Re: [Chicken-users] Bug in MD5 egg?

[felix]

Alejandro Forero Cuervo wrote:

> Hi.
>
> Sometimes, when  calculating a  MD5 sum in  a big  directory with
> lots of files, I get a  stack overflow error.  I modified my code
> to contain  only tail  recursive calls  but it  didn't help  so I
> suspect the MD5 egg might have a bug.  I took a quick peek at the
> code of the MD5 egg and couldn't find the problem.
>
>
> The following code *sometimes* triggers the error:
>
> (require 'md5)
>
> (let loop ((f (remove directory? (directory "."))))
>   (if (not (null? f))
>     (let ((in (open-input-file (car f))))
>       (format #t "~A: ~A~%" (car f) (md5:digest in))
>       (close-input-port in)
>       (loop (cdr f)))))
>
> Can anyone please try this in  a big directory with lots of files
> (in my case  it is a directory with 3600  files totaling 47m) and
> confirm  the error?   Please  do  run it  a  few  (10?) times  in
> different csi processes, as,  strangely, the error doesn't always
> occur.

Oh, if you have a fairly recent Chicken, try to invoke the
compiler with

-disable-stack-overflow-checks

This disables the checks, alternatively run the executable
(or csi, whatever you use here) with `-:o'.

From what I see in the generated compiler output of md5.scm,
the only place where a stack-overflow check occurs is in
md5:update, which couldn't really be triggered, unless
the code calling md5:digest already has built up a large stack.

What is the default nursery size on your system? (run csi
and enter ",r", or look at the value of
C_DEFAULT_TARGET_STACK_SIZE in the generated c_defaults.h
in the Chicken build directory).


cheers,
felix