Re: AC_REQUIRE problems

[Stepan Kasal]

Hi Paul,

On Sat, Feb 12, 2005 at 11:22:01PM -0800, Paul Eggert wrote:
> No, you'd invoke "make" just once (just as you invoke "configure" just
> once).  "configure" would invoke "make" with an appropriate makefile.
> The makefile would do all the real work.
> 
> Make (1) handles dependencies well,

Yes, this can handle the _dependencies_ well.

But make parses the whole makefile(s) at the begining and it's not easy
to modify the dependencies as make has learnt them.

I'm afraid that some of the tests might need other tests' results as
their input.  How can they find out?  To influence the current run, you
have to create files, or modify their timestamps.  Or you can run a
secondary make.  Both techniques are very slow, so you should use them
only in exceptional cases.

In short, my doubts is the same as in my previous mail:
> > Is make indeed suitable declarative language for this purpose?
> > Will we have a small file for each macro which was run, in order to tell
> > Makefile that it doesn't have to run it again?
> > Or will we write the results to a makefile, which will the make have to
> > re-read after each test?

Last, but not least: if you solve this problems in theoretical level, you
probably won't find a way to translate current configure.am's to the new
tool.  In fact, you would be starting a new project, which will have to
struggle hard for popularity.  As you require that the users (I mean
maintainers using your new tool) program in a declarative language, your
situation is very hard.

I think the solution which uses functions to implement dependencies is
much more practical. Current sources (configure.ac) will be usable without
any change.

Have a nice day,
        Stepan