Valgrind warnings with -1.8.6

Valgrind warnings with -1.8.6

[Andrew Gaylard]

Hi,

I'm linking with libguile on Linux, built from source, to extend my
C application with Scheme. I'm seeing a lot of warnings from
valgrind.  The same warnings appear when using Guile standalone.
Should I be worried?

The following steps show the problem:

Use this as hello-world.scm:

#!/usr/local//bin/guile \
-e main -s
!#

(define (main args)
  (display "hello world")
  (newline))

...And run it like this:

$ valgrind --tool=memcheck --gen-suppressions=all --num-callers=30
--track-fds=yes --trace-children=yes --log-file=hello-world.vg
--show-below-main=yes --leak-check=full --show-reachable=yes
--leak-resolution=high ./hello-world.scm

...Now hello-world.vg shows 47 warnings relating to uninitialised values:

$ grep '^==[0-9]*== [a-zA-Z]' ./hello-world.vg | awk '/FILE
DESCRIPTORS/ { x=0 } x==1 {print} /My PID/ { x=1 }' | sort | uniq -c
     33 ==12696== Conditional jump or move depends on uninitialised value(s)
     14 ==12696== Use of uninitialised value of size 4
      1 ==12696== Warning: bad signal number 0 in sigaction()

Thanks,
Andrew

Valgrind warnings with -1.8.6

[Andrew Gaylard]

Hi,

I'm linking with libguile on Linux, built from source, to extend my
C application with Scheme. I'm seeing a lot of warnings from
valgrind.  The same warnings appear when using Guile standalone.
Should I be worried?

The following steps show the problem.

Use this as hello-world.scm:

#!/usr/local//bin/guile \
-e main -s
!#

(define (main args)
 (display "hello world")
 (newline))

...And run it like this:

$ valgrind --tool=memcheck --gen-suppressions=all --num-callers=30
--track-fds=yes --trace-children=yes --log-file=hello-world.vg
--show-below-main=yes --leak-check=full --show-reachable=yes
--leak-resolution=high ./hello-world.scm

...Now hello-world.vg shows 47 warnings relating to uninitialised values:

$ grep '^==[0-9]*== [a-zA-Z]' ./hello-world.vg | awk '/FILE
DESCRIPTORS/ { x=0 } x==1 {print} /My PID/ { x=1 }' | sort | uniq -c
    33 ==12696== Conditional jump or move depends on uninitialised value(s)
    14 ==12696== Use of uninitialised value of size 4
     1 ==12696== Warning: bad signal number 0 in sigaction()

Thanks,
Andrew

Re: Valgrind warnings with -1.8.6

[Andreas Rottmann]

Andrew Gaylard <ag@computer.org> writes:

> Hi,
>
> I'm linking with libguile on Linux, built from source, to extend my
> C application with Scheme. I'm seeing a lot of warnings from
> valgrind.  The same warnings appear when using Guile standalone.
> Should I be worried?
>
No, this is normal AFAIK. The garbage collector in Guile messes up
valgrind. At some point, I had a working suppression file for that, but
it's likely to be bitrotted into unusability.

Regards, Rotty

Re: Valgrind warnings with -1.8.6

[Neil Jerram]

Andreas Rottmann <a.rottmann@gmx.at> writes:

> Andrew Gaylard <ag@computer.org> writes:
>
>> Hi,
>>
>> I'm linking with libguile on Linux, built from source, to extend my
>> C application with Scheme. I'm seeing a lot of warnings from
>> valgrind.  The same warnings appear when using Guile standalone.
>> Should I be worried?
>>
> No, this is normal AFAIK. The garbage collector in Guile messes up
> valgrind. At some point, I had a working suppression file for that, but
> it's likely to be bitrotted into unusability.

That is fundamentally true; however, we have recently addressed some
genuine Valgrind warnings in the 1.8.x development branch.  Those fixes
will show up if/when we release a 1.8.7, but in the meantime you may
like to look at the 1.8.x branch in Git, to see if any of the warnings
fixed match the ones that you are seeing.

      Neil

Re: Valgrind warnings with -1.8.6

[Ludovic Courtès]

Hello,

Neil Jerram <neil@ossau.uklinux.net> writes:

> That is fundamentally true; however, we have recently addressed some
> genuine Valgrind warnings in the 1.8.x development branch.

These were Helgrind warnings, not Memcheck warnings, right?

There's a suppression file there:

  http://www.mail-archive.com/guile-devel@gnu.org/msg00743.html

If it happens to work well, maybe we could ship it with the next 1.8
release.

Thanks,
Ludo'.

Re: Valgrind warnings with -1.8.6

[Paul Emsley]

Ludovic Courtès wrote:
> Hello,
>
> Neil Jerram <neil@ossau.uklinux.net> writes:
>
>   
>> That is fundamentally true; however, we have recently addressed some
>> genuine Valgrind warnings in the 1.8.x development branch.
>>     
>
> These were Helgrind warnings, not Memcheck warnings, right?
>
> There's a suppression file there:
>
>   http://www.mail-archive.com/guile-devel@gnu.org/msg00743.html
>
> If it happens to work well, maybe we could ship it with the next 1.8
> release.
>   

More suppressions:

http://coot.googlecode.com/svn/trunk/guile.supp

(I don't know if they are useful - and may be specific cases of above).

Paul.

Re: Valgrind warnings with -1.8.6

[Neil Jerram]

ludo@gnu.org (Ludovic Courtès) writes:

> Hello,
>
> Neil Jerram <neil@ossau.uklinux.net> writes:
>
>> That is fundamentally true; however, we have recently addressed some
>> genuine Valgrind warnings in the 1.8.x development branch.
>
> These were Helgrind warnings, not Memcheck warnings, right?

Yes, good point!

     Neil

Re: Valgrind warnings with -1.8.6

[Andrew Gaylard]

On Mon, Apr 6, 2009 at 5:05 PM, Andrew Gaylard <ag@computer.org> wrote:
> Hi,
>
> I'm linking with libguile on Linux, built from source, to extend my
> C application with Scheme. I'm seeing a lot of warnings from
> valgrind.  The same warnings appear when using Guile standalone.
> Should I be worried?

After some digging, I'm a bit closer.  Building from today's git
shows fewer problems:

grep '^==[0-9]*== [a-zA-Z]' ./hello-world.vg |
awk '/FILE DESCRIPTORS/ {x=0} x==1 {print} /My PID/ {x=1}' |
 sort | uniq -c
     25 ==3551== Conditional jump or move depends on uninitialised value(s)
     15 ==3551== Use of uninitialised value of size 4
      1 ==3551== Warning: bad signal number 0 in sigaction()

Also, something that may or may not be related...
I've seen that a build without optimisation, i.e.:

make distclean ; ./autogen.sh ; CFLAGS=-g ./configure ;  make

fails like this:

make[2]: Entering directory `/home/apg/tmp/guile-git/guile/module'
/bin/mkdir -p `dirname ice-9/psyntax-pp.go`
../meta/uninstalled-env guile-tools compile -o "ice-9/psyntax-pp.go"
"ice-9/psyntax-pp.scm"
ERROR: Stack overflow
make[2]: *** [ice-9/psyntax-pp.go] Error 1

But doing the same without specifying the CFLAGS
(i.e. using the default CFLAGS  of "-g -O2") builds fine:

make distclean ; ./autogen.sh ; ./configure ;  make

(This is on Linux (Ubuntu-8.04).)

- Andrew

Re: Valgrind warnings with -1.8.6

[Ludovic Courtès]

Hello,

Andrew Gaylard <ag@computer.org> writes:

> After some digging, I'm a bit closer.  Building from today's git
> shows fewer problems:

You're referring to `master', right?

> grep '^==[0-9]*== [a-zA-Z]' ./hello-world.vg |
> awk '/FILE DESCRIPTORS/ {x=0} x==1 {print} /My PID/ {x=1}' |
>  sort | uniq -c
>      25 ==3551== Conditional jump or move depends on uninitialised value(s)
>      15 ==3551== Use of uninitialised value of size 4
>       1 ==3551== Warning: bad signal number 0 in sigaction()

Cool.

> I've seen that a build without optimisation, i.e.:

That is expected: the evaluator and/or bytecode interpreter eat up too
much stack space with `-O0'.

Thanks,
Ludo'.

Re: Valgrind warnings with -1.8.6

[Andy Wingo]

On Thu 30 Apr 2009 23:09, ludo@gnu.org (Ludovic Courtès) writes:

> Andrew Gaylard <ag@computer.org> writes:
>
>> I've seen that a build without optimisation, i.e.:
>
> That is expected: the evaluator and/or bytecode interpreter eat up too
> much stack space with `-O0'.

Actually I didn't expect that, after the recent getrlimit changes.
Perhaps we revert to building psyntax-pp.go after the compiler is
expanded.

Cheers,

Andy
-- 
http://wingolog.org/