Re: caching variables

[Ralf Wildenhues]

Hello Sam,

* Sam Steingold wrote on Wed, Sep 02, 2009 at 05:26:46PM CEST:
> I want to cache several dependent variables, and I cannot figure out
> how to do that correctly.

* Sam Steingold wrote on Wed, Sep 02, 2009 at 07:51:20PM CEST:
> does the cache variable name matter?
> e.g., ac_cv_have_readline is not declared cached in
> http://clisp.cvs.sourceforge.net/*checkout*/clisp/clisp/src/m4/readline.m4
> but I do see it in config.cache.

Generally, configure currently saves all variables matching *_cv_*
(glob notation), and considers a variable cached also if its value
is empty.  It uses AS_VAR_TEST_SET to test if the variable is cached.
Being very precise, we don't document that you can use cache variables
without one of the AC_CACHE_{CHECK,VAL} macros, but I'm not sure whether
a future Autoconf version may make this a requirement.

> specifically:
> 
> AC_DEFUN([AC_FOO],[dnl
> AC_ARG_WITH([foo],
> AC_HELP_STRING([--with-foo],[use this FOO installation]),
> [ac_cv_use_foo="$withval"], [ac_cv_use_foo=default])

By overwriting the cache variable with the value obtained from
--with-foo, you cannot tell any more whether the variable was cached
or you just read a command line argument.

> ac_cv_have_foo=no

Here you just overwrite the value obtained above again.

> if test "$ac_cv_use_foo" = "no";

This will then always be true.

> then AC_MSG_NOTICE([not checking for FOO])


> I want all the ac_cv_* vars to be cached.
> how do I arrange for that?
> i.e., should I put the "if $ac_cv_foo_command bar;" inside AC_CACHE_CHECK?
> would those _nested_ AC_CACHE_CHECKs work?

You should have one
  AC_CACHE_CHECK(MESSAGE, CACHE-ID, COMMANDS)

or one
  AC_CACHE_VAL(CACHE-ID, COMMANDS)

for one of the CACHE-IDs.  The important thing is that you don't set
that CACHE-ID outside of COMMANDS, and the COMMANDS should only set the
CACHE-ID (and the other, dependent ones, I guess).

Thus, this:

> FOO=$ac_cv_foo_command; AC_SUBST(FOO)
> FOO_BAR=$ac_cv_foo_bar; AC_SUBST(FOO_BAR)
> FOO_ZOT=$ac_cv_foo_zot; AC_SUBST(FOO_ZOT)

should happen outside (after) the AC_CACHE_{CHECK,VAL}.

Hope that helps.  I think there are a couple of examples in gnulib.

Cheers,
Ralf