Re: [PATCH] Fix dummy module being included even when not needed

[Benjamin Cama]

Le vendredi 08 novembre 2013 à 09:06 -0700, Eric Blake a écrit :
> On 11/08/2013 08:31 AM, Benjamin Cama wrote:
> > My goal is to avoid building any library at all when I am only using
> > gnulib for some utils like gitlog-to-changelog or git-version-gen
> > (furthermore when I don't even check for a compiler at all; my project
> > is shell scripting only).
> 
> GNU Autoconf is another such project (all use of gnulib is limited to
> non-code modules); right now it just manually copies files because
> gnulib-tool doesn't quite do the right thing (see the 'make
> gnulib-update' rule encoded into autoconf's cfg.mk).

Thanks for the hint. I looked at the needed files, and appart from
config.guess, config.sub and install-sh (in the havelib-tests module),
which are not really part of gnulib anyway, all the other ones are
included in modules not depending on any compiled code. Maybe Autoconf
would need a separate new module for the aforementioned files.

> > I thought that this was the purpose of this
> > test, but i was mistaken. Is there a way to do what I want? I don't
> > think so. Do you think this would be a worthwhile possibility?
> 
> Absolutely - patches are welcome to make gnulib-tool useful in a
> non-code setting.

What about the following one? My criterion for not building the library
is “no conditional am code and no include nor link directive”. You still
need to pass "--avoid dummy" not to get the library.

I would also advocate a more radical approach not to have a dummy module
at all, but I am not sure it's not needed in all use cases. Furthermore,
its criterion for including the dummy package is even less strict than
mine (no lib_SOURCES augmentation only); should I rather stick to the
same definition? If anyone could explain me the use of a dummy library,
I could have a better idea of what to do.

Regards,
-- 
Benjamin Cama <address@hidden>

0001-Don-t-build-a-library-if-no-code-is-compiled.patch
Description: Text Data