Re: stamp-h* and make distcheck

[Akim Demaille]

>>>>> "Ralf" == Ralf Corsepius <address@hidden> writes:

Ralf> If using AM_CONFIG_HEADERS located in subdirectories, make
Ralf> distcheck breaks because of not correctly handling stamp*-files.

Ralf> ..  make[1]: Leaving directory `stamp-test-0.0/=build' ERROR:
Ralf> files left after distclean: ./include/stamp-h1 make: ***
Ralf> [distcheck] Error 1

Ralf> This is with autoconf/cvs from subversion as of this morning and
Ralf> automake/CVS from cygnus/sources as of Sunday.

Ralf> Example enclosed below.

Thanks for the excellent bug report!  Your tarball is precious.

But did this ever worked?  I tried with 1.5, and it fails the same
way.

The problem, AFAICS is that you don't have a include/Makefile.am,
which is probably where Automake would have cleaned it.  Adding one
makes it succeed.

It turns out I don't like either the way distclean works.  IMHO, what
is created by config.status is to be removed by config.status, and in
particular from the top level Makefile.am.

So I'm not willing to handle this issue now, but to delay it: I will
equip 2.53's config.status with --clean, (something texi2dvi should do
too IMHO), and have Automake next generation, relying on 2.53, stop
cleaning up config.status's files, but ask it to do it.

Is it OK for everyone?

Of course one problem is running make distclean when there is no
config.status.  But if you start thinking about such cases, you ought
to think about the case when there is no Makefile either but there is
config.status.  In either case, one loses.