Allow writing to readonly properties during cloning

[nicolas-grekas]

Here is my try at fixing cloning + readonly. This patch allows readonly properties to be set or unset once during cloning, ie inside __clone(). This makes it straightforward to clone a property when cloning an object. For withers, some coordination is required between __clone and wither methods. See comments for what I mean.

I would love to see this merged as a bugfix in 8.0 (even-though we discuss improved syntax for withers, which don't need to solve for this PR.)

[bwoebi]

I generally agree that clone should generally reset the immutability of readonly.

While I think it makes sense to add a new syntax akin to $foo = clone $bar { prop1 = $something }; or similar (and should probably also apply to new,) this would need to go through the RFC process and should not be tied directly to this feature.

[cmb69]

I would love to see this merged as a bugfix in 8.0 (even-though we discuss improved syntax for withers.)

I don't think there is a bug. The respective RFC mentions:

The clone-based implementation will result in an error, because the x property on the cloned object is already initialized, and modification is thus rejected. This is by design: The property does indeed get modified post-initialization, and the fact that this is only “temporary” is ultimately irrelevant.

So this was a well-known "limitation" of that RFC.

[TysonAndre]

php > class Foo { public readonly int $x = 0; }
Fatal error: Readonly property Foo::$x cannot have default value in php shell code on line 1

I think the test failures might be related to the fact php internally expects properties with defined values not to have the type IS_PROP_UNINIT, but I'm not completely certain. UNINIT means the value would be IS_UNDEF. In the failing clone tests, they'd have type int for $this->foo = 0, so it sees the property's type is defined and emits the error.

I also don't remember how to properly solve that or what other edge cases there may be

[nicolas-grekas]

Thanks for chiming in @TysonAndre.
I've figured our that there is another field in the zval that can be used as a flag for what we need: u1.v.u.extra, which looks unused.
I've updated the PR to used it.
I've also figured out other places where this flag should be checked so that my test case passes.
There's a segfault on tests/classes/clone_003.phpt which I don't understand. Anyone would for me please?

[nicolas-grekas]

Closing in favor of #9497, thanks @kocsismate for taking over!