[ruby-dev:52119] [Ruby master Bug#20767] VM_ASSERT fails on clear_method_entry_by_id_in_class() under a specific situation

["alanwu (Alan Wu) via ruby-dev" <ruby-dev@ml.ruby-lang.org>]

Issue #20767 has been updated by alanwu (Alan Wu).

Status changed from Closed to Open

Reopening since the assert failed again on CI:
http://ci.rvm.jp/logfiles/brlog.trunk-repeat20-asserts.20241029-211036#L2454

----------------------------------------
Bug #20767: VM_ASSERT fails on clear_method_entry_by_id_in_class() under a specific situation
https://bugs.ruby-lang.org/issues/20767#change-110290

* Author: tagomoris (Satoshi Tagomori)
* Status: Open
* ruby -v: ruby 3.4.0dev (2024-09-27T17:45:22Z vm_assertion_cme_o.. 027ef60500) +PRISM [arm64-darwin23]
* Backport: 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN
----------------------------------------
VM_ASSERT in `clear_method_entry_by_id_in_class()` (vm_method.c) fails when:

* Kernel#require is refined
* Kernel#require is replaced by `alias_method` and `define_method`
* User code calls `require`
* The aliased method entry is undefined

Under this situation, `VM_ASSERT(cme->owner == T_CLASS)` fails in `clear_method_entry_by_id_in_class`.
This CI failure shows it: https://github.com/ruby/ruby/actions/runs/11081754435/job/30793895707?pr=11715

In this case, the cme had:

* def->type: CFUNC
* defined_class: Kernel
* owner: Kernel

I couldn't find the root cause of why Kernel is set on cme->owner, and why the owner is checked here.
How should we fix this problem?

* Just delete the VM_ASSERT() (because the cme is just to be invalidated)
* Fix the root cause of the invalid&unexpected cme->owner



-- 
https://bugs.ruby-lang.org/
_______________________________________________
ruby-dev mailing list -- ruby-dev@ml.ruby-lang.org
To unsubscribe send an email to ruby-dev-leave@ml.ruby-lang.org