and with libtool.
> 
> Maybe worth a scribble in TODO?

Yep.

Regards,
Ralf