I'm having some issues using MethodCache instead of CacheMap.
The problem I have right now with MethodCache is that there's no way to
use it for inline caching. In order for my inline cache to work, I need
to be able to cache at the call site. But in order for that to be safe I
need to be informed when the cache should be cleared. Checking if I
should clear it on each call defeats the purpose.
The old CacheMap allowed code to register for such events. When a new
method was defined in a hierarchy, anyone who had cached that method or
the same-named method from a superclass would be notified they should
flush their caches. With a small modification, this also works with call
sites; if I cache a method I could rest assured that changes would
correctly cause me to flush.
There's currently no way to do this with the MethodCache. However
MethodCache *is* a bit faster.
Here are some numbers for method invocation with MethodCache versus
CacheMap:
CacheMap:
Test interpreted: 100k loops calling self's foo 100 times
2.587000 0.000000 2.587000 ( 2.587000)
2.894000 0.000000 2.894000 ( 2.894000)
1.908000 0.000000 1.908000 ( 1.908000)
1.900000 0.000000 1.900000 ( 1.899000)
1.892000 0.000000 1.892000 ( 1.892000)
1.889000 0.000000 1.889000 ( 1.889000)
1.892000 0.000000 1.892000 ( 1.892000)
1.886000 0.000000 1.886000 ( 1.887000)
1.874000 0.000000 1.874000 ( 1.874000)
1.906000 0.000000 1.906000 ( 1.906000)
MethodCache:
Test interpreted: 100k loops calling self's foo 100 times
2.437000 0.000000 2.437000 ( 2.437000)
2.592000 0.000000 2.592000 ( 2.593000)
1.697000 0.000000 1.697000 ( 1.697000)
1.697000 0.000000 1.697000 ( 1.697000)
1.700000 0.000000 1.700000 ( 1.700000)
1.707000 0.000000 1.707000 ( 1.707000)
1.702000 0.000000 1.702000 ( 1.702000)
1.695000 0.000000 1.695000 ( 1.695000)
1.702000 0.000000 1.702000 ( 1.702000)
1.714000 0.000000 1.714000 ( 1.714000)
So I'm trying to reconcile the situation. I certainly don't want to lose
MethodCache performance, but I need to have the callback for inline
caching to be efficient. The cache flush has to be triggered outside the
call path, rather than having an additional guard in the code.
Any ideas?
- Charlie
---------------------------------------------------------------------
To unsubscribe from this list please visit:
http://xircles.codehaus.org/manage_email
