Jim Mussared
Jim Mussared
> Have you done any profiling to see which slots are accessed the most? That's a good question... All the analysis I did (see #7542) was entirely on code size...
@dlech I tested this and the result was +232 bytes (+288 bytes as predicted for adding the slot and changing the definitions, then -56 bytes to use `type->locals_dict` directly) and...
> From a practical point of view: could the relevant commit, or otherwise some explanation here, include the basic instructions needed to convert from the current implementation to the new...
Updated with two additional changes: Merged the `getiter` and `iternext` slots. On PYBV11: - 22 types use mp_identity_getiter, mp_stream_unbuffered_iter iternext - 19 types use mp_identity_getiter, custom iternext - 22 type...
@robert-hh I promised @stinos too that I would write some instructions. Haven't had a chance to come back to this since the PR was finalised earlier this arvo. (At the...
> Only the property is lost, that you could stop at any point in the commit tree and still have a state, that can be compiled and run. @robert-hh Can...
> And before I forget it (again): One really good aspect of this PR is, that it reduces the code size. At the SAMD port it's about 2.5 k. Thanks!...
> You mention the getiter slot, but what about the iternext slot in this case. There is now only an `iter` slot and its behaviour depends on the four possible...
So in that specific case, now just the flag is set and the iter slot should be unset. (There was about 20 instances of this in PYBV11, so 80 bytes...