Disallow invoking GMP constructor

[nielsdos]

Fixes GH-10155

Note: I targeted master because I wasn't sure whether this could be considered a BC break.

[Girgias]

This is a bug please backport to 8.1

[nielsdos]

I've update the PR, 8.2 and master additionally need this PR: #10160

[cmb69]

I'm slightly concerned about no longer allowing new GMP(); maybe we should keep the ability to call the constructor without argument for PHP 8.1?

[Girgias]

I'm slightly concerned about no longer allowing new GMP(); maybe we should keep the ability to call the constructor without argument for PHP 8.1?

I don't know, because if anyone is instantiating it via new they probably have a bug as nothing would actually work except if it's meant to be initialized to 0, which seems to be very unlikely.

[cmb69]

[…] except if it's meant to be initialized to 0, which seems to be very unlikely.

When it comes to the creativity of users, almost nothing is unlikely. After all, GMP objects can be constructed normally since they've been invented. Given that there is no need to hurry (next QAs are scheduled for Jan, 19th), it might be a good idea to ask on internals about this change (and maybe on other channels as well). Maybe I'm just too overcautious.

People using new GMP() with and without argument does seem to occur according to sourcegraph: https://sourcegraph.com/search?q=context:global+new%5CsGMP%5C%28%5Cd*%5C%29+lang:php&patternType=regexp&case=yes&sm=1

At least the first occurrences are about namespaced GMP classes, so that wouldn't affect this change. Searching for \GMP (FQN) yields no results. The truth might be somewhere in between.

[Girgias]

[…] except if it's meant to be initialized to 0, which seems to be very unlikely.

When it comes to the creativity of users, almost nothing is unlikely. After all, GMP objects can be constructed normally since they've been invented. Given that there is no need to hurry (next QAs are scheduled for Jan, 19th), it might be a good idea to ask on internals about this change (and maybe on other channels as well). Maybe I'm just too overcautious.

People using new GMP() with and without argument does seem to occur according to sourcegraph: https://sourcegraph.com/search?q=context:global+new%5CsGMP%5C%28%5Cd*%5C%29+lang:php&patternType=regexp&case=yes&sm=1

At least the first occurrences are about namespaced GMP classes, so that wouldn't affect this change. Searching for \GMP (FQN) yields no results. The truth might be somewhere in between.

sigh at this point we may as well just implement the constructor properly, so one can initialize a GMP object this way. Because I find it just strange to allow instantiating it while not passing any arguments. In any case, this is going to be a nightmare to document.

[cmb69]

sigh at this point we may as well just implement the constructor properly, so one can initialize a GMP object this way. Because I find it just strange to allow instantiating it while not passing any arguments. In any case, this is going to be a nightmare to document.

Like I said, "maybe I'm just too overcautious". :)

[Girgias]

From the ML discussion, it seems the preferred option is to make a proper constructor that is like gmp_init(). @nielsdos do you want to tackle this?

[nielsdos]

From the ML discussion, it seems the preferred option is to make a proper constructor that is like gmp_init().

That makes the most sense to me as well.

@nielsdos do you want to tackle this?

Sure, I'll do it soon-ish :)

[nielsdos]

Superseded by #10225