Re: Dependency problem?

[Andrew Moise]

On Fri, Jul 11, 2003 at 02:27:54PM -0400, Larry Jones wrote:
> Andrew Moise writes:
> > 
> >   So I just noticed that when I apply the context-diff patch that I sent
> > to the list a couple of days ago (which changes a handful of files in
> > diff/), the cvs binary isn't rebuilt until I do 'make clean' or 
> > 'rm src/cvs'.  Changing a file in src/ (commit.c, at least) _does_
> > result in the executable being rebuilt, of course.
> 
> It's always worked fine for me -- are you sure you were running make in
> the top-level directory rather than in the src subdirectory?

  Yep.

> The
> Makefile in the src directory should define a cvs_DEPENDENCIES variable
> that includes the ancillary libraries (including the diff library) which
> is then referenced as a dependency for cvs$(EXEEXT), so the binary
> should be rebuilt if it's older than the diff library.

  Ah.  In that case, I've discovered the problem.  From src/Makefile:

cvs$(EXEEXT): $(cvs_OBJECTS) $(cvs_DEPENDENCIES)

  (... good ...)

cvs_DEPENDENCIES = $(cvs_client_objects)

  (... uh-huh ...)

cvs_client_objects =

  (... oh dear.)
  Does configure make your Makefile like this, too?  A little poking
around in the autoconf stuff enlightened me not at all.

> >   (BTW, I can't help but notice the silence which greeted my earlier
> > mail... this is the right list to send patches to, no?  Or was that patch
> > so offensive that no one felt the need to comment on it? :-)
> 
> Right place, but the diff code is imported from GNU diff, so we usually
> avoid touching it (or even understanding it).  Hopefully, the GNU diff
> maintainer (who is known to hang around here) will eventually get around
> to looking at it and have some words of wisdom for us.

  That's a sensible approach, yes.  The patch I sent is the wrong
thing (TM), since it adds a bit of special-case nonsense to the diff
code, but as I said I don't have a sufficient understanding of the code
to remove the newline-addition hackery, and the current behavior is
definitely problematic.
  The problem _is_ specific to cvs; diff is not affected.