Bruno Haible <br...@clisp.org> writes:
>> +uninstall-hook: uninstall-relocwrapper
>> +uninstall-relocwrapper:
>> +if RELOCATABLE_VIA_LD
>> + @:
>> +else
>> + if test $(RELOCATABLE) = yes; then \
>> + case '$(EXEEXT)' in \
>> + .bin*) ;; \
>> + *) $(MAKE) uninstall EXEEXT=.bin$(EXEEXT) ;; \
>> + esac; \
>> + fi
>> +endif
>
> Hmm, I don't like targets that contain empty commands in the default
> case (when no relocatability is requested): it looks like bloat. I'm
> adding a conditional 'RELOCATABLE_VIA_WRAPPER' that should allow you
> to simplify this code.
OK, thanks.
> Btw, in gnulib contributions, please indent shell statements by 2 spaces,
> not 4 spaces.
OK, I updated my local copy.
>> --- a/doc/relocatable-maint.texi
>> +++ b/doc/relocatable-maint.texi
>> @@ -145,6 +145,14 @@ foo_LDFLAGS = `$(RELOCATABLE_LDFLAGS) $(bindir)`
>> endif
>> @end example
>>
>> +Also, only in the @file{Makefile.am} at the top level of your source
>> +tree, add the following. Replace @samp{build-aux}, if necessary, by
>> +the directory where @file{gnulib-tool} places auxiliary build tools:
>> +
>> +...@example
>> +include build-aux/relocwrapper.am
>> +...@end example
>
> I would prefer to avoid Makefile.am fragments and 'include' statements.
> The reason is that here we start to get dependencies on the file layout
> of the package.
>
> The automake documentation says:
>
> There are two forms of `include':
>
> `include $(srcdir)/file'
> Include a fragment that is found relative to the current source
> directory.
>
> `include $(top_srcdir)/file'
> Include a fragment that is found relative to the top source
> directory.
>
> and your example is not one of these. Then, if the package's configure.ac
> invokes AC_CONFIG_AUX_DIR, the string 'build-aux/' needs to be replaced
> accordingly. This does not make it easier for the package maintainer.
I've changed this to
include $(top_srcdir)/build-aux/relocrwapper.am
in my loacl copy.
> Instead, I would better see an AC_SUBSTed and AM_SUBST_NOTMAKEd variable
> that expands to
>
> uninstall-hook:
>
> if RELOCATABLE_VIA_WRAPPER is false, and to
>
> uninstall-hook: uninstall-relocwrapper
> uninstall-relocwrapper: ; case '$(EXEEXT)' in .bin*) ;; *) $(MAKE)
> $(AM_MAKEFLAGS) uninstall EXEEXT=.bin$(EXEEXT) ;; esac
>
> if RELOCATABLE_VIA_WRAPPER is true. Unfortunately, I don't see how to
> reduce this to a single line; this would make it unnecessary to rely on
> AM_SUBST_NOTMAKE - but how?
>
> Such a variable could be used in any Makefile.am, without thinking about
> how to access 'build-aux/'.
I agree that this is a nicer approach, but it does not work. If
Automake cannot see that uninstall-hook is in use when it
processes Makefile.am, then it does not call it. So this
approach will only work in `Makefile.am's that already have an
uninstall-hook.
> Btw, make recursions in automake generated Makefile.ins use
> $(MAKE) $(AM_MAKEFLAGS)
> not plain
> $(MAKE)
Thanks, I've fixed that in my local copy.
Here is another approach that I had not thought of before.
What if we added the following to
modules/relocatable-prog-wrapper:
Makefile.am:
if RELOCATABLE_VIA_WRAPPER
uninstall-hook: uninstall-relocwrapper
uninstall-relocwrapper:
if test $(RELOCATABLE) = yes; then \
case '$(EXEEXT)' in \
.bin*) ;; \
*) cd $(top_builddir) && \
$(MAKE) $(AM_MAKEFLAGS) EXEEXT=.bin$(EXEEXT) \
AM_MAKEFLAGS='$(AM_MAKEFLAGS) EXEEXT=.bin$(EXEEXT)' \
uninstall ;; \
esac; \
fi
.PHONY: uninstall-relocwrapper
endif
I haven't tested it yet, and I'm a little worried about whether
this is unwarranted abuse of AM_MAKEFLAGS.
--
Ben Pfaff
http://benpfaff.org
