[fortran-dev, patch] Parsed format data caching

Jerry DeLisle jvdelisle@verizon.net
Tue Mar 31 03:17:00 GMT 2009


Bud Davis wrote:
>  
> --- On Mon, 3/30/09, Paul Richard Thomas <paul.richard.thomas@gmail.com> wrote:
>   
>> Well, if not Treaps, what about a simple binary tree?  I
>> guess that
>> for the applications that we are talking about, balancing
>> is not too
>> much of an issue?
>>
>> Cheers
>>
>> Paul
> 
> 
> although not very sexy, the plain old boring ordered list (or even un-ordered list) would work fine for these tiny values of N.
> 
> ordered list, average number of compares would be 8 .  in the noise...
> 
> kudos to jerry for the caching idea..it is a good one.
> 

I want to clarify a few things here.  The idea to cache the format data came 
from a discussion with Janne on IRC.  So Janne should be recognized for planting 
the seed.

Secondly, I used no existing hashing libraries.  I did not want this 
implementation to be encumbered in any way and I think it should stay that way. 
   The whole discussion on the licensing issue is moot regarding format caching.

My preference is we roll our own on the memory checking as well.  It really is 
not that difficult to do.  I believe strongly in the KISS principle here.  I 
simply don't believe that what we are trying to do calls for full blown and 
complex hashing, treaps, or whatever.  For memory checking, is this not a 
runtime option?  Does it have to be totally optimal?  Or am I missing something 
here?

Regards,

Jerry

Jerry




More information about the Fortran mailing list