Re: AC_REQUIRE problems

[Dan Manthey]

Ah, I do stand corrected on several accounts.


On Tue, 15 Feb 2005, Paul Eggert wrote:

> Dan Manthey <address@hidden> writes:
>
> > Broadly speaking, the act of configuration is that of setting variables to
> > describe the build environment.
>
> Autoconf often sets the variables by putting them into a file, e.g.,
> something like config.h.  "make" could do this just as easily.
        Yes.  Then again, this would become the _only_ way of storing
results, which seems like a loss.

>
> > If there is no way to request a test before it is needed, then there
> > is no benefit to running it asynchronously.
>
> True, but most of autoconf usage is a simple static dependency graph.
> Autoconf has the power to do dynamic dependencies, but in practice
> this feature is relatively rarely used (and is often used wrongly).
> Hardly anybody would care if it went.
        I would care quite a bit, and surely it's a common case that
several alternatives are tried.  Currently, this constitutes dynamic
dependencies, though I suppose that could be worked around.

> > The only reason we're discussing in now anyway is that it comes
> > along with using Make for dependencies.
>
> No, I'm discussing it because "configure" Is Way Too Slow.
        Ah, true.  I'm curious what is known about the speed of Make in
comparison.  It seems that it can actually create just as many or more
child processes than a shell script.  Is that not the major factor in
speed?

> > (2) Have you ever tried putting a here document in Make?
>
> That part could be done by the shell before invoking "make".
        One of the major advantages of here documents is that they need
never be explicitly named nor cleaned up.  Communicated a possibly _very_
large number of files to the called Make would not have this advantage.
Great care, at least, would be needed, I think.

>
> Admittedly this all would be some work to do.  But Autoconf really
> should do a better job with parallelism.  I'm now seeing many cases
> where it takes longer to run "configure" than to run the following
> "make"!  It's getting ridiculous.
>
        Do you think that the comparatively small idea of handling
dependencies with functions will have any speed benefits?

-Dan