Libguestfs - May 2013 - Re: hivex: revert much of " Build and test Perl bindings out-of-tree." (887466794a40f57b84ce32524e1a650d6674c7e6)?

While trying to build the official Debian packages for libguestfs 1.22,
I noticed that instead of working around ExtUtils::MakeMaker's
limitations, I had made quite a mess of building the Perl bindings. I
have tried from different sides to fix things, but could not get it
quite right:

- EU::MM expects Makefile.PL to be in the same directory as the generated
  Makefile (We work around that using sed.)

- In principle, there seem to be two ways to do things:

  1. Call make -f builddir/Makefile-pl from srcdir (that's what we do
     now), override INST_*. Problems with this:
     - XS is still built in srcdir.
     - there is actually code in ExtUtils::Install which is used from the
       generated Makefile-pl that bases its decisions where to install
       *.pm files on whether there are any files within ./blib/arch/
       (hard-coded)! I have found no way to override this.

  2. Call make -f Makefile-pl from builddir. I tried that and thought
     that I would be able to patch absolute paths for source files
     (*.xs, lib/*) and generated files (*.c, *.xs, blib/*) into the
     generated Makefile-pl. Much of a perfectly good weekend went to
     waste over that attempt.

I realized that I was actively working against ExtUtils::MakeMaker's
design. Separating source and build trees has never been a concern. Even
if I got it right once, I'd expect things to break in interesting ways
in the future.
>From a look at the documentation, Module::Build does not seem to have a
solution, either.

I give up and propose that we go back to the hackish, yet simple
previous solution and adopt that for libguestfs as well:

    -[ $(srcdir) != $(builddir) ] && cp -rsu $(abs_srcdir)/.
$(builddir)/.

I agree that this is ugly, but works and it won't hurt when building
libguestfs from within the source directory.

Cheers,
-Hilko

Is no patch attached or following this message?  I don't seem to see
it in my email nor on the mailing list archives on the web.

Anyway, a straightforward 'git revert' is fine, ACKed.

If the patch is more involved than that, I would prefer to see it first.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
virt-df lists disk usage of guests without needing to install any
software inside the virtual machine.  Supports Linux and Windows.
http://people.redhat.com/~rjones/virt-df/

* Richard W.M. Jones:
> Is no patch attached or following this message? I don't seem to see it
> in my email nor on the mailing list archives on the web.
>
> Anyway, a straightforward 'git revert' is fine, ACKed.
No, other than expressing a bit of frustration, I just wanted to know if
also using that symlink hack for libguestfs would be okay.

Cheers,
-Hilko