Re: RFE: configure -> dependency list on exit.

[Hugh Sasse Staff Elec Eng]

On Thu, 24 Feb 2005, Dan Manthey wrote:

> I have a modest proprosal that may capture the desired features, and
        [...]
> We've been discussing in another thread on this list that Autoconf tests
> are usually preformed unconditionally because of issues with calling
> prerequisite macros conditionally and the inability of m4 to understand sh
> if-else blocks.  If we presume for the moment that problem is solved
> (which I think is a fair assumption given the discussion of the other
> thread), then we could have in configure.ac something like the following:
>
> AC_PROG_LEX
> if test $LEX = :; then
>  # No lex was found, so we can't use yacc.
>  AC_MSG_END([A lexer is needed for yacc.]
> AS_PACKAGE_SUGGEST([flex 2.5.x],[http://gnu.org/whereever_flex_is.html])
> AS_PACKAGE_SUGGEST([lex]))

        [OT] I don't understand this extra ')' but ...
> else
>  AC_PROG_YACC
>  ...
> fi

I like this, it is clear.  Could the shell logic be produced by a
Macro, though, to help people get it right?
> Beyond the handling of conditional macros, this only requires AC_MSG_END
> (which I assume would be a trivial matter outputting messages to a temp

That would be a good solution.

> file which is later output), and possibly AS_PACKAGE_SUGGEST, a fascility
> quite orthogonal to the logic that determines which tests are run.  It
> would have the job of appropriately describing a recommended place to get
> a package.  This could (possibly) do anything from simple string
> formatting to some sort of database lookup.

Yes, that would be a good addition, though I think it will meet
resistance because of the documentation overhead.  Anwyway, having
it as a standard macro allows future flexibility, such as the Free
Software Directory adopting GNOWSYS or enquiring of some future vast
artificial intelligence :-).
> The only problem this doesn't solve is that of generating a DEPENDENCIES
> files.  If this is the critical need, then how about:  (This is clearly
> not a final proposal since this is terribly ugly.)
>
> AC_DEPENDS([A lexer is needed for yacc.]
> AS_PACKAGE_SUGGEST([flex 2.5.x],[http://gnu.org/whereever_flex_is.html])
> AS_PACKAGE_SUGGEST([lex]),
>           [AC_PROG_LEX],
>           [test $LEX != :],
> [AC_PROG_YACC
> ...
> ])
>
> At this point the Autoconf macro accounting mechanism can note that
> AC_PROG_LEX was called to prepare to test whether to run AC_PROG_YACC.
> Personally, I don't think the added complexity of this second choice is
> warranted by its small benefit of generating a rather brittle description
> that's already captured in the prose output by AC_MSG_END.

I don't think we need to generate a DEPENDENCIES file, although it
would be nice to generate supporting documentation automatically.
I think that provided the AS_MSG_END macro produces a clear message
with dependency information when configure is run, and does so at
the end as you say, this would be a very significant improvement.
> I'm not convinced that the dependencies are complex enough to need
> anything more than sane use of if-else, but in the few cases where it is,
> that can be expressed with the form of AC_REQUIRE that takes code to
> execute as well as an identifier to have provided:
>
> AC_DEFUN([AC_PROG_YACC_THAT_NEEDS_LEX],
> [AC_REQUIRE([_YACC_CHECK_LEX],
> [AC_PROG_LEX
> if test $LEX = :; then
> AC_MSG_END([Yacc needs lex])
> fi])
> if test $LEX != :; then
>  AC_PROG_YACC
> fi])

If the case for a failure is always a : (nicely a NO-OP in ksh at
least, and, it seems, bash) then there is scope for something like

#define  AC_IF_FAILED(LEX) if test $LEX != :; then
#define  AC_ELSE_FAILED(LEX) else
#define  AC_END_FAILED(LEX) fi

to clarify what the 'else' and 'fi' refer to, and automate the
construction of the test clause.  [I always seem to munge test
commands!]

> (Okay, that's a poor example, but I hope it shows that the needed tools
> already exist.)

Thank you.  I think you have cut through a lot of the confusion that
was springing up.  I like it when people manage to simplify things
so neatly.
> -Dan
        Thank you,
        Hugh