Thanks for fixing up the stats. It's interesting that the specialisation hits went up after merging the instructions.

FYI, I'm working on specializations for:

• property
• method
• non overriding descriptor
• mutable class

The increased rate of specialization hits might be due to exponential backoff. We get fewer specialization failures.

It might be due to better specialization forms too. Looking at the stats, the old instance attribute methods and load class attribute failures are now gone thanks to the instruction merge :).

I'm sure some is due to improved specialization, but be aware that exponential backoff means that the success ratio will appear to be higher than it really is.

It is the hit ratio that really matters.