Re: 1.6.1 todo (was: Re: Patch: FYI: PR 304)

[Alexandre Duret-Lutz]

>>> "Tom" == Tom Tromey <address@hidden> writes:

 >>>>>> "adl" == Alexandre Duret-Lutz <address@hidden> writes:
 adl> I don't understand all the issues involved in depcomp.  For
 adl> instance, why does it computes dependencies in the background?

 Tom> Performance.  It's probably the case that it is faster to run the two
 Tom> compilations in parallel than to run them sequentially.

 Tom> This might be a mistake; we might find strange bugs doing this. 

It looks like PR 307 is one of them.

[...]

 adl> extra6     -  fails on Tru64 too, see the discussion with Nicolas
 adl> Joly on bug-automake.

 Tom> This is the CONFIG_SITE thing, right?  Let's just check in your
 Tom> proposed change (either of them).

No.  The ansi5.test failure was about site configuration, but
it's fixed in CVS (with ./configure --quiet).  

extra6.test is a different issue where Tru64 Make magically
create empty directories in the build tree.  It can be fixed by
teaching distdir how to merge directories from . and $(srcdir).
I've sent a patch for this.  Actually, this is a new feature
which fixes extra6.test as a side-effect.

 Tom> I don't think we need to worry about all the cases that
 Tom> autoconf does.

I don't understand what you are referring to.

 adl> pr87       -  a distcheck issue I've not investigated

 Tom> This is closed.  I thought there was another problem referring to it,
 Tom> but I couldn't find it.

Sorry, this list is all about _testcases_ which fails on Solaris.
pr87.test does fail, but this is probably unrelated to PR/87.

 adl> ccnoco     -  PR/310
 adl> pr224      -  PR/310
 adl> subobj6    -  PR/310
 adl> yacc8      -  PR/310
 adl> yaccvpath  -  I've not looked at it.  Another VPATH horror, I
 adl> guess.
 adl> None of them would fail when using GNU make (with CVS Automake).

 Tom> If we can support VPATH for SunOS (I usually write "Solaris", since
 Tom> "SunOS" forever means SunOS 3.x/4.x to me), then that would be nice.
 Tom> However, I don't think this is a 1.6.1 issue.  For now we could just
 Tom> require GNU make in these tests.

Fixing PR/310 will takes as much time as adding
$require_gnu_make to these tests :)

 Tom> I suspect we'll have trouble making VPATH work reliably.  Maybe we
 Tom> could have some assertion that says "I want VPATH to work on Solaris,
 Tom> warn me if I somehow break it".

-Wsolaris-vpath-issues :)

 Tom> Then we could disable features (e.g., nobase_) which are
 Tom> not easy or possible to make work in that scenario.

 Tom> That's a lot to do.  It's easier to just prohibit this.  Still, if
 Tom> someone feels like doing the work, I'm for it.

 adl> subdir5    -  XPASS.  I've not looked at this one

 Tom> Hari fixed this, though the patch isn't in yet.

When I awaken this morning I wondered why Hari's patch would fix
this subdir5.test failure.  (Yep, I've finally bought the new
ePillow.)  This is somethign that happens with Solaris make but
not with GNU make [1], how can it be related to AUTOHEADER and
AUTOCONF?

[1] http://sources.redhat.com/ml/bug-automake/2002/msg00209.html
-- 
Alexandre Duret-Lutz