Re: Linux 2.0.32 gcc2.8.1 swarm1.2 (re: negative double weirdness)

[Marcus G. Daniels]

>>>>> "SE" == Steve Emsley <address@hidden> writes:

SE> I built Swarm without optimisation and the error went away
SE> ... Why? I suppose I shouldn't ask. 

The explanation probably has something to do with different register
usage for shared libraries.  It would be useful to post (complete or
trimmed) source code that would demonstrate the problem, so that
someone can find the particular optimization flag that triggers the
bug and post a useful bug report/fix to the GCC bug mailing list.

SE> By turning off optimization what am I losing? 

Speed.

SE> I'm fortunate that I was debugging my model at the time of the
SE> upgrade so the error was easily isolated. If my full model was
SE> running such an error could have possibly become published!. Well,
SE> as easily as anything gets published :) How does this sort of
SE> thing impact on the aim of creating an environment conducive to
SE> reproducible experiments?

This is an example of why having a common [simulation] package like
Swarm important: it's easier for many people to shake out problems in
a few targets than it is for a few people to shake out problems in
many targets.  (Even though in this case, the evidence suggests a bug
in GCC, not in Swarm.)

SE> Perhaps in addition to using heatbugs/market etc. we
SE> should try to create a test suite that picks up this sort of
SE> error. O.K. this is something to devolve to the user group. I'll
SE> try to find time to reproduce my error in a handy test-script.

Absolutely.  I agree that it is very important goal.


                  ==================================
   Swarm-Support is for discussion of the technical details of the day
   to day usage of Swarm.  For list administration needs (esp.
   [un]subscribing), please send a message to <address@hidden>
   with "help" in the body of the message.
                  ==================================