Fix for assertCount expected size for Generator

[kubawerlos]

#3302

[sebastianbergmann]

This does not solve the underlying problem that assertCount() modifies the generator.

[kubawerlos]

I don't know if it is possible in PHP.

the only sensible thing to do here would be to disallow assertCount() on generators.

Should we follow this path? If so, target it to master as BC break or to 7.3 as bugfix?

[JeroenDeDauw]

@sebastianbergmann I don't see why that would be a problem. Going through a generator always does this in PHP, so it is expected behavior. Disallow it in assertCount would just force people to create such assertions themselves, which seems like loss-loss to me.

[epdenouden]

From a developer perspective: whatever the underlying language feature does internally, I would expect an assertCount to work as the name suggests and not give me "size 0" errors.

[guilliamxavier]

Please consider:

$gen2 = function () { yield 1; yield 2; };
$gen3 = function () { yield 1; yield 2; yield 3; };

$count2 = new Count(2);
Assert::assertTrue($count2->evaluate($gen2(), '', true));
Assert::assertFalse($count2->evaluate($gen3(), '', true)); // this should still work

$count3 = new Count(3);
Assert::assertFalse($count3->evaluate($gen2(), '', true));
Assert::assertTrue($count3->evaluate($gen3(), '', true)); // this should still work

Maybe need to reset $this->countOfGenerator = null; at the beginning of function matches?

Also please consider #3302 (comment) (i.e. there are non-Countable, Traversable things that are not instanceof Generator but suffer the same kind of problem)...

[JeroenDeDauw]

I can confirm that this PR is breaking that behavior and does not clean up the already existing confusion between Generators and Iterables that cannot be rewinded. I'll submit a modified solution in a bit.

Edit: here it is: #3316

[sebastianbergmann]

Superseded by #3316.