Always push the current user_error/exception_handler to the stack,
even when it is empty, so restore_error_handler() always works as
expected.
The user_error_handler is especially temporarily empty when we are inside
the error handler, which caused inconsistent behaviour before.
Unlink the current stack frame before freeing CVs or extra args.
This means it will no longer show up in back traces that are
generated during CV destruction.
We already did this prior to destructing the object/closure,
presumably for the same reason.
Finally blocks in generators may be invoked during shutdown, in
which case we don't have a stack frame. Similar to what
zend_call_function does, we still need to rethrow these exceptions,
otherwise they will be hidden (and leak).
The "return" in the for loop should have been a break on the switch,
otherwise the result is just ignored... but because it prevents
evaluation of the other operand, it also violates the invariant that
everything has been constant evaluated, resulting in an assertion
failure.
The for loop isn't correct in any case though, because it's not legal
to determine the result based on just the second operand, as the
first one may have a side-effect that cannot be optimized away.
NULL out the execute_data before destroying it, otherwise GC may
trigger while the execute_data is partially destroyed, resulting
in double-frees.
The handling of call stack unfreezing is a bit awkward because it's
a ZEND_API function, so we can't change the signature.
The properties HT may be a GC root itself, so we need to remove it.
I'm not sure this issue actually applies to PHP 7.2, but committing
it there to be safe. As seen from the test case, the handling here
is rather buggy on 7.2.
In PHP 7.3 shadow properties are no longer duplicated. Make sure we
only release them if the property was defined on the parent class,
which means that it changed from private->shadow, which is where
duplication does happen.
When cleaning nops in the dfa pass, we were always keeping the
smart branch inhibiting nop that occurs directly before the jump
instruction. However, as we skip unreachable blocks entirely, it
may happen that we need to keep a nop that occurs further back,
prior to the unreachable blocks. Account for that case now.
We should really do something about the smart branch situation,
this is very fragile...
If we perform a class fetch that is not marked as exception safe,
convert exceptions thrown by autoloaders into a fatal error.
Ideally fetching the interfaces would be exception safe, but as it
isn't right now, we must abort at this point.
Nikita Popov