Re: Excessive memory usage of large hash_maps

[pcarlini at unitus dot it]

Quoting Paul Dubuc <pdubuc@cas.org>: 
 
> If patching it is the best approach, would adding more prime numbers 
> to the table be a good solution?  
> We have a cases when large hash tables temporarily increase size over  a 
> given threshold, their size doubles and the process uses excessive memory. 
> It also appears that when one hash table is copied to another the large 
> capacity is duplicated even though the number of elements in the source 
> container may have decreased significantly below the high water mark. 
 
Hi again, Paul. 
The issues you are bringing to the fore are real and in fact, if you look 
under Sec. E of the document I mentioned before, you will notice that a 
specific novelty of Matt's proposal is the "load factor control" which 
replaces the previous "resize" way the user had to control the size of 
the hashtable. 
 
Note, in particular (to partially answer your second point), that, in 
Matt's proposal, both the copy contructor and the copy assignment operator 
don't copy the load factor but only the maximum load factor thus making 
possible for a conforming implementation to obviate to the problem. 
 
All of this will be definitely dealt with in the next few weeks as part of 
the effort that I mentioned before. I hope you will continue to provide your 
important feedback then. 
 
Paolo. 
 
P.S. In the process will be obviously also fixed the inhability of the current 
hashed containiers to work with long long types.