The GMP class was introduced as part of the
Internal operator overloading and GMP improvements
RFC.
This RFC also converted the GMP extension to use an object instead of a resource. One of the first proposals of its kind. Since then numerous resources have been converted to what we call opaque objects. [1][2][3] And the finalization of the process was codified in the Resource to object conversion RFC which was accepted in late 2023.
One element of the resource to object migration is to make the object opaque,
which in general implied, being final, being not serializable,
not constructible via new, cannot be cast, and to not implement any methods.
The GMP class is a notable exception to many of the above restrictions.
It is a non-final class, that can be instantiated via new, and is serializable.
The serialization aspect was intended as part of the initial RFC,
but the other aspects were not.
Indeed, the GMP class did not have a proper constructor to initialize the object until
PHP 8.2.4 after this fact was discovered and discussed on the mailing list in late December 2022. [4]
As such, we deem the fact that GMP was not made final an oversight and want to correct this.
Make the GMP class final.
Any userland class that extends GMP would now throw a compile time error.
From an analysis of private and publicly available code bases with Exakat no one has ever extended this class, as such the BC Break is mostly theoretical.
In the discussion of the new BCMath\Number class introduced via the
Support object type in BCMath
RFC, it was explicitly discussed if the class should be final or not.
And the consensus was that it should be final due to concerns about overloading the methods.
However, people seemingly want to “enhance” GMP to achieve userland operator overloading
by hooking into it's logic when extending it.
This is in total contradiction with the above RFC discussion which happened a few months ago.
This resembles more of a “hack” than anything else and support for a proper form of operator overloading [5][6]
should be revisited instead.
Another alternative would be to create a dedicated PHP extension with an object that overloads the do_operator object handler.
Therefore, we think it is crucial to close this avenue for discussion once and for all.
Next minor version, which is PHP 8.4.
As per the voting RFC a yes/no vote with a 2/3 majority is needed for this proposal to be accepted.
Voting started on 2024-07-13 and will end on 2024-07-27.
[1] “PHP RFC: Migration Hash Context from Resource to Object”: https://wiki.php.net/rfc/hash-context.as-resource
[2] “Convert ext/xml to use an object instead of resource” PHP Internals Mailing List thread: https://externals.io/message/104361
[3] “Resource to opaque object migration” php/php-tasks issue: https://github.com/php/php-tasks/issues/6
[4] “What to do with opaque GMP objects which allows instantiation” PHP Internals Mailing List thread: https://externals.io/message/119216
[5] “PHP RFC: Userspace operator overloading”: https://wiki.php.net/rfc/userspace_operator_overloading
[6] “PHP RFC: User Defined Operator Overloads”: https://wiki.php.net/rfc/user_defined_operator_overloads