This issue tracker has been migrated to GitHub, and is currently read-only.
For more information, see the GitHub FAQs in the Python's Developer Guide.

Author vstinner
Recipients Mark.Shannon, neonene, paul.moore, rhettinger, steve.dower, tim.golden, vstinner, zach.ware
Date 2021-09-06.21:49:38
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <>
> Perhaps these critical code sections should have been left as macros. It is difficult to assuring system wide inlining across modules.

These functions were not converted recently to static inline function. For example, Py_INCREF() was already a static inline function in Python 3.9. I don't think that any decision should be taken before performances have been analyzed in depth.

I'm not convinced that there is really a performance regression. I'm not sure how benchmarks are run on Windows.

> I measured overheads of (1)~(4) on my own build whose eval-loop uses macros instead of them.

I don't understand how to read the table.

Usually, I expect a comparison between a reference build and a patch build, but here you seem to use 3.10a7 as the reference to compare results. I'm not sure that geometric means can be compared this way.
Date User Action Args
2021-09-06 21:49:38vstinnersetrecipients: + vstinner, rhettinger, paul.moore, tim.golden, Mark.Shannon, zach.ware, steve.dower, neonene
2021-09-06 21:49:38vstinnersetmessageid: <>
2021-09-06 21:49:38vstinnerlinkissue45116 messages
2021-09-06 21:49:38vstinnercreate