Re: [Chicken-users] profile: collecting stats for multiple runs

[felix winkelmann]

On Mon, 28 Feb 2005 18:31:08 -0500, Alejandro Forero Cuervo
<address@hidden> wrote:
> 
> Would you be willing to accept a patch to modify this?  I think it is
> far more useful to have the profiler add new information to the
> existing file (instead of overwriting it).
> 
> I decided to take some time and create a patch implementing this.  See
> the file attached.  Basically, I do the following:
> 
> - Open the PROFILE file in append mode (so I won't overwrite previous
>   info).
> 
> - Modify chicken-profile.scm to be smart enough to merge information
>   from multiple runs, using a hash table.
> 
> If a user wants the previous semantics, all he needs to do is rm
> PROFILE before running his program (another option would be to add a
> --profile-overwrite paramter to Chicken?  I would vote against is, at
> it adds complexity to Chicken's interface without being too useful).
> What do you think?

Very good. I'll apply your patches ASAP. But I'm not sure whether it is 
perhaps a bit surprising, if a user has an old PROFILE lying around, and
several runs produce different output (for example when continuously
recompiling a benchmark).
What about an alternative to the -profile option, something like
"-accumulate-profile"?

> 
> BTW, I haven't actually got around to testing my changes.  They could
> be wrong (I'm on a very slow machine, where building Chicken can take
> hours; I might try them latter).

They look good. I'll do a bit of testing.

> 
> Thank you!  You are an excellent free software maintainer!

Thanks a lot. [blushes] :-) 

> 
> Alejo.
> http://bachue.com/alejo
> 
> Ps: The structure of sort-by-calls, sort-by-time and sort-by-avg is
> very similar.  Would you want me to capture it in a procedure and
> mayhaps simplify things a bit?  Heh, I'm a bit of a perfectionist,
> sometimes...

Sure, I'm quite aware that my code is sometimes a bit unstructured.
Patches are always welcome.


cheers,
felix