Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Alex Buell]

On Fri, 26 Jun 1998, Horst von Brand wrote:

> > [cc1 eats all memory and crashes]
> 
> I saw something like that with an egcs snapshot, but had no time to dig
> further.

Further tests with -O1 got the same results that -O2 did. Removing the -O
option from the Makefile in Lynx worked. There must be something strange
in LYMainLoop.c that is causing the optimiser to go amok. :o(

Cheers,
Alex
--
 /\_/\  Legalise cannabis now! 
( o.o ) Smoke some cannabis today!
 > ^ <  Peace, Love, Unity and Respect to all.

http://www.tahallah.demon.co.uk

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Andreas Schwab]

Alex Buell <alex.buell@tahallah.demon.co.uk> writes:

|> On Fri, 26 Jun 1998, Horst von Brand wrote:
|> 
|> > > [cc1 eats all memory and crashes]
|> > 
|> > I saw something like that with an egcs snapshot, but had no time to dig
|> > further.
|> 
|> Further tests with -O1 got the same results that -O2 did. Removing the -O
|> option from the Makefile in Lynx worked. There must be something strange
|> in LYMainLoop.c that is causing the optimiser to go amok. :o(

There is nothing strange.  It's just huge, at least after expanding all
the macros from <bits/string2.h>.  Use -D__NO_STRING_INLINES.

Andreas.

-- 
Andreas Schwab                                      "And now for something
schwab@issan.informatik.uni-dortmund.de              completely different"
schwab@gnu.org

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Alex Buell]

On 29 Jun 1998, Andreas Schwab wrote:

> |> Further tests with -O1 got the same results that -O2 did. Removing the -O
> |> option from the Makefile in Lynx worked. There must be something strange
> |> in LYMainLoop.c that is causing the optimiser to go amok. :o(
> 
> There is nothing strange.  It's just huge, at least after expanding all
> the macros from <bits/string2.h>.  Use -D__NO_STRING_INLINES.

*Sigh* I see. Do I really have to expand my machine's memory? :o(

Cheers,
Alex
--
 /\_/\  Legalise cannabis now! 
( o.o ) Smoke some cannabis today!
 > ^ <  Peace, Love, Unity and Respect to all.

http://www.tahallah.demon.co.uk

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Alex Buell]

On 27 Jun 1998, Andreas Jaeger wrote:

> glibc 2.0.94 uses optimized inline functions.  Inlining these
> functions might case this.  Try -D__NO_STRING_INLINES.

I tested this with Lynx. No joy. Problem is still there, and is only fixed
by removing the -O2 optimising option from the Makefiles in lynx-2.8.

I think this is more a symptom of low resources actually. I _may_ decide
to enlarge the swap file, and see if it will be happier with that. 

Cheers,
Alex
--
 /\_/\  Legalise cannabis now! 
( o.o ) Smoke some cannabis today!
 > ^ <  Peace, Love, Unity and Respect to all.

http://www.tahallah.demon.co.uk

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Alex Buell]

On Sat, 27 Jun 1998, Franz Sirl wrote:

> I had the same problem with lynx, egcs and glibc-2.1 snapshots on LinuxPPC, but
> the latest snapshots fix this. I compiled the RedHat-5.1 SRPM out of the box
> recently with ~egcs-2.91.34.

Hmm. Okay, it does look as if the soon forthcoming egcs-1.1 will cure this
particular problem.

Cheers,
Alex
--
 /\_/\  Legalise cannabis now! 
( o.o ) Smoke some cannabis today!
 > ^ <  Peace, Love, Unity and Respect to all.

http://www.tahallah.demon.co.uk

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Franz Sirl]

Am Sat, 27 Jun 1998 schrieb Alex Buell:
>On Sat, 27 Jun 1998, Franz Sirl wrote:
>
>> I had the same problem with lynx, egcs and glibc-2.1 snapshots on LinuxPPC, but
>> the latest snapshots fix this. I compiled the RedHat-5.1 SRPM out of the box
>> recently with ~egcs-2.91.34.
>
>Hmm. Okay, it does look as if the soon forthcoming egcs-1.1 will cure this
>particular problem.

Uhm, sorry, I just saw I put a -D__NO_STRING_INLINES into the CFLAGS to get rid
of a lot of warnings, I will (that will take some time) recheck without it.

Franz.

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Andreas Jaeger]

>>>>> Alex Buell writes:

 > Hi guys,
 > I could hardly believe my eyes tonight when I finally compiled lynx-2.8
 > with egcs-1.0.3a, using glibc-2.0.94, on kernel v2.1.107. 

 > What happened was that for some reasons (yet undetermined), the cc1
 > trundled off merrily on its way compiling LYMainLoop.c (in the src
 > subdirectory), and it seemed to get stuck somewhere. I then went into top
 > to see what it was doing and I was astonished to see that the swap space
 > was FULL, no more swap space available (this on a 32MB i486 machine with a
 > 26MB swap file - it never used up more than 6MB normally and only uses up
 > more than that rarely when looking at some stupid java-based web sites)
 > and Linux was thrashing the hard disk so hard I was very afraid for the
 > poor thing's safety.  According to top, it had well over 49MB of virtual
 > memory used up on compiling that source file. Of course when the real
 > memory dropped to less than 512k, the compiler finally aborted with a
 > sig-11 error (please don't quote me the reference to that sig-11 error web
 > site - I have never had a sig-11 in the last 18 months that I have been
 > using Linux - and this is the first sig-11 error I have seen).

glibc 2.0.94 uses optimized inline functions.  Inlining these
functions might case this.  Try -D__NO_STRING_INLINES.

(see also the glibc FAQ entry: 3.8. I've got errors compiling code
that uses certain string functions.  Why?)

Andreas
-- 
 Andreas Jaeger   aj@arthur.rhein-neckar.de    jaeger@informatik.uni-kl.de
  for pgp-key finger ajaeger@alma.student.uni-kl.de

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Franz Sirl]

Am Fri, 26 Jun 1998 schrieb Alex Buell:
>I've had a similar report from another guy on the egcs mailing list. He
>says that this is definitely a problem with the egcs compiler - to fix
>this, either remove the -O2 or change to -O1. Either way, works fine with
>that. But it's going to be interesting to find out which code sequence in
>Lynx sent the compiler into an endless optimising loop. 

I had the same problem with lynx, egcs and glibc-2.1 snapshots on LinuxPPC, but
the latest snapshots fix this. I compiled the RedHat-5.1 SRPM out of the box
recently with ~egcs-2.91.34.

Franz.

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Jean-Pierre Radley]

Alex Buell bruited (on 26Jun):
| On Fri, 26 Jun 1998, Andrea Arcangeli wrote:
| 
| > Could you compile the kernel with gcc-2.7.2 to be sure?
| 
| This problem isn't the kernel this time.
|  
| > Are you compiling lynx with the -g option of gcc? If so remove it. Also
| > check the cache size when the swap become very big, latest 2.1.x eat a lot
| > of memory with the cache and they work bad on machine with low memory.
| > It' s a complexity problem, to go faster you need to waste more memory... 
| 
| I've had a similar report from another guy on the egcs mailing list. He
| says that this is definitely a problem with the egcs compiler - to fix
| this, either remove the -O2 or change to -O1. Either way, works fine with
| that. But it's going to be interesting to find out which code sequence in
| Lynx sent the compiler into an endless optimising loop. 

Hmm. Here, I compile with -O3, and have no problem at all.

-- 
Jean-Pierre Radley <jpr@jpr.com>  XC/XT Custodian   Sysop, CompuServe SCOForum

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[D.A. Harris]

On Fri, Jun 26, 1998 at 03:16:28AM -0400, Alex Buell wrote:
.. [snip] ..
> 
> Has anyone successfully compiled lynx-2.8 with egcs-1.0.3a? Or has anyone
> observed this behaviour? If not, can you please try compiling lynx-2.8 on
> your respective machines and see what the results are? 
> 
.. [snip] ..
> 
> Cheers,
> Alex


I didn't have any problems compiling it up lynx-2.8 with egcs-1.0.3a, 
2.1.106, and gnulib 2.0.7pre3.  Course I wouldn't think that kernel 
should matter much in the behavior of the compiler, unless there was some
sort of kernel memory leak.  The 2.1.107 has some other funky problems, 
so I just switched back to 2.1.106.

-- 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dale Harris  <rodmur@csuchico.edu>   PGP KeyID: E26EC5FD      
http://www.ecst.csuchico.edu/~rodmur/
|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|+|

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Alex Buell]

On Fri, 26 Jun 1998, D.A. Harris wrote:

> I didn't have any problems compiling it up lynx-2.8 with egcs-1.0.3a, 
> 2.1.106, and gnulib 2.0.7pre3.  Course I wouldn't think that kernel 
> should matter much in the behavior of the compiler, unless there was some
> sort of kernel memory leak.  The 2.1.107 has some other funky problems, 
> so I just switched back to 2.1.106.

How much memory do you have, and how large is your swap file?

Cheers,
Alex
--
 /\_/\  Legalise cannabis now! 
( o.o ) Smoke some cannabis today!
 > ^ <  Peace, Love, Unity and Respect to all.

http://www.tahallah.demon.co.uk

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Alex Buell]

On Fri, 26 Jun 1998, Myrdraal wrote:

> > I could hardly believe my eyes tonight when I finally compiled lynx-2.8
> > with egcs-1.0.3a, using glibc-2.0.94, on kernel v2.1.107. 
[snip]
> I also had that problem (With like, umm, the previous egcs snapshot.) -
> I would guess that the problem is an egcs bug. My fix was to compile without
> optimization or with plain -O, I forget which.
>  -Myrdraal

Looks like a bug in the egcs optimiser. Lynx had an optimisation level of
2 (-O2). Hmm. 

Cheers,
Alex
--
 /\_/\  Legalise cannabis now! 
( o.o ) Smoke some cannabis today!
 > ^ <  Peace, Love, Unity and Respect to all.

http://www.tahallah.demon.co.uk

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Alex Buell]

On Fri, 26 Jun 1998, Andrea Arcangeli wrote:

> Could you compile the kernel with gcc-2.7.2 to be sure?

This problem isn't the kernel this time.
 
> Are you compiling lynx with the -g option of gcc? If so remove it. Also
> check the cache size when the swap become very big, latest 2.1.x eat a lot
> of memory with the cache and they work bad on machine with low memory.
> It' s a complexity problem, to go faster you need to waste more memory... 

I've had a similar report from another guy on the egcs mailing list. He
says that this is definitely a problem with the egcs compiler - to fix
this, either remove the -O2 or change to -O1. Either way, works fine with
that. But it's going to be interesting to find out which code sequence in
Lynx sent the compiler into an endless optimising loop. 

Cheers,
Alex
--
 /\_/\  Legalise cannabis now! 
( o.o ) Smoke some cannabis today!
 > ^ <  Peace, Love, Unity and Respect to all.

http://www.tahallah.demon.co.uk

Re: egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Jean-Pierre Radley]

Alex Buell bruited (on 26Jun):
| 
| Has anyone successfully compiled lynx-2.8 with egcs-1.0.3a? Or has anyone
| observed this behaviour? If not, can you please try compiling lynx-2.8 on
| your respective machines and see what the results are? 

I compiled the dev-release 17 of Lynx 2.8.1 (which just appeared
yesterday) with the snapshot of egcs updated by cvs about fifteen hours
ago. It worked just as painlessly as every other Lynx compilation always
has.

This is in SCO OSR 5..0.4.

-- 
Jean-Pierre Radley <jpr@jpr.com>  XC/XT Custodian   Sysop, CompuServe SCOForum

egcs-1.0.3a weirdness with lynx-2.8 on v2.1.107

[Alex Buell]

Hi guys,

I could hardly believe my eyes tonight when I finally compiled lynx-2.8
with egcs-1.0.3a, using glibc-2.0.94, on kernel v2.1.107. 

What happened was that for some reasons (yet undetermined), the cc1
trundled off merrily on its way compiling LYMainLoop.c (in the src
subdirectory), and it seemed to get stuck somewhere. I then went into top
to see what it was doing and I was astonished to see that the swap space
was FULL, no more swap space available (this on a 32MB i486 machine with a
26MB swap file - it never used up more than 6MB normally and only uses up
more than that rarely when looking at some stupid java-based web sites)
and Linux was thrashing the hard disk so hard I was very afraid for the
poor thing's safety.  According to top, it had well over 49MB of virtual
memory used up on compiling that source file. Of course when the real
memory dropped to less than 512k, the compiler finally aborted with a
sig-11 error (please don't quote me the reference to that sig-11 error web
site - I have never had a sig-11 in the last 18 months that I have been
using Linux - and this is the first sig-11 error I have seen).

Has anyone successfully compiled lynx-2.8 with egcs-1.0.3a? Or has anyone
observed this behaviour? If not, can you please try compiling lynx-2.8 on
your respective machines and see what the results are? 

What the heck was it doing to that source code anyway?! :o) No harm done
anyway - I'm surprised Linux withstood the demands made of it - Windows 95
would have thrown in the towel ages ago... 

Cheers,
Alex
--
 /\_/\  Legalise cannabis now! 
( o.o ) Smoke some cannabis today!
 > ^ <  Peace, Love, Unity and Respect to all.

http://www.tahallah.demon.co.uk