Re: AC_INIT design bug or automake thinko?

[Bruno Haible]

Stepan Kasal wrote:
> But there is no need to set the version during the run of configure;
> it is known when Autoconf runs. ... Actually, Autoconf requires that the 
> version is a literal, not shell variable.

This is only a minor issue, but still: Why does autoconf want to have
the version number at autoconf-time? It has no influence on the shell
script code that needs to be expanded into 'configure'.

Sam asked it in http://lists.gnu.org/archive/html/autoconf/2005-03/msg00054.html
and got no answer.

> Could you replace your version.sh with version.m4, which would
> contain:
> m4_define([gettext_version], [1.2.3])
> ?
> 
> Then you would have:
> m4_include([version.m4])
> AC_INIT([GNU gettext], gettext_version, address@hidden)

Thanks for the tip. The major problem, however, is the one of syntax.
Even to an experienced programmer, having to write an m4 function call
in order to declare a value is _unnatural_. The natural syntax for
declaring values is
      IDENTIFIER = VALUE

Makefile.am gets this right. sh script snippets got this right as well.
Things like
      AC_INIT([hello], [2.1])
or
      AM_GNU_GETTEXT_VERSION([0.14.6])
are programmer-unfriendly hacks that are only motivated by the lack of
a central parametrization file standard in GNU. (We do have a standard means
for parametrization per-directory; this is the Makefile.am. But nothing
that has the right syntax and exists only once per package.)

A combination of your technique with the right syntax could look like
this:

  esyscmd([ if ! test -f version.sh; then cd ..; fi; ./version.sh; echo 
"m4_define([gettext_version], [$version])" ])
  AC_INIT([GNU gettext], gettext_version, address@hidden)

Bruno