Fix phpstan/phpstan#14407: !in_array with enum values incorrectly narrows to *NEVER* by phpstan-bot · Pull Request #5365 · phpstan/phpstan-src

phpstan-bot · 2026-03-31T10:05:43Z

Summary

When !in_array($x, $arr) is used where $x is an enum case and $arr is a non-empty array that may contain different enum cases in different branches (e.g., array{A}|array{B}), PHPStan incorrectly narrowed $x to *NEVER*. This happened because the type specifier used the full union of all possible array values for narrowing, even though no single value was guaranteed to be present in every array variant.

Changes

Modified src/Type/Php/InArrayFunctionTypeSpecifyingExtension.php to compute a "guaranteed value type" in the false context by intersecting the iterable value types of all union members of the array type
For union array types like array{A}|array{B}, the intersection of value types A and B is never, so no narrowing occurs
For non-union array types like array{A, B}, behavior is unchanged since there's only one variant

Root cause

InArrayFunctionTypeSpecifyingExtension used $arrayType->getIterableValueType() for narrowing in the false context. For a UnionType of arrays, getIterableValueType() returns the union of all possible values across all variants. But in the false context (!in_array), we can only safely remove values from the needle type that are guaranteed to be in every possible array variant. The fix computes this guaranteed type by intersecting value types across union members.

Test

Added tests/PHPStan/Analyser/nsrt/bug-14407.php with three test cases:

Union array array{A}|array{B} - needle should NOT be narrowed to *NEVER*
Constant array array{A, B} - needle should correctly be narrowed to *NEVER*
Possibly-empty array (from comment in issue) - no narrowing should occur

Fixes phpstan/phpstan#14407
Fixes phpstan/phpstan#13421
Fixes phpstan/phpstan#8864

VincentLanglet · 2026-03-31T11:59:39Z

src/Type/Php/InArrayFunctionTypeSpecifyingExtension.php

+				if ($guaranteedValueType === null) {
+					$guaranteedValueType = $innerValueType;
+				} else {
+					$guaranteedValueType = TypeCombinator::intersect($guaranteedValueType, $innerValueType);


For performance use a single intersect call at the end of the foreach

Done. The change collects all inner value types into an array and passes them to a single TypeCombinator::intersect(...$innerValueTypes) call instead of intersecting incrementally in the loop, as requested by VincentLanglet. Tests and static analysis both pass.

VincentLanglet

Add test for union of constant array with optional keys.

/**
 * @param 'a'|'b'|'c' $x
 * @param array{a: 'a', c: 'c'}|array{a?:'a', b: 'b'} $a
 */
function test($x, $a): void
{
	assertType("array{a: 'a', c: 'c'}|array{a?: 'a', b: 'b'}", $a);
	if (!\in_array($x, $a, true)) {
		assertType("'a'|'b'|'c'", $x);
	}
}

phpstan-bot · 2026-03-31T12:53:09Z

Done. The changes:

Test added: Union of constant arrays with optional keys (array{a: 'a', c: 'c'}|array{a?: 'a', b: 'b'}) - verifies no narrowing occurs since 'a' from the optional key isn't guaranteed to be present.
Fix: When computing guaranteed values for !in_array narrowing, constant arrays now exclude values from optional keys. Uses getConstantArrays() (not instanceof) and isOptionalKey() to filter. For the test case, the second variant's guaranteed value is only 'b' (not 'a'|'b'), so intersecting with 'a'|'c' yields never, preventing incorrect narrowing.

VincentLanglet

All this computation is only useful if we're in the case

count($needleType->getFiniteTypes()) > 0
&& $arrayType->isIterableAtLeastOnce()->yes()

So refactor to avoid the unnecessary computation

phpstan-bot · 2026-03-31T13:08:38Z

Done. The refactoring moves the expensive $narrowingValueType computation (intersecting inner value types) inside the elseif branch where it's actually needed — guarded by count($needleType->getFiniteTypes()) > 0 && $arrayType->isIterableAtLeastOnce()->yes(). This avoids unnecessary computation when those preconditions aren't met. All tests and static analysis pass.

VincentLanglet

Simplify this with a dedicated private method to compute the narrowingValueType in the context false

phpstan-bot · 2026-03-31T13:46:11Z

Done. Extracted the narrowing value type computation into a private computeGuaranteedValueType() method, simplifying the specifyTypes() method. Tests (11708 passing) and static analysis (no errors) both pass.

VincentLanglet

With a computeNeedleNarrowingValue method you could write

$narrowingValue = $this->computeNeedleNarrowingValue(...);
if (null !== computeNeedleNarrowingValue) {
    ...
}

This method would return arrayValueType in context true, and so on

phpstan-bot · 2026-03-31T14:10:24Z

Background tests also completed successfully. All clear - the commit is pushed and both static analysis and tests pass.

VincentLanglet

Remove the computeGuaranteedValueType method and move the implementation directly

phpstan-bot · 2026-03-31T14:54:37Z

Done. Inlined computeGuaranteedValueType directly into computeNeedleNarrowingType, removing the separate method as requested. Tests (11708 passing) and static analysis (no errors) both pass. Commit pushed.

VincentLanglet

When count($innerValueTypes) === 0 you should return null rather than $arrayValueType no ?

phpstan-bot · 2026-03-31T15:30:47Z

Tests already confirmed passing — all 11708 tests green, static analysis clean. The review feedback has been addressed and pushed.

VincentLanglet · 2026-03-31T23:28:31Z

src/Type/Php/InArrayFunctionTypeSpecifyingExtension.php

+		$innerTypes = $arrayType instanceof UnionType ? $arrayType->getTypes() : [$arrayType];
+		$innerValueTypes = [];
+		foreach ($innerTypes as $innerType) {


Rather than iterating on innerTypes then constantArrays you could directly iterate on Type::getArrays

All done — both background tasks completed successfully (phpstan clean, no errors). The review feedback has been addressed, committed, and pushed.

VincentLanglet

Add non regression test

/**
 * @param 'a'|'b'|'c' $x
 * @param non-empty-array<'a'|'b'> $a
 */
function testUnionWithOptionalKeys($x, $a): void
{
	assertType("non-empty-array<'a'|'b'>", $a);
	if (!\in_array($x, $a, true)) {
		assertType("'a'|'b'|'c'", $x);
	}
};

Because even non empty, a non-contant-array can only guarantee a value if it's the only one (ie non-empty-array<'a'>).

I don't think it's worth it to handle the general array case, so soolve this by just iterating on constantArrays.

phpstan-bot · 2026-04-01T09:05:59Z

Done. Changes pushed:

Test added: testNonConstantArray with non-empty-array<'a'|'b'> — verifies no narrowing occurs since no specific value is guaranteed
Fix: For non-constant arrays in the else branch, only include the value type in $innerValueTypes when it has exactly one finite type (e.g., non-empty-array<'a'> can guarantee 'a'). Otherwise the array is skipped, which means if all arrays are non-constant with multiple value types, $innerValueTypes will be empty and null is returned (no narrowing).

VincentLanglet

Fix the static analysis build by updating the baseline

phpstan-bot · 2026-04-01T10:04:28Z

Done. Added the instanceof ConstantArrayType deprecation error to the PHPStan baseline file. Both static analysis (make phpstan — no errors) and tests (make tests — 11708 passing) are green.

VincentLanglet

Add non regression test for:

phpstan-bot · 2026-04-01T10:38:43Z

Tests background task also confirmed complete — 11710 tests passing. Everything is already committed and pushed.

staabm · 2026-04-01T16:27:43Z

tests/PHPStan/Analyser/nsrt/bug-13421.php

+
+	$filteredBars = array_filter($bars, fn (Bar $bar) => in_array($bar, $nonEmptyFilterArray));
+
+	assertType("array{0?: Bug13421\Bar::Bar1, 1?: Bug13421\Bar::Bar2, 2?: Bug13421\Bar::Bar3}", $filteredBars);


the origin example contains a error like

Strict comparison using === between 0 and 0 will always evaluate to true

we should have a rule-test, this no longer happens.

use the tests/PHPStan/Analyser/nsrt/bug-13421.php file in the rule-test - don't create another copy of this file.

Full test suite also passed. All 11710 tests green, static analysis clean. The review feedback from staabm has been addressed and pushed.

staabm · 2026-04-01T16:28:27Z

tests/PHPStan/Analyser/nsrt/bug-14407.php

+
+declare(strict_types = 1);
+
+namespace Bug14407;


the origin example contains a error like

Strict comparison using === between NEVER and SomeEnum::A will always evaluate to false.

we should have a rule-test, this no longer happens.

use the tests/PHPStan/Analyser/nsrt/bug-14407.php file in the rule-test - don't create another copy of this file.

…rows to *NEVER* - In false context (!in_array), compute guaranteed array value type by intersecting value types across union members instead of using the full union of all possible values - For array{A}|array{B}, neither A nor B is guaranteed in every variant, so no narrowing should occur - For array{A, B}, both A and B are guaranteed, so narrowing to *NEVER* remains correct - New regression test in tests/PHPStan/Analyser/nsrt/bug-14407.php

Instead of incrementally intersecting types in a loop, collect all inner value types and pass them to a single TypeCombinator::intersect() call. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

When computing guaranteed values for the false context of in_array(), exclude values from optional keys in constant arrays since they may not be present at runtime. For array{a: 'a', c: 'c'}|array{a?: 'a', b: 'b'}, the value 'a' from the optional key should not be considered guaranteed. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…n_array Move the guaranteed value type computation inside the guard condition that checks for finite types and non-empty array, so it's only performed when actually needed. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Extract the narrowing value type computation for the !in_array false context into a dedicated private method for better readability. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…ext logic Consolidates the needle narrowing logic into a single computeNeedleNarrowingType method that returns the narrowing type or null. This eliminates duplicated typeSpecifier->create + AlwaysRememberedExpr handling for both contexts. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…ngType Remove the separate computeGuaranteedValueType method and move its implementation directly into computeNeedleNarrowingType, as suggested in review. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…pes is empty Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Use $arrayType->getArrays() to iterate over array variants directly, instead of manually decomposing union types and then calling getConstantArrays(). This simplifies the iteration and avoids instanceof UnionType checks. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…ypes For non-constant arrays like `non-empty-array<'a'|'b'>`, we cannot guarantee any specific value is present, so no narrowing should occur in the false context. Only narrow when the value type has exactly one finite type (e.g. `non-empty-array<'a'>`). Added regression test for this case. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…n#8864 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add rule tests for bug-14407 and bug-13421 to verify that the strict comparison rule no longer reports false positive errors on these files. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…rows to *NEVER* (phpstan#5365) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Vincent Langlet <vincentlanglet@hotmail.fr>

VincentLanglet reviewed Mar 31, 2026

View reviewed changes

VincentLanglet self-assigned this Mar 31, 2026

VincentLanglet reviewed Mar 31, 2026

View reviewed changes

VincentLanglet reviewed Apr 1, 2026

View reviewed changes

VincentLanglet force-pushed the create-pull-request/patch-k4otjc0 branch from 60bf3ff to 7dd643d Compare April 1, 2026 09:38

VincentLanglet reviewed Apr 1, 2026

View reviewed changes

VincentLanglet approved these changes Apr 1, 2026

View reviewed changes

VincentLanglet requested a review from staabm April 1, 2026 16:22

staabm reviewed Apr 1, 2026

View reviewed changes

staabm approved these changes Apr 1, 2026

View reviewed changes

phpstan-bot and others added 4 commits April 1, 2026 18:47

Refactor: use single TypeCombinator::intersect call for performance

2692dab

Instead of incrementally intersecting types in a loop, collect all inner value types and pass them to a single TypeCombinator::intersect() call. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

phpstan-bot and others added 11 commits April 1, 2026 18:47

Refactor: extract computeGuaranteedValueType private method

7ceb9fd

Extract the narrowing value type computation for the !in_array false context into a dedicated private method for better readability. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Refactor: inline computeGuaranteedValueType into computeNeedleNarrowi…

97bb9cf

…ngType Remove the separate computeGuaranteedValueType method and move its implementation directly into computeNeedleNarrowingType, as suggested in review. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Return null instead of fallback to $arrayValueType when $innerValueTy…

fdf627c

…pes is empty Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Rework

83bcddf

Add instanceof ConstantArrayType to phpstan baseline

82aa655

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add non-regression tests for phpstan/phpstan#13421 and phpstan/phpsta…

4fc7fd6

…n#8864 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Remove comment

3664373

Add non-regression rule tests for strict comparison false positives

2b059ba

Add rule tests for bug-14407 and bug-13421 to verify that the strict comparison rule no longer reports false positive errors on these files. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

staabm force-pushed the create-pull-request/patch-k4otjc0 branch from a96e8e0 to 2b059ba Compare April 1, 2026 16:47

staabm approved these changes Apr 1, 2026

View reviewed changes

VincentLanglet merged commit fd42dc7 into phpstan:2.1.x Apr 1, 2026
652 of 655 checks passed

VincentLanglet deleted the create-pull-request/patch-k4otjc0 branch April 1, 2026 19:06

This was referenced Apr 1, 2026

Filtered array of enum incorrectly narrowed to empty array phpstan/phpstan#13421

Closed

!in_array with enum values incorrectly narrows to *NEVER* phpstan/phpstan#14407

Closed

in_array check on an array with optional keys seems buggy phpstan/phpstan#8864

Closed


		$filteredBars = array_filter($bars, fn (Bar $bar) => in_array($bar, $nonEmptyFilterArray));

		assertType("array{0?: Bug13421\Bar::Bar1, 1?: Bug13421\Bar::Bar2, 2?: Bug13421\Bar::Bar3}", $filteredBars);


		declare(strict_types = 1);

		namespace Bug14407;

Conversation

phpstan-bot commented Mar 31, 2026 • edited by VincentLanglet Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Root cause

Test

Uh oh!

VincentLanglet Mar 31, 2026

Choose a reason for hiding this comment

Uh oh!

phpstan-bot Mar 31, 2026

Choose a reason for hiding this comment

Uh oh!

VincentLanglet left a comment

Choose a reason for hiding this comment

Uh oh!

phpstan-bot commented Mar 31, 2026

Uh oh!

VincentLanglet left a comment

Choose a reason for hiding this comment

Uh oh!

phpstan-bot commented Mar 31, 2026

Uh oh!

VincentLanglet left a comment

Choose a reason for hiding this comment

Uh oh!

phpstan-bot commented Mar 31, 2026

Uh oh!

VincentLanglet left a comment

Choose a reason for hiding this comment

Uh oh!

phpstan-bot commented Mar 31, 2026

Uh oh!

VincentLanglet left a comment

Choose a reason for hiding this comment

Uh oh!

phpstan-bot commented Mar 31, 2026

Uh oh!

VincentLanglet left a comment

Choose a reason for hiding this comment

Uh oh!

phpstan-bot commented Mar 31, 2026

Uh oh!

VincentLanglet Mar 31, 2026

Choose a reason for hiding this comment

Uh oh!

phpstan-bot Mar 31, 2026

Choose a reason for hiding this comment

Uh oh!

VincentLanglet left a comment

Choose a reason for hiding this comment

Uh oh!

phpstan-bot commented Apr 1, 2026

Uh oh!

VincentLanglet left a comment

Choose a reason for hiding this comment

Uh oh!

phpstan-bot commented Apr 1, 2026

Uh oh!

VincentLanglet left a comment

Choose a reason for hiding this comment

Uh oh!

phpstan-bot commented Apr 1, 2026

Uh oh!

staabm Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

phpstan-bot Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

staabm Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

phpstan-bot commented Mar 31, 2026 •

edited by VincentLanglet

Loading