Internal error: abs(PHP_INT_MIN) #14946

Bug report

Analysing abs(PHP_INT_MIN) crashes with an internal error:

Internal error: PHPStan\Type\Constant\ConstantIntegerType::__construct():
Argument #1 ($value) must be of type int, float given,
called in src/Type/Constant/ConstantIntegerType.php on line 88

ConstantIntegerType::toAbsoluteNumber() does new self(abs($this->value)), but PHP's abs(PHP_INT_MIN) returns a float, because -PHP_INT_MIN is not representable as an int:

var_dump(abs(PHP_INT_MIN)); // float(9.2233720368548E+18)

Note that abs(-9223372036854775808) on the second line of the playground link does not crash — that literal is already a float, so it takes a different code path.

IntegerRangeType::toAbsoluteNumber() has the same problem in two places ($this->min * -1 and $this->max * -1), so these crash too:

/** @var int<-9223372036854775808, 0> $a */
abs($a);   // Internal error: IntegerRangeType::fromInterval(): Argument #2 ($max) must be of type ?int, float given

/** @var int<-9223372036854775808, -1> $b */
abs($b);   // same

/** @var int<min, -9223372036854775808> $c */
abs($c);   // Internal error: ... Argument #1 ($min) ...

Code snippet that reproduces the problem

https://phpstan.org/r/fa1c992f-684d-4abd-a6fc-b248b18180c6

Expected output

No internal error. abs(PHP_INT_MIN) should be inferred as a float:

\PHPStan\dumpType(abs(-9223372036854775807 - 1)); // 9.223372036854776E+18

For integer ranges, I would expect the existing approximation to be kept, so that a lower bound of PHP_INT_MIN behaves like an unbounded one (abs(int) is already int<0, max> today, and abs(int<min, 0>) is already int<0, max>):

/** @var int<-9223372036854775808, 0> $a */
\PHPStan\dumpType(abs($a));  // int<0, max>

/** @var int<-9223372036854775808, -1> $b */
\PHPStan\dumpType(abs($b));  // int<1, max>

int<min, -9223372036854775808> holds exactly one value, so abs() of it is that same float.

I have a fix ready and will open a PR right away.

Did PHPStan help you today? Did it make you happy in any way?

Yes — I found this while investigating how PHPStan models integer width, and PHPStan's own error messages pointed me straight at the offending line.