Remove get_id() hotspot

[bonzini]

Use lazy_property for get_id(). To avoid touching code all over the place, keep get_id() but make it access the actual lazy property (which will be simply an entry in self.dict)

[dcbaker]

I’d honestly love to see many of the get_* functions go. They either need a better name, or end up being a useless layer of indirection to return a property, which comes with a performance cost.

incidentally I had been looking at this and the. Got bogged down adding type annotations to the vs backend…

[bonzini]

They either need a better name, or end up being a useless layer of indirection to return a property, which comes with a performance cost.

Absolutely. I'd also love to have a lazy_property variant that stores the outcome of a method in the __dict__ of a parameter, thus being a cheaper and cleaner variant of lru_cache(max_size=None) (for example if you have N>1 parameters but N-1 depend on the remaining 1). But it seemed to me that this was not the right place to do a full refactoring of get_*.

If you'd like me to introduce an id property then that's good, though it's probably a good idea anyway to fix lazy_property vs. __.

[dcbaker]

I think i'd rather have id exposed as a public property, even if we don't start using it yet.

I'm fine with the __set_name__ change, this code predates __set_name__ by a lot.