Re: [Boost-testing] Toolset Autoconfiguration [was: Martin Wille x86_64 V2 - python]

From: David Abrahams (dave_at_[hidden])
Date: 2007-03-26 15:35:07

on Mon Mar 26 2007, Roland Schwarz <roland.schwarz-AT-chello.at> wrote:

> Roland Schwarz wrote:
>> So perhaps the autoconfigure should use this function too, and then
>> pass only the toolset (i.e. gcc) and version (i.e. 3.4.4) to the using
>> rule? At least this is what "using" is expecting to get as input.
>
> Too fast :-( This won't work.
>
> The function relies on the subfeatures already been defined (which
> happens inside "using".)

Right.

> So the notation xxx-yyy-zzz means yyy and zzz being subfeatures of
> xxx (whithout a defined order). They are implicitly looked up by
> value.
>
> But if this is true, the autoconfig poses a chicken-egg problem,
> and none can tell how to parse the xx-yyy-zzz.

Right.  If someone would establish some protocols that would limit
the representation of toolset names, it might be possible, but so far
nobody has done that.

> Indeed, gcc-mingw-3.4.4 produces
> <toolset>gcc/<toolset-gcc:flavor>mingw/<toolset-gcc:version>3.4.4
> which also is a correct expansion.
>
> But the autoconfigure never can know which is the "version" part.
>
> I think I begin to understand why Volodya insisted on leaving the
> "toolset=" less syntax in place.
>
> Hmm, if we had a rule similar to expand-no-defaults which tries to
> deduce the toolset and version, but does not bomb out if a subfeature
> cannot be found this could be used to find the correct input to
> the using rule.
>
> What do you think?

That helps, but still doesn't answer all the open questions raised in
my other post.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com