[Bug-gnulib] Re: What is AC_CHECK_FUNCS_ONCE?

[Bruno Haible]

Hi Akim,

In http://mail.gnu.org/archive/html/bison-patches/2003-03/msg00057.html
you wrote:

> I would like to warn the authors of this _ONCE stuff.

That's me. Thanks for the warning :-)

> some macro named AC_CHECK_FUNCS_ONCE which is not an Autoconf macro,
> and therefore should not be named AC_.

You know better than me how to realize this macro in Autoconf. Can you
put AC_CHECK_FUNCS_ONCE into autoconf, or - even better - make
AC_CHECK_FUNCS behave like it by default?

The raison d'être of this macro is that we have 8 modules in gnulib
which each need AC_CHECK_FUNCS(isascii), and similarly for other
functions, headers, and declarations. Without any countermeasure a
package which invokes these autoconf macros will:

  1) have an unnecessarily large configure file. Remember, ca. 1 KB
     for each AC_CHECK_FUNCS invocation. This bloats up every GNU
     package in the long run.

  2) cause the user/installer to wonder when he sees:

     $ ./configure
     ...
     checking for isascii... yes
     ...
     checking for isascii... (cached) yes
     ...
     checking for isascii... (cached) yes
     ...
     checking for isascii... (cached) yes
     ...
     checking for isascii... (cached) yes
     ...
     checking for isascii... (cached) yes
     ...
     checking for isascii... (cached) yes
     ...
     checking for isascii... (cached) yes
     ...

     My thoughts when I see this from a single configure file (not even
     from different configure files in subdirectories!) is that it
     must be insane underneath.

AC_CHECK_FUNCS_ONCE is an answer to 1). You could fix 2) by just
omitting the message if the value is in the cache.

> Now, sit back, and look at this:
> 
> @example
> $ @kbd{./configure --with-sprintf}
> checking for sprintf... no
> @end example
> 
> @noindent
> although @code{sprintf} @emph{is} present!
> 
> What happened is that Autoconf knows that checking for a function
> requires a compiler for the current language (here, C), so it actually
> expands something similar to the following @file{configure.ac}:
> 
> @example
> AC_INIT
> 
> AC_ARG_WITH([fprintf])
> if test "x$with_fprintf" = xyes; then
>   AC_PROG_CC
>   AC_CHECK_FUNCS(fprintf)
> fi
> 
> AC_ARG_WITH([sprintf])
> if test "x$with_sprintf" = xyes; then
>   AC_CHECK_FUNCS(sprintf)
> fi
> @end example

This is a bug in autoconf, because

  1) You haven't documented which AC_* macros are callable inside
     'if' and which are not? (Some certainly are, like AC_DEFINE,
     right?)

  2) Why should a configure file test for 'fprintf' when it doesn't
     need the result? Or in the case of error.m4:

       AC_FUNC_ERROR_AT_LINE
       dnl Note: AC_FUNC_ERROR_AT_LINE does AC_LIBSOURCES([error.h, error.c]).
    -  if test $ac_cv_lib_error_at_line = no; then
    -    jm_PREREQ_ERROR
    -  fi
    +  jm_PREREQ_ERROR
     ])

    Why should the configuration spend time to check the prerequisites
    of a file that will not be compiled on the particular platform?
    Gimme a real reason, other than "autoconf currently doesn't work
    that way".

Bruno