Re: Remove the autoreconf-provided INSTALL, so that we regenerate it.

[Jim Meyering]

Ralf Wildenhues <address@hidden> wrote:

> * Jim Meyering wrote on Sun, Nov 11, 2007 at 09:16:46PM CET:
>>
>>      Remove the autoreconf-provided INSTALL, so that we regenerate it.
>>      * GNUmakefile (dummy): Remove INSTALL.
>
> Thanks.
>
>> --- a/GNUmakefile
>> +++ b/GNUmakefile
>> @@ -47,13 +47,14 @@ include Makefile
>>  # for others: rerunning autoconf and recompiling everything isn't cheap.
>>  # The $(MAKE) clean is required, to remove all traces of the previous
>>  # version string, which would otherwise cause a few test failures.
>> +# Remove the autoreconf-provided INSTALL, so that we regenerate it.
>>  ifeq (0,$(MAKELEVEL))
>>    _is-dist-target = $(filter dist% alpha beta major,$(MAKECMDGOALS))
>>    ifneq (,$(_is-dist-target))
>>      _curr-ver := $(shell build-aux/git-version-gen $(srcdir) .version)
>>      ifneq ($(_curr-ver),$(VERSION))
>>        $(info INFO: running autoreconf for new version string: $(_curr-ver))
>> -      dummy := $(shell rm -rf autom4te.cache; autoreconf -i && $(MAKE) 
>> clean)
>> +      dummy := $(shell rm -rf autom4te.cache; autoreconf -i && $(MAKE) 
>> clean && rm -f INSTALL)
>
> I have a question or two for this line: First, can we do `autoreconf -v
> -i', so autotools rerun triggers are visible?  I ask because I sometimes
> bootstrap with an automake that is not the first in $PATH (released
> Automake vs. development version), so it helps seeing when I forgot to
> adjust $PATH at time.

Sure, I'll add it.

> Second, is the `rm -rf autom4te.cache;' needed?

Yes, most definitely.
Otherwise, the version update (the whole goal of that rule)
never propagates out to e.g., config.status.
There's probably a better way to get the same effect, but
I haven't looked for it.

> Third, isn't there an `cd $(srcdir)' or two missing?

Now that you mention it, I do see at least two are missing.