Re: still getting errors with CRM

On Wed, Apr 23, 2008 at 2:07 PM, Paolo <[email protected]> wrote:
>  (not so) wild guess: it's a 64bit issue. I'd rather not waste too much time on
>  that, till the 'long* -> int*_t' conversion has been done.
>  Till then, you may want to try Gerrit's 'branch' which is supposed to track
>  closely mainline on 'stable' classifiers and have a number of 64bit (possible)
>  issues fixed.
>  Try that one, see if your loop (maybe go up to 10000) bombs out again.

Thanks for the confidence in that branch, but I had a short chat with
Jason just now; be warned that the CSS files produced by Bill's on a
64-bit are in /some/ places not precisely '32-bit' where they should,
because the code writes 'long's to file. It's highly dependent on the
classifier used, etc.etc.etc., but the point is that GerH crm114 may
NOT like your existing CSS files on a 64-bit box if you created them
with 'vanilla' crm114, because the int-size is a little bit, erh, off.

So in this case, to ensure we're not adding a bunch of variables into
this unholy mix, I suggest we take the source code Jason used to
create his crm114 binary, make a copy of that too, then edit it in a
(very) few places to see if we can get more info out of there when
this error occurs.
As it's a time-related thing - at least seems so to me - it also means
any change made to the system may 'skew' the critical timing and 'make
the error disappear'. I've tackled these kind of issues before and
found that when you hit something like this and a re close to anything
that resembles 'reproducability' on them, you better tread very
carefully. Even recompiling code with other optimization settings,
e.g. to allow debugging, can throw the whole thing off and make it go
away, forcing you to wait for another 'happening'.

Sooooo.... bottom line today: my suggestion:

1) tar.gz the whole shebang (CSS files, crm binary, scripts, the email
files) to a backup somewhere because the time will be there when we
need to restore the 'test environment' to the point where it all
started: today.

2) Fetch the source code for the crm114 binary running on Jason's
machine right now.

3) recompile it in the 'test environment', copy the new bin into the
proper place so we can have another for {0...1000} run and see if the
timing error still pops up - better get good prayer beads, because I
can only hope it's still there after this initial action. If not,
straight /back/ to the drawing board.

4) ... may try to augment the error message, recompile and see if
another run keeps producing the error. Shell account(s) may be handy
so Paolo and/or me can dig around in there. Might need a little
coordination too, so we know from each other what we're doing and
thinking there. Anyway, let's see if number 3 there is a 'relative'
success (= error stays around) and take it from there.


Jason, again, very good work at isolating this; others may not have
this issue, but at least this is something we can all learn from and
it can only improve crm114.

And, yes, I'd love to try my version on there too, but I shelve that
ego-trip for once we've pinned down this sucker /properly/. GerH
doesn't come with a 'vanilla CSS 64-bit CSS file loader/converter' for
vanilla crm114 CSS files yet so it's only adding to the chaos right
now if we did use that one.

Ger



>
>  --
>  paolo
>
>
>
>  -------------------------------------------------------------------------
>  This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
>  Don't miss this year's exciting event. There's still time to save $100.
>  Use priority code J8TL2D2.
>  http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
>  _______________________________________________
>  Crm114-general mailing list
>  [email protected]
>  https://lists.sourceforge.net/lists/listinfo/crm114-general
>
>



-- 
Met vriendelijke groeten / Best regards,

Ger Hobbelt

--------------------------------------------------
web: http://www.hobbelt.com/
 http://www.hebbut.net/
mail: [email protected]
mobile: +31-6-11 120 978
--------------------------------------------------

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone