[antlr-interest] classloading time for old DFA

[Terence Parr]

On Jun 5, 2006, at 2:33 AM, Martin Probst wrote:

> Hi,
>
>> Looks like loading that 1.9M of class files takes about 500ms of  
>> that 600ms discrepancy between old and new.  So, once loaded, the  
>> old DFA aren't that much slower.
>
> I think you should still go with the new system - that's a lot less  
> code to optimize for the JIT, a lot less class files, a lot less GC  
> on those files.

Definitely.  No going back now ;)

Further, I'm thinking of doing this for even the acyclic DFA so we  
have one mechanism in code gen and in resulting parsers...so even LL 
(2) decisions will result in a little DFA (but not an object).  LL(1)  
decisions should be a switch, but I'm not sure exactly how to make  
that happen easily.

Ter