Re: gettext-tools with g-prefix (was Re: xgettext with UI files support)

[Jonathan Perkin <jperkin%joyent.com@localhost>]

* On 2015-08-30 at 13:02 BST, Richard PALO wrote:

> Le 30/06/15 20:12, Richard PALO a écrit :
> > Le 30/06/15 20:02, Jonathan Perkin a écrit :
> >> * On 2015-06-30 at 18:29 BST, Richard PALO wrote:
> >>
> >>> Le 30/06/15 19:09, Jonathan Perkin a écrit :
> >>>>> Yes, but how to check with pbulk reports on seeming successes but
> >>>>> with mixed pkgsrc and native gettext tools (that is, *bad*).
> >>>>
> >>>> I'm not sure what you mean here.  Can you give some concrete examples
> >>>> of failure so that we can pinpoint exactly what the cause is?
> >>>>
> >>>
> >>> See my reply from 17 Jun 2015 23:39:50 +0200:
> >>>> For a example, take a look at databases/gnome-mime-data wrt gmsgfmt.
> >>
> >> By 'concrete examples' I was hoping for build logs, setup, etc.  It's
> >> quite hard to figure out what the problem is without the ability to
> >> reproduce it or see any data.
> >>
> >> That said, piecing together information from various mails, I think
> >> what you're saying is that, despite the correct tool being pulled into
> >> tools/bin, the /usr/bin/g* version is being chosen by configure first,
> >> rendering our $PATH ordering irrelevant.
> >>
> >> That doesn't explain the request to add more USE_TOOLS (if they were
> >> missing then we should see failures on SmartOS?), but a simple
> >> solution for this should just be that we add some TOOLS_ALIASES to
> >> gettext.mk, i.e.:
> >>
> >> diff --git a/mk/tools/gettext.mk b/mk/tools/gettext.mk
> >> index b8d3b80..977a59e 100644
> >> --- a/mk/tools/gettext.mk
> >> +++ b/mk/tools/gettext.mk
> >> @@ -148,6 +148,7 @@ TOOLS_DEPENDS.${_t_}?	=${_TOOLS_DEP.gettext-tools}:../../devel/gettext-tools
> >>  TOOLS_CREATE+=	${_t_}
> >>  TOOLS_FIND_PREFIX+=	TOOLS_PREFIX.${_t_}=${TOOLS_DEPENDS.${_t_}:C/:.*//}
> >>  TOOLS_PATH.${_t_}=	${TOOLS_PREFIX.${_t_}}/bin/${_t_}
> >> +TOOLS_ALIASES.${_t_}=	g${_t_}
> >>  .    endif
> >>  .  endif
> >>  .endfor
> >>
> >> This will create g-prefixed symlinks in .tools/bin for all the gettext
> >> tools, and should ensure they get picked before the native versions.
> >>
> >> If I'm still misunderstanding the problem, it might be best to hop
> >> onto Freenode/#pkgsrc IRC rather than going back and forth via the
> >> list any further.
> >>
> >> Cheers,
> >>
> > 
> > the proposed patchset for devel/gettext-tools correctly creates the g-prefixed program names in $PREFIX/bin and 
> > puts the native gnu named links in PKGGNUDIR as with other gnu packages in pkgsrc.
> > 
> > the patchset for mk/tools/gettext.mk does the alias as well for the g-prefixed names.
> > 
> > the advantage is now the gettext-tools *correctly* hide the native counter parts on systems such
> > as SunOS that put the g-prefixed name in /usr/bin, which is precisely what gnu configure seems to 
> > frequently look for.
> > 
> 
> I'd like to push this now along with the ~115 packages needing msgmerge and/or xgettext added to USE_TOOLS.  This should allow enough time to find the rest via bulk if not yet done.

Please could you post a patch for review?  The previous ones had
unrelated changes included, making it hard to review, especially as I
still don't understand why the USE_TOOLS changes are necessary.  I
didn't see an answer to my question about why we don't see failure on
SmartOS - if packages require those tools, and we do not ship them in
/usr/bin, then surely we'd see bulk build failures?

Thanks,

-- 
Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com