Fix regression in overload resolution picking eta-expanded method #21847
Conversation
|
I didn't look at what change or changes caused the code to regress, seeing as it was so many versions ago. So I just took this as it stood. We seem to have a |
There was a problem hiding this comment.
I get the reasoning behind the change, and the code is fine. But this would be a rather sweeping spec change in a very sensitive area. We risk
- code changing behavior at runtime because another alternative is now chosen
- compiletime blowouts since converting to function types when comparing alternatives can be expensive and we know that alternative comparison can already be very bad.
So my gut feeling is not to go ahead with this. It's too risky. When it comes to overloading resolution we should give high priority to the status quo and only change something when we have bad regressions, like existing programs changing runtime behavior.
I left a comment on the original issue why this happens. The problem is, the program in #21727 originally compiled only because of a very dubious addition of unknown origins that has since been removed. Since it seems OpenCI was OK with this removal, I would say let's treat that as the status quo to preserve.
ok |
Fixes #21727