Do not report maybe in RandomIntParametersRule when either argument is an unbounded IntegerRangeType

[phpstan-bot]

Summary

RandomIntParametersRule was reporting false-positive maybe errors for random_int() calls where one or both arguments had unbounded integer range types (e.g. int<0, max>). This happened because the rule already skips plain int (IntegerType) but treated unbounded IntegerRangeType values differently, even though they carry the same level of uncertainty about the relationship between min and max.

Changes

• Modified src/Rules/Functions/RandomIntParametersRule.php to suppress maybe reports when either the min or max argument type is an IntegerRangeType with an unbounded end (null min or null max). Definite (yes) errors are still reported even for unbounded ranges.
• Added regression test in tests/PHPStan/Rules/Functions/data/bug-14468.php covering:
  • The original issue: random_int() inside a while ($min < $max) loop with strlen-based max
  • Both arguments with unbounded max (int<0, max>)
  • Both arguments with unbounded min (int<min, 0>)
  • Unbounded min arg with bounded max arg
  • Bounded min arg with unbounded max arg
  • Unbounded ranges where the result is definitely wrong (int<5, max> vs int<min, 3>) — still correctly reported

Root cause

The rule checks $maxType->isSmallerThan($minType) and reports when the result is maybe (with reportMaybes enabled). For unbounded ranges like int<0, max>, isSmallerThan trivially returns maybe because unbounded ranges overlap with almost everything. The rule already had a bail-out for plain int (IntegerType) at lines 59-63, but IntegerRangeType with one null bound was not covered by that bail-out.

The fix extends the maybe suppression to also cover unbounded IntegerRangeType, making it consistent with the existing IntegerType handling.

Searched for analogous rules using isSmallerThan — RandomIntParametersRule is the only consumer, so no other rules are affected.

Test

• testBug14468 in RandomIntParametersRuleTest — verifies no errors for unbounded range combinations and still reports errors for definitely-wrong unbounded combinations.
• All existing tests (testFile, testBug6361) continue to pass unchanged.

Fixes phpstan/phpstan#14468

[VincentLanglet]

Dunno if we should confirm this with ondrej...

[staabm]

I remember we did similar thibgs somewhere else