Re: [Boost-build] Consistent toolset versioning requirements :(

From: Vladimir Prus (ghost_at_[hidden])
Date: 2006-10-11 04:18:23

On Tuesday 10 October 2006 17:59, David Abrahams wrote:
>   error: gcc initialization: parameter 'version' inconsistent
>   error: no value was specified in earlier initialization
>   error: an explicit value is specified now
>
> We've discussed this problem before.  Any site administrator who
> invokes "using gcc ;" in site-config.jam prevents users from
> initializing gcc with a version.  Any site administrator who adds a
> new "using" invocation to site-config.jam (even a versioned one)
> potentially breaks the user-config.jam of any users.
>
> The only safe thing for everyone concerned is to never invoke "using"
> without a version identifier:
>
>     using gcc : default ;
>
> but after all, isn't that just redundant?  The system could simply do
> that for me.  And finally, it won't work for the Python toolset, for
> example, which expects that 2nd argument to be a real, meaningful
> toolset version.
>
> Why do we still have this requirement that "using" be invoked
> consistently?

I think there were technical reasons. Specifically, some flag settings depends 
on toolset setting and so must be a condition, so that they are not used for 
other toolset.

If we have

     flags gcc.compile WHATEVER <toolset>gcc      : ......
     flags gcc.compile WHATEVER <toolset>gcc,<toolset:gcc-version>3.4 : ......

both flag settings will be used, which means that default-initialize gcc will 
add his settings to all version-initiialized gccs. Now, we have a mechanism 
to say:

     flags gcc.compile WHATEVER
               <toolset>gcc,<toolset:gcc-version>      : ......
     flags gcc.compile WHATEVER 
                <toolset>gcc,<toolset:gcc-version>3.4 : ......

So removing the consistency requirement should be possible. I'm a bit worried 
about "gcc" and "gcc-4.1" being names of exactly the same compiler, but 
perhaps that's fine.

Code-wise, all that's needed is change common.check-init-parameters to avoid 
the check, use the value-less syntax for features that are not specified.

As usual, help in doing that would be appreciated. If you don't have the time, 
can you file an issue?

Thanks,
Volodya