[Mondrian] Re: Mondrian map factory classes

Eric McDermid mcdermid at stonecreek.com
Fri Jul 17 02:40:13 EDT 2009


On Jul 16, 2009, at 5:21 PM, Julian Hyde wrote:
>
> A simple rule to use a HashMap for measures, Flat3Map for everything  
> else
> should do it. Log a jira case if you don't want to take this on now.

Simple enough, and already implemented (check-in coming as soon as my  
test run finishes).  This will definitely improve the situation,  
though it still won't be ideal.

In particular, there are some times where a Flat3Map is more optimal  
for measures, but that only affects 10% of the cases involving  
measures.  Since measures themselves are only 10% of the overall set  
of cases, this means we're going from a situation where 90% of the  
members use a Hashmap when a Flat3Map would be more efficient to one  
in which only 1% of the cases do so.  Not perfect, but certainly not  
bad at all.

We'll need to revisit this rule someday if we add more undeclared  
properties to non-measure members, of course.  I'll still log a Jira  
case to remind us that we can make the behavior more robust in the  
long term if we can get an accurate property count, but I don't see it  
as a high priority.

  -- Eric




More information about the Mondrian mailing list