Re: egcs-971105 config.guess and gcc/config.guess have diverged

Re: egcs-971105 config.guess and gcc/config.guess have diverged

[Kaveh R. Ghazi]

 > From: Jim Wilson <wilson@cygnus.com>
 > 
 > 	Ooops, what I meant to say was egcs' config.guess and
 > 	gcc-2.7.90's config.guess have diverged.  And I think it does matter
 > 	given the recent failures I had on alphaev56. 
 > 
 > Yes it matters.  The goal is to try to keep them in sync, but
 > occasional short term divergences will be unavoidable.  We
 > unfortunately did not have any merges from the FSF since early Aug, so
 > anything added to the FSF gcc2 sources after that will not be in EGCS.
 > The alphaev56 support was added after that point, and hence is not in
 > EGCS yet.
 > 
 > This stuff is so new that we may not be able to support it in the
 > initial EGCS release.
 > 
 > Incidentally, after starting a branch for the initial EGCS release,
 > the first thing we did was merge with the most recent FSF snapshot, so
 > the post release EGCS sources will be up-to-date with the FSF as of
 > the Oct 21 snapshot.  The EGCS snapshots have since been coming from
 > the release branch.  As soon as the release is out, we will make
 > snapshots from the mainline sources again, and the alphaev56 support
 > will be there.
 > 
 > Jim


Jim,

	I don't think the problem is, as you imply above, that support
is missing.  The issue is that ev5* support is there and broken.  At
least I couldn't bootstrap my ev5 with egcs out of the box.

	If the support doesn't work at all, then at the very least,
config.guess should set the target as alpha-dec-osf4.0 so a working
compiler could be built, assuming ev5's are backward compatible.

	I think it _would_ actually work except that config.guess
guesses wrongly.  When I supply the target to configure and build
alpha-dec-osf4.0 (least common denominator) or alphaev5-dec-osf4.0b, it
works.  When I rely on the target config.guess chooses,
alphaev56-dec-osf4.0, it crashes during bootstrap. 

	So fixing config.guess should be all that's required.  If you
merged in the alpha diffs (quite short) from testgcc-971104, it would
be fine.  (Don't use the one from testgcc-971111 because the snapshot
has the wrong copy.)

		--Kaveh
--
Kaveh R. Ghazi				Project Manager
ghazi@caip.rutgers.edu			ICon CMT Corp.

Re: egcs-971105 config.guess and gcc/config.guess have diverged

[Richard Henderson]

On Wed, Nov 12, 1997 at 04:48:23PM -0500, Kaveh R. Ghazi wrote:
> 	I don't think the problem is, as you imply above, that support
> is missing.  The issue is that ev5* support is there and broken.  At
> least I couldn't bootstrap my ev5 with egcs out of the box.

It's not ev5 thats the problem but ev56 (since it is ev5's that I have),
almost certainly caused by the relatively untested byte-word extension.

We should probably not enable MASK_BYTE_OPS by default in configure.in
until we've got it fixed.


r~

Re: egcs-971105 config.guess and gcc/config.guess have diverged

[Jeffrey A Law]

  In message < 19971112150154.29821@dot.cygnus.com >you write:
  > On Wed, Nov 12, 1997 at 04:48:23PM -0500, Kaveh R. Ghazi wrote:
  > > 	I don't think the problem is, as you imply above, that support
  > > is missing.  The issue is that ev5* support is there and broken.  At
  > > least I couldn't bootstrap my ev5 with egcs out of the box.
  > 
  > It's not ev5 thats the problem but ev56 (since it is ev5's that I have),
  > almost certainly caused by the relatively untested byte-word extension.
  > 
  > We should probably not enable MASK_BYTE_OPS by default in configure.in
  > until we've got it fixed.
Kaveh -- can you try removing MASK_BYTE_OPS in gcc/{configure,configure,in}
for the alpha56 target, then try to bootstrap the ev56?

If that works we can use it for the release; for the mainline we've already
sucked in the work from ev56 work from the FSF tree.

jeff

Re: egcs-971105 config.guess and gcc/config.guess have diverged

[Kaveh R. Ghazi]

 > From: Richard Henderson <rth@dot.cygnus.com>
 > 
 > On Wed, Nov 12, 1997 at 04:48:23PM -0500, Kaveh R. Ghazi wrote:
 > > 	I don't think the problem is, as you imply above, that support
 > > is missing.  The issue is that ev5* support is there and broken.  At
 > > least I couldn't bootstrap my ev5 with egcs out of the box.
 > 
 > It's not ev5 thats the problem but ev56 (since it is ev5's that I have),
 > almost certainly caused by the relatively untested byte-word extension.
 > 
 > We should probably not enable MASK_BYTE_OPS by default in configure.in
 > until we've got it fixed.
 > r~



 > From: Jeffrey A Law <law@cygnus.com>
 > 
 >   In message < 19971112150154.29821@dot.cygnus.com >you write:
 >   > On Wed, Nov 12, 1997 at 04:48:23PM -0500, Kaveh R. Ghazi wrote:
 >   > > 	I don't think the problem is, as you imply above, that support
 >   > > is missing.  The issue is that ev5* support is there and broken.  At
 >   > > least I couldn't bootstrap my ev5 with egcs out of the box.
 >   > 
 >   > It's not ev5 thats the problem but ev56 (since it is ev5's that I have),
 >   > almost certainly caused by the relatively untested byte-word extension.
 >   > 
 >   > We should probably not enable MASK_BYTE_OPS by default in configure.in
 >   > until we've got it fixed.
 > Kaveh -- can you try removing MASK_BYTE_OPS in gcc/{configure,configure,in}
 > for the alpha56 target, then try to bootstrap the ev56?
 > 
 > If that works we can use it for the release; for the mainline we've already
 > sucked in the work from ev56 work from the FSF tree.
 > jeff


Gentlemen,

	I am still not convinced when you say I am using an ev56.  (I
know it sounds kind of silly that I can't identify what system I am
using.  I am logging in remotely to a guest account so I can't actually
look at the physical box to check anything.)

	From "sizer -c" I get:
 > cpu             "DEC_KN20AA"
in case that helps ID the chip.

	How do you explain the discrepancy between
testgcc-971104/config.guess (which says alphaev5-dec-osf4.0b) and
egcs/config.guess (which says alphaev56-dec-osf4.0)?

	In a message from rth:
( http://www.cygnus.com/ml/egcs/1997-Nov/0434.html ) I was told that the
gcc2 config.guess gets it right.  So I say the entire problem lies in
the fact that egcs/config.guess gets my chip wrong which of course
causes the bootstrap to fail.



	Regarding testing without setting MASK_BYTE_OPS, I think I've
already verified that avoiding setting this works, in a round about way. 
What I did was bootstrap with "configure alphaev5-dec-osf4.0b" to test
if that worked (it did) since I suspected that my system was in fact an
ev5, not an ev56 as egcs claimed. 

	Looking in configure, the difference between alphaev5 and ev56
is only the MASK_BYTE_OPS.  So having a successful bootstrap
configured as alphaev5-dec-osf4.0b should give us confidence that
taking out the MASK_BYTE_OPS makes it work on my system.

	However I don't think this looking at the problem this way is
the right issue to examine, as I explained above.

		--Kaveh
--
Kaveh R. Ghazi				Project Manager
ghazi@caip.rutgers.edu			ICon CMT Corp.

Re: egcs-971105 config.guess and gcc/config.guess have diverged

[Jim Wilson]

	Ooops, what I meant to say was egcs' config.guess and
	gcc-2.7.90's config.guess have diverged.  And I think it does matter
	given the recent failures I had on alphaev56. 

Yes it matters.  The goal is to try to keep them in sync, but occasional
short term divergences will be unavoidable.  We unfortunately did not have
any merges from the FSF since early Aug, so anything added to the FSF gcc2
sources after that will not be in EGCS.  The alphaev56 support was added
after that point, and hence is not in EGCS yet.

This stuff is so new that we may not be able to support it in the initial
EGCS release.

Incidentally, after starting a branch for the initial EGCS release, the first
thing we did was merge with the most recent FSF snapshot, so the post release
EGCS sources will be up-to-date with the FSF as of the Oct 21 snapshot.
The EGCS snapshots have since been coming from the release branch.  As soon as
the release is out, we will make snapshots from the mainline sources again,
and the alphaev56 support will be there.

Jim

Re: egcs-971105 config.guess and gcc/config.guess have diverged

[Kaveh R. Ghazi]

 > 	In egcs-971105, config.guess and gcc/config.guess have diverged.
 > Does it matter?
 > 
 >                 --Kaveh

	Ooops, what I meant to say was egcs' config.guess and
gcc-2.7.90's config.guess have diverged.  And I think it does matter
given the recent failures I had on alphaev56. 

		--Kaveh
--
Kaveh R. Ghazi				Project Manager
ghazi@caip.rutgers.edu			ICon CMT Corp.

egcs-971105 config.guess and gcc/config.guess have diverged

[Kaveh R. Ghazi]

	In egcs-971105, config.guess and gcc/config.guess have diverged.
Does it matter?

		--Kaveh
--
Kaveh R. Ghazi				Project Manager
ghazi@caip.rutgers.edu			ICon CMT Corp.