Method lookup caching doesn't happen until a method has actually been called. Because the @ISA manipulation happens in a BEGIN block, this should occur before any method calls, thus before any method lookups have been cached.

Also, according to perlobj any changes to @ISA should invalidate the cache. There may be bugs lurking somewhere in perl's method lookup caching mechanism, but none that seem to be triggered by the technique I've discussed.

After all, the situation where package $X manipulates package $Y's <c>@ISA at BEGIN-time is really common. (Consider the case where $X eq "base" or $X eq "parent". Sure, this is happening at package $Y's request, but from Perl's perspective the same thing is happening.)

With Moose it happens at run-time, and this doesn't seem to cause any problems.

... BEGIN ... perlobj ... parent ... perspective ..

By the time you load Data::Dumper or whatever, resolution method could have happened

But yeah, I did say I vaguely recall, esp because I didn't feel like digging up history or pinning down the details :)

As for parent... perspective is exactly the point. If you're bothering to write a module to share, you might as well do it right, which is not dickering around with other peoples class data :) delegate (AUTOLOAD or ...)

At least if you monkey patch, warnings will issue "Subroutine %s redefined"

With Moose it happens at run-time, and this doesn't seem to cause any problems.

Sure it does :) Moose Cookbook - Meta Recipe 5 Won't Run

I also vaguely recall method lookup caching could survive @ISA updates.