Description

markshannon

opened on Aug 24, 2022

Issue body actions

In order to better support shim frames, trampolines, Cython frames and other such bits and pieces, we need to relax the requirement that the f_func field is a function.
The only reason for it to be a function is that it needs to hold a strong reference to builtins and globals, as the frame only has a borrowed reference to those. Any object with those references will do.
In fact, for code that has no builtins or globals (many shims and trampolines) there is no need for those references and f_func could even be None.

Activity

1. markshannon

   added

   performancePerformance or resource usagePerformance or resource usage

   on Aug 24, 2022
2. iritkatriel commented on Aug 24, 2022

   

   iritkatriel

   on Aug 24, 2022

   Member

   More actions

   Does it really need to allow any object? Could you instead define a small number of object types that it can be?
3. markshannon commented on Aug 24, 2022

   

   markshannon

   on Aug 24, 2022

   MemberAuthor

   More actions

   I don't see any advantage to restricting the possible types. Allowing any object is no worse for any code reading from f_funcobj and more flexible for anything storing to it.

   Given the values of f_globals and f_builtins can no longer be trusted, I think I'll make them void * just to make it clear that if you didn't set them, then you can't trust them.

   Reacted by Irit Katriel
4. markshannon

   added a commit that references this issue on Aug 25, 2022

   GH-96237: Allow non-functions as reference-holder in frames. (GH-96238)

   c09fa75
5. carljm commented on Feb 14, 2023

   

   carljm

   on Feb 14, 2023

   Member

   More actions

   Currently whatever is in f_funcobj also needs to have a func_closure. #101310 demonstrates one way to remove this requirement.
6. carljm commented on Feb 14, 2023

   

   carljm

   on Feb 14, 2023

   Member

   More actions

   Should this issue be closed, given the attached PR was merged? Or is there more planned here?
7. markshannon commented on Apr 3, 2023

   

   markshannon

   on Apr 3, 2023

   MemberAuthor

   More actions

   Superseded by #100987
8. markshannon

   closed this as not plannedon Apr 3, 2023

Issue actions