Here's a comparison of a purely hash-based ivar map versus one that just
does a linear search.
For objects with only a few instance variables, the linear search
mechanism is going to be faster. Perhaps what we need is a map that's
optimized so that small sizes are linear searches and after a cert
Two changes in attached patch:
- attr accessors should store interned name of variable
- attr accessor methods can be "fast"
Substantial improvement (this is with the standard ivars hashmap):
BEFORE:
100k loop of 100 attr accesses and assign to local; one ivar
2.695000 0.00 2.695000
Werner Schuster (murphee) wrote:
A solution would be to simply overload the parse method in the
org.jruby.Ruby class to call the three argument one with null as last
arg (since this does seem to work).
This is simply not true.
I would first of all argue that everything you get when you req
Thomas E Enebo wrote:
> I can only say that that was a mistake. We had no intention of
> breaking API between minor release versions. What's done is done
> though, we will try and be more careful in the future.
>
OK - good to hear that it was an oversight not a conscious decision.
BTW: it mig
I can only say that that was a mistake. We had no intention of
breaking API between minor release versions. What's done is done
though, we will try and be more careful in the future.
-Tom
On 8/25/07, Werner Schuster (murphee) <[EMAIL PROTECTED]> wrote:
> howdy,
>
> I have an issue regarding th
howdy,
I have an issue regarding the stability of JRuby in minor releases.
Case in point: I just ran the tests of JParseTree with the new JRuby 1.0.1.
Result: 100 % tests failed.
Theses tests work perfectly with JRuby 1.0.
The reason:
The JRuby.parse went from two required args to three require
