Re: FYI: test old m4 interface [280]

[Ralf Wildenhues]

Hi Gary,

* Gary V. Vaughan wrote on Fri, Sep 30, 2005 at 06:28:57PM CEST:
> Ralf Wildenhues wrote:
> >
> >And another one. :-/
> 
> Excellent... the wall between here and 2.0 is slowly getting thinner...

I dearly hope so.  :)

> >The first failure is fixed by the first hunk in the patch below:
> >the sub-configure is invoked by the toplevel configure
> >(the AC_CONFIG_SUBDIRS _is_ working iff the libltdl package is in the
> >directory named `libltdl').
> >
> >The extra configure caused(!) autotools reruns which then changed(!)
> >the symlinked-to files!
> 
> Ahah!  Thankyou... I saw that go by in one of my pre-submission tests,
> and then couldn't figure out how to make it happen again.
> 
> >Maybe we should disallow `libtoolize --ltdl' without `--install'?
> 
> That's effectively the same as having --ltdl always do an implicit
> --install.  Not necessarily a bad idea.  We could do that for the
> alpha release, and see if it causes any problems during testing...

Erm, I meant --copy, not --install.  Sorry.

> >OK, with that in place, it gets through configuration and builds the
> >libltdl object files, then fails with
> >| pmake
> >| cd libltdl && pmake libltdlc.la
> >| ...
> >| make: don't know how to make ./dlopen.la. Stop
> >| *** Error code 2
> >
> >Why?  The usual non-GNU make foo vs ./foo problem:
> >| LT_DLLOADERS =  ./dlopen.la
> >| dlopen.la: $(dlopen_la_OBJECTS) $(dlopen_la_DEPENDENCIES)
> >| libltdl_la_DEPENDENCIES = $(LT_DLLOADERS) $(LTLIBOBJS)
> 
> Okay, I think I can fix that quite easily.  Patch to follow.

Cool.

> >But _then_, I don't like the need to call aclocal at all.  I tried to
> >eliminate it, the second ugly hunk below is what I could come up with.
> 
> Cool!  I like that alot :-)
> 
> >Seems we have introduced a silent dependency on aclocal >= 1.2.  :-)
> 
> Which you just removed?  Or elsewhere?

No, not elsewhere.  But I meant something different:
If I apply my proposed patch, our test will be fine, but users will
still have to use `aclocal'.  Shouldn't we provide an AM_CONDITIONAL
replacement for them in that case?  Is it possible at all to provide 
this only in the case where the user does not use aclocal -- is there
some trick to do this?

> >Do you think it is ok to rip off Automake code like this in the test?
> 
> Sure.  It's all GPL.  I would add a comment to say where it came from
> though, if only incase we want to resynch it in a few years...

OK.

For now, I have applied the first hunk only, since the rest requires
more discussion.

Cheers,
Ralf

        * tests/old-m4-iface.at (Makefile.in): Do not reconfigure
        libltdl directory.

Index: tests/old-m4-iface.at
===================================================================
RCS file: /cvsroot/libtool/libtool/tests/old-m4-iface.at,v
retrieving revision 1.2
diff -u -r1.2 old-m4-iface.at
--- tests/old-m4-iface.at       30 Sep 2005 10:09:58 -0000      1.2
+++ tests/old-m4-iface.at       2 Oct 2005 08:52:37 -0000
@@ -113,7 +113,7 @@
 all: $(TARGETS)
 
 libltdl/libltdlc.la:
-       cd libltdl && ./configure && $(MAKE)
+       cd libltdl && $(MAKE)
 
 module.la: module.lo
        $(LTLINK) module.lo -module -avoid-version -rpath /dev/null