assertSame broken after 9.5.14 version

[pablorsk]

Q	A
PHPUnit version	9.5.14
PHP version	8.1.10
Installation Method	Composer

Summary

Assuming the following test:

public function testAssertSame()
{
    static::assertSame(40.7, 40.8-0.1);
}

In version 9.5.13 this test works perfect:

.                                                                   1 / 1 (100%)

But, in version 9.5.14, it breaks

1) SomeTest::testAssertSame
Failed asserting that 40.699999999999996 is identical to 40.7.

Any ideas? I try to set serialize_precision = 16, but the problem continued.

[sebastianbergmann]

CC @marijnvanwezel @mvorisek

[sebastianbergmann]

This may be fixed by #4972.

[mvorisek]

assertSame pass only if the values are strictly the same, in your case, the values are not the same, you need to compare them with a small delta using assertEqualsWithDelta

[bancer]

@sebastianbergmann #4972 did not fix the problem:

[mvorisek]

@bancer the assertion is correctly not met - https://3v4l.org/5CiYa

[bancer]

@bancer the assertion is correctly not met - https://3v4l.org/5CiYa

Well, everybody could assume that 40.7 is mathematically the same value as 40.8-0.1.

[mvorisek]

you can use $delta in assertEquals, assertSame checks (simply said) for 1:1 binary value

[anders-photowall]

I ran in to this problem today after doing what is supposed to be a safe composer update since we have caret notation and minor and patch updates should never have breaking changes. This very much is a breaking change and not even on a minor but patch version which I find extremely surprising. Was this an unintended and unknown breaking change? Or should the phpunit lib be considered unsafe and potentially breaking to update even patch changes on? That should be very clearly stated in the documentation in that case.

[marijnvanwezel]

I think that strictly speaking, the change could be considered a patch only. The previous behaviour (<= 9.5.13) was incorrect, as an assertion could pass even if it clearly shouldn't have. Comparisons between floats should not be done with assertSame, but it might be good to more clearly mention this in the documentation.

That being said, since many people relied on this incorrect behaviour, including it in a major release might have been a better approach.