Re: [PATCH] There's no point in setting up the libltdl build when we're using an installed copy

[Ralf Wildenhues]

Hello Ari,

thanks for the report and patch.

* Ari Entlich wrote on Mon, Mar 08, 2010 at 05:58:59AM CET:
> The subject says it all. Comments appreciated.

A very minor nit: mails are easier to grok if the contents don't refer
to the subject.  Patches are a lot easier to understand with a
corresponding ChangeLog entry.  (Sorry, couldn't resist.)

I see two problems with this patch, but haven't tested it so far.
First, in some cases, _LTDL_MODE_DISPATCH may expand to the empty string
(or whitespace only), which will then cause a shell syntax error.

More fundamentally though, when we don't configure the subpackage at
all, then a 'make dist' won't package up the included libltdl, which
means the package tarball can't be reproduced.  Not totally sure about
this, but changing this would be an API change and require a big
warning, so probably a new configure switch.

Lastly, the change to the comment isn't fully correct (but this is
really a nit, too), since the _LTDL_MODE_DISPATCH macro also has effect
at m4/autoconf run time, that means it will have an effect even if only
inside a shell conditional branch that is not used at configure time.

Cheers,
Ralf

> --- a/libltdl/m4/ltdl.m4
> +++ b/libltdl/m4/ltdl.m4
> @@ -384,8 +384,11 @@ dnl Don't require this, or it will be expanded earlier 
> than the code
>  dnl that sets the variables it relies on:
>  _LT_ENABLE_INSTALL
>  
> -dnl _LTDL_MODE specific code must be called at least once:
> -_LTDL_MODE_DISPATCH
> +dnl _LTDL_MODE specific code must be called at least once if
> +dnl we're using the included copy of libltdl:
> +if test x"$with_included_ltdl" = xyes; then
> +  _LTDL_MODE_DISPATCH
> +fi