Re: [PATCH 2 of 2] Drop mkinstalldirs, use install-sh -d instead

["Matthias Andree" <matthias.andree@xxxxxx>]

Am 08.07.2009, 17:59 Uhr, schrieb Patrick Welche <prlw1@xxxxxxxxx>:

> On Tue, Jul 07, 2009 at 10:48:07PM -0000, Matthias Andree wrote:
> > Drop mkinstalldirs, use install-sh -d instead.
> >
> > This requires that we use autoconf 2.60 and automake 1.10 or newer to
> > avoid race conditions in older versions of mkinstalldirs and install-sh.
>
> I totally agree on using $(MKDIR_P), I'm just surprised that there  
> doesn't
> seem to be a AC_PROG_MKDIR_P in configure.ac, and why not also use
> $(MKDIR_P) in the intl/Makefile.in too? (That's just from eye-balling
> the diff...)

Hi Patrick,

it's implicit called through AM_INIT_AUTOMAKE, so no explicit reference  
there. Try autoconf --trace=AC_PROG_MKDIR_P to see where it comes from :-)

However, the intl/* and particularly po/Makefile.in.in diffs make me  
reconsider this patch. I'm no longer sure we should actually apply it at  
this time. Reason for my new reservation is that even gettext 0.17 (the  
current upstream version) still relies on mkinstalldirs. Rocco's observed  
"mkinstalldirs not reinstalled" failure is caused by mutt using a derived  
gettext version rather than the original (with MUTT_ prefixes to the  
regular macros) - autoreconf believes that gettext weren't used.

Perhaps we should shelve this part and rather make sure that we get mutt's  
gettext into shape, aka. see if the local patches are still needed with  
newer gettext versions - and then see to updating these scripts to recent  
versions...

Another issue to consider is the inclusion/updating policy for the get*  
scripts. Currently, we are including (in code and repository) the full  
code. The advantage is that the maintainer doesn't have to care,  
disadvantages are that it entails manual maintenance overhead and can  
cause m4/, po/ and intl/ stuff to be updated and cause half-bogus  
"M)odified" status in hg status and inadvertent commits.
OTOH, with some core committers not using the latest gettext version, it  
seems more of a requirement to keep copies in the repo.

Perhaps all of this was a premature and misfit attempt of mine to get the  
standard GNU stuff out of the repo.

Opinions? Should we pursue the removal of GNU standard scripts from the  
repo? I don't fear differing gettext versions too much.

--
Matthias Andree