Re: [Caml-list] Segv while compiling ocaml-3.04

Re: [Caml-list] Segv while compiling ocaml-3.04

[Christophe Raffalli]

Luc Maranget a écrit :
> 
> >
> >
> > I am wondering if it is make of ocamlrun that triggers the segv as the problem
> > appears only when make is calling ocamlrun but not when calling ocamlrun
> > directly
> >
> > Does anyone knows of such a make bug ?
> >
> Yes I have similar problems diring various compilations in Caml.
> I concluded that my Linux (RedHat 7.2) and make is guilty
> 
> % make --version
> GNU Make version 3.79.1, by Richard Stallman and Roland McGrath.
> Built for i386-redhat-linux-gnu
> Copyright (C) 1988, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 2000
>         Free Software Foundation, Inc.
> This is free software; see the source for copying conditions.
> There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
> PARTICULAR PURPOSE.
> 
The core (as said in my previous post) is produced by ocamlrun not make.
 
> BTW, I discovered the following workaround : ``make -n | sh ''

I can try that !

> > --
> > Christophe Raffalli
> 
> --Luc Maranget

-- 
Christophe Raffalli
Université de Savoie
Batiment Le Chablais, bureau 21
73376 Le Bourget-du-Lac Cedex

tél: (33) 4 79 75 81 03
fax: (33) 4 79 75 87 42
mail: Christophe.Raffalli@univ-savoie.fr
www: http://www.lama.univ-savoie.fr/~RAFFALLI
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Xavier Leroy]

Update on the gcc 2.96 issue:

A patch is now available, along with an updated source RPM and an
updated RedHat 7.2 binary RPM, all at the usual place:
        http://caml.inria.fr/ocaml/distrib.html

For those who are worried about it, the purity of the OCaml source
code isn't affected: the patch simply selects gcc compilation options
(-fomit-frame-pointer) that seem to work around the bug.

As to which versions of gcc are affected: gcc 2.96 from RH 7.2 and
Mandrake 8.1 are, I'm certain that 2.95.2 is not, and believe that 3.0
isn't either, but this has to be checked carefully.

A bug report was filed with RedHat.  (The bug was already in their
database, but I added a report for good measure.)  If someone knows
how to file a bug report for Mandrake 8.1 (which seems to be affected
as well), I can provide details.

This is actually the second time that RedHat's modified gcc breaks
OCaml.  I won't engage in RedHat bashing here, since that's rather
pointless.  We'll probably deal with this issue like we did the first
time: test the gcc version in configure and issue a strongly worded
warning, telling the users to continue at their own risks.  Just
refusing to compile for gcc 2.96 would be excessive, since apparently
not all versions of 2.96 have the bug...

- Xavier Leroy
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Pixel]

Xavier Leroy <xavier.leroy@inria.fr> writes:

[...]

> For those who are worried about it, the purity of the OCaml source
> code isn't affected: the patch simply selects gcc compilation options
> (-fomit-frame-pointer) that seem to work around the bug.

I find it quite funny that *adding* an optimisation makes things work. Since
2.96 (also true for 3.0), gcc has been much better at tail-call'ing,
tail-recursion. Most of the bugs appearing in 2.96 were caused by using "-O3"
together with "-fomit-frame-pointer". The solution was to *remove*
"-fomit-frame-pointer"

For information, i've had problems with ocaml 3.02 built with -O3
-fomit-frame-pointer. -O3 works nicely.

> As to which versions of gcc are affected: gcc 2.96 from RH 7.2 and
> Mandrake 8.1 are,

I've not been able to reproduce under 8.1 
(taking your srpm and removing the patch)

Mandrake and Redhat have been using gcc 2.96 for quite a long time now.
Mandrake 8.0 and redhat 7.1 have it. But beware that these gcc's are heavily
patched and evolved a lot.

[...]

> I'm certain that 2.95.2 is not, and believe that 3.0
> isn't either, but this has to be checked carefully.

3.0 is not very stable. Redhat's rawhide has switched to pre-3.1 cvs version.

About version gcc versions, check http://www.bero.org/gcc296.html
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Xavier Leroy]

> > As to which versions of gcc are affected: gcc 2.96 from RH 7.2 and
> > Mandrake 8.1 are,
> 
> I've not been able to reproduce under 8.1 
> (taking your srpm and removing the patch)

You need to look at the assembly code generated for the MD5Transform
function.  More explanations on the gcc bug, and a smaller test suite,
can be found in my bug report on RedHat's bugzilla (bug #57760).

> Mandrake and Redhat have been using gcc 2.96 for quite a long time now.
> Mandrake 8.0 and redhat 7.1 have it. But beware that these gcc's are heavily
> patched and evolved a lot.

Right.  This bug isn't in RH 7.1, and popped in RH 7.2.

> But beware that these gcc's are heavily patched and evolved a lot.

>From my point of view, they evolve from bad to worse :-)  
I miss the robust gcc 2.95.2 from RH 6.2 and Mandrake 7.2, which never
failed me.

- Xavier Leroy
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

[Caml-list] Segv while compiling ocaml-3.04

[Christophe Raffalli]

I failed to compile ocaml-3.04 under Mandrake Linux 8.12 (proc ATHLON 1500+
(1.3GHz :-) chipset VIIAKT

First I shoud say that  ocaml-3.03-alpha compiles with no problem (same machine
same time) !

After ./configure, make world crashes for 3.04 in 

../boot/ocamlrun ../boot/ocamlc -g -nopervasives -c oo.mli
make[1]: *** [oo.cmi] Erreur de segmentation
make[1]: *** Deleting file `oo.cmi'
make[1]: Quitte le répertoire `/usr/local/src/ocaml-3.04/stdlib'
make: *** [coldstart] Erreur 2

But then If I compile the file myself it works:

[root@boris ocaml-3.04]# cd stdlib
[root@boris stdlib]# ../boot/ocamlrun ../boot/ocamlc -g -nopervasives -c oo.mli

And then the next crash is in 

boot/ocamlrun boot/ocamlc -I boot -I utils -I parsing -I typing -I bytecomp -I
asmcomp -I driver -I toplevel -c parsing/parsetree.mli
make: *** [parsing/parsetree.cmi] Erreur de segmentation
make: *** Deleting file `parsing/parsetree.cmi'

And so on, after a dozen of similar crashes, I manage to end the make world.
They are no crashes during make opt
But the problem starts again with make ocamlopt.opt

very strange !
Is it a problem with gnu make that was not there before ?


-- 
Christophe Raffalli
Université de Savoie
Batiment Le Chablais, bureau 21
73376 Le Bourget-du-Lac Cedex

tél: (33) 4 79 75 81 03
fax: (33) 4 79 75 87 42
mail: Christophe.Raffalli@univ-savoie.fr
www: http://www.lama.univ-savoie.fr/~RAFFALLI
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Pixel]

Christophe Raffalli <Christophe.Raffalli@univ-savoie.fr> writes:

> After ./configure, make world crashes for 3.04 in 
> 
> ../boot/ocamlrun ../boot/ocamlc -g -nopervasives -c oo.mli
> make[1]: *** [oo.cmi] Erreur de segmentation
> make[1]: *** Deleting file `oo.cmi'
> make[1]: Quitte le répertoire `/usr/local/src/ocaml-3.04/stdlib'
> make: *** [coldstart] Erreur 2

at least it works here on cooker (mandrake devel version, post 8.1)

i've only been having some pbs with building for a location and installing
somewhere else (especially ld.conf and camlp4)

while i'm at it, would this be useful for someone else?
--------------------------------------------------------------------------------
--- ocaml-3.00/emacs/ocamltags.in.pix	Mon Nov 29 14:03:30 1999
+++ ocaml-3.00/emacs/ocamltags.in	Tue Dec 12 06:06:28 2000
@@ -1,4 +1,4 @@
-":" ; @EMACS@ -batch -l $0 "$@" ; status=$? ; : '--*-Emacs-Lisp-*--' <<';'
+":" ; @EMACS@ --batch --no-init-file --no-site-file -l $0 "$@" ; status=$? ; : '--*-Emacs-Lisp-*--' <<';'
 
 ;; Copyright (C) 1998 Ian Zimmerman <itz@transbay.net>
 ;;  This program is free software; you can redistribute it and/or
--------------------------------------------------------------------------------
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Christophe Raffalli]

I am wondering if it is make of ocamlrun that triggers the segv as the problem 
appears only when make is calling ocamlrun but not when calling ocamlrun
directly

Does anyone knows of such a make bug ?

-- 
Christophe Raffalli
Université de Savoie
Batiment Le Chablais, bureau 21
73376 Le Bourget-du-Lac Cedex

tél: (33) 4 79 75 81 03
fax: (33) 4 79 75 87 42
mail: Christophe.Raffalli@univ-savoie.fr
www: http://www.lama.univ-savoie.fr/~RAFFALLI
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Xavier Leroy]

> After ./configure, make world crashes for 3.04 in 
> 
> ../boot/ocamlrun ../boot/ocamlc -g -nopervasives -c oo.mli
> make[1]: *** [oo.cmi] Erreur de segmentation
> make[1]: *** Deleting file `oo.cmi'
> make[1]: Quitte le répertoire `/usr/local/src/ocaml-3.04/stdlib'
> make: *** [coldstart] Erreur 2
> 
> But then If I compile the file myself it works:
> 
> [root@boris ocaml-3.04]# cd stdlib
> [root@boris stdlib]# ../boot/ocamlrun ../boot/ocamlc -g -nopervasives -c oo.mli

We've observed similar hard-to-reproduce segmentation faults on a
Linux RedHat 7.2 machine.  After some tracking, it turns out to be a
bug in gcc version 2.96 20000731 -- the "unofficial" gcc that RH 7.2
uses; I don't know which version of gcc your Mandrake has.

For those who are interested, the bug is that gcc forgets to allocate
a stack frame in a leaf function, thus accessing the stack "below" the
stack pointer.  If there are no interrupts during the leaf function,
it still works; but any interrupt wipes the memory area just below the
stack pointer, and the function crashes...

We're trying to figure out a workaround, and will release updated RPMs
and SRPMs.

- Xavier Leroy
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Christophe Raffalli]

 
> We've observed similar hard-to-reproduce segmentation faults on a
> Linux RedHat 7.2 machine.  After some tracking, it turns out to be a
> bug in gcc version 2.96 20000731 -- the "unofficial" gcc that RH 7.2
> uses; I don't know which version of gcc your Mandrake has.

My bug is very easy to reproduce, but I indeed use gcc 2.96 (coming with
Mandrake 8.1 by default).
I can install gcc 3.0, will it work ?

-- 
Christophe Raffalli
Université de Savoie
Batiment Le Chablais, bureau 21
73376 Le Bourget-du-Lac Cedex

tél: (33) 4 79 75 81 03
fax: (33) 4 79 75 87 42
mail: Christophe.Raffalli@univ-savoie.fr
www: http://www.lama.univ-savoie.fr/~RAFFALLI
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Ian Zimmerman]

Xavier> We've observed similar hard-to-reproduce segmentation faults
Xavier> on a Linux RedHat 7.2 machine.  After some tracking, it turns
Xavier> out to be a bug in gcc version 2.96 20000731 -- the
Xavier> "unofficial" gcc that RH 7.2 uses; I don't know which version
Xavier> of gcc your Mandrake has.

Xavier> For those who are interested, the bug is that gcc forgets to
Xavier> allocate a stack frame in a leaf function, thus accessing the
Xavier> stack "below" the stack pointer.  If there are no interrupts
Xavier> during the leaf function, it still works; but any interrupt
Xavier> wipes the memory area just below the stack pointer, and the
Xavier> function crashes...

Xavier> We're trying to figure out a workaround, and will release
Xavier> updated RPMs and SRPMs.

<unlurk>
Is it worth polluting the source for the sake of an out-and-out vendor
bug?  I'd say no.
</unlurk>

-- 
Ian Zimmerman, Oakland, California, U.S.A.
GPG: 433BA087  9C0F 194F 203A 63F7 B1B8  6E5A 8CA3 27DB 433B A087
In his own soul a man bears the source
from which he draws all his sorrows and his joys.
Sophocles.
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Florian Hars]

On Thu, Dec 20, 2001 at 02:49:17PM -0800, Ian Zimmerman wrote:
> Is it worth polluting the source for the sake of an out-and-out vendor
> bug?  I'd say no.

There is a clean solution to this problem, as implemented in the
configure-script of mplayer (severly edited, the full code is available
at http://mplayerhq.hu/cgi-bin/cvsweb.cgi/main/configure):

if test "$_skip_cc_check" != yes ; then
  echocheck "$_cc version"
  cc_version=`$_cc -v 2>&1 | sed -n 's/^.*version \([aegcygnustp-]*[0-9.]*\).*$/\1/p'`
  case $cc_version in
    2.95.[2-9]|2.95.[2-9].[0-9]|3.[0-9]|3.[0-9].[0-9])
      cc_version="$cc_version, ok"
      ;;
    *)
      cc_version="$cc_version, bad"
      cc_verc_fail=yes
      ;;
  esac
  echores "$cc_version"
  if test "$cc_verc_fail" ; then
    cat <<EOF

*** Please downgrade/upgrade C compiler to gcc-2.95.x or gcc-3.x version! ***

        GCC 2.96 IS NOT AND WILL NOT BE SUPPORTED BY US !
EOF
    die "Bad gcc version"
  fi
else
cat <<EOF


Yours, Florian.
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Ian Zimmerman]

itz> Is it worth polluting the source for the sake of an out-and-out
itz> vendor bug?  I'd say no.

Florian> There is a clean solution to this problem, as implemented in
Florian> the configure-script of mplayer (severly edited, the full
Florian> code is available at
Florian> http://mplayerhq.hu/cgi-bin/cvsweb.cgi/main/configure):

Yes, I support something like this wholeheartedly :)

-- 
Ian Zimmerman, Oakland, California, U.S.A.
GPG: 433BA087  9C0F 194F 203A 63F7 B1B8  6E5A 8CA3 27DB 433B A087
In his own soul a man bears the source
from which he draws all his sorrows and his joys.
Sophocles.
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[William Chesters]

Florian Hars writes:
 > On Thu, Dec 20, 2001 at 02:49:17PM -0800, Ian Zimmerman wrote:
 > > Is it worth polluting the source for the sake of an out-and-out vendor
 > > bug?  I'd say no.
 > 
 > There is a clean solution to this problem, as implemented in the
 > configure-script of mplayer [...]
 > 
 >   if test "$cc_verc_fail" ; then
 >     cat <<EOF
 > 
 > *** Please downgrade/upgrade C compiler to gcc-2.95.x or gcc-3.x version! ***
 > 
 >         GCC 2.96 IS NOT AND WILL NOT BE SUPPORTED BY US !
 > EOF
 >     die "Bad gcc version"

Luckily the ocaml team are much too sensible to go this way.  For
those who are tempted, take a look at

    http://www.linuxworld.com/site-stories/2001/1214.mplayer.html

"... boorish developers who think of themselves as a sort of
intellectual jeunesse doree ... The attitude I mentioned earlier was
noticeable in the comments on the project's page at freshmeat, ... in
the MPlayer team's remarks about gcc 2.96 ... Actually, the team's
infamous anti-Red Hat rhetoric has been toned down slightly. At Red
Hat's request, some false statements the developers made about 2.96
have been removed from MPlayer's Web site. The team still lets you
know at every opportunity what they think of Red Hat, which gets
tiresome fast."

Basically this approach is absolutely guaranteed to give the
impression that you are over-fussy, arrogant, inconsiderate of real
users' needs etc.  RedHat is too big to ignore and get away with it:
every time they pull some stunt like this, lots of developers waste
lots of time and credibility attempting to convince users that it's
All Red Hat's Fault---eventually succeeding only to discover that the
users still blame them not RH.

My &euro;1E-2.
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Sven]

On Fri, Dec 21, 2001 at 10:25:52AM +0100, William Chesters wrote:
> Florian Hars writes:
>  > On Thu, Dec 20, 2001 at 02:49:17PM -0800, Ian Zimmerman wrote:
>  > > Is it worth polluting the source for the sake of an out-and-out vendor
>  > > bug?  I'd say no.
>  > 
>  > There is a clean solution to this problem, as implemented in the
>  > configure-script of mplayer [...]
>  > 
>  >   if test "$cc_verc_fail" ; then
>  >     cat <<EOF
>  > 
>  > *** Please downgrade/upgrade C compiler to gcc-2.95.x or gcc-3.x version! ***
>  > 
>  >         GCC 2.96 IS NOT AND WILL NOT BE SUPPORTED BY US !
>  > EOF
>  >     die "Bad gcc version"
> 
> Luckily the ocaml team are much too sensible to go this way.  For
> those who are tempted, take a look at
> 
>     http://www.linuxworld.com/site-stories/2001/1214.mplayer.html
> 
> "... boorish developers who think of themselves as a sort of
> intellectual jeunesse doree ... The attitude I mentioned earlier was
> noticeable in the comments on the project's page at freshmeat, ... in
> the MPlayer team's remarks about gcc 2.96 ... Actually, the team's
> infamous anti-Red Hat rhetoric has been toned down slightly. At Red
> Hat's request, some false statements the developers made about 2.96
> have been removed from MPlayer's Web site. The team still lets you
> know at every opportunity what they think of Red Hat, which gets
> tiresome fast."
> 
> Basically this approach is absolutely guaranteed to give the
> impression that you are over-fussy, arrogant, inconsiderate of real
> users' needs etc.  RedHat is too big to ignore and get away with it:
> every time they pull some stunt like this, lots of developers waste
> lots of time and credibility attempting to convince users that it's
> All Red Hat's Fault---eventually succeeding only to discover that the
> users still blame them not RH.

But i remember something similar about gcc 2.7.x or at least one of them that
didn't work well with ocaml back then (96-97 time frame i think). There was
just a small test in the configure script and if it detected a not working
gcc, there was a notice of problems with it and it just stopped.

You may very well be able to say the same thing in a more milder way, and it
would not cause any kind of problem, i think.

Friendly,

Sven Luther
-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr

Re: [Caml-list] Segv while compiling ocaml-3.04

[Luc Maranget]

> <unlurk>
> Is it worth polluting the source for the sake of an out-and-out vendor
> bug?  I'd say no.
> </unlurk>

Hum, I would ne be that confident on the origin of the bug.
Anyway, only Makefiles sould get altered.


--Luc


-------------------
Bug reports: http://caml.inria.fr/bin/caml-bugs  FAQ: http://caml.inria.fr/FAQ/
To unsubscribe, mail caml-list-request@inria.fr  Archives: http://caml.inria.fr