Improve isOffsetAccessible->no

[rajyan]

Feature request

https://3v4l.org/b3Vem
As we can see here, there are two types of isOffsetAccessible->no. Whether it throws a fatal error in isset/coalesce/empty ($scope->isUndefinedExpressionAllowed) or not.
I think we can improve NonexistentOffsetInArrayDimFetchRule by separating these.

Also, a lot of people are not aware why StrictMixedType throws an error with the code below, maybe we can add some information in the document

/** @var mixed $mixed */
isset($mixed['foo']);

Goal

https://phpstan.org/r/1087a148-ae5b-477f-9003-215fefc3437f
No error in line 7, 21
Throw error in 12

Design

• Add method something like isOffsetAccessThrowsFatalError(): TernaryLogic (want a better naming...) to Types
• Return TernaryLogic::Yes for isOffsetAccessThrowsFatalError in MixedType/StrictMixedType if object type is subtracted

Related issues

These issues should be closed after documented

• recursive traversal of data: array|mixed (level=9) #7229
• Cannot access offset on mixed when using isset() #8388
• Null coalesce operator on nested array returns unexpected error #8069

The test should be reverted after this is implemented

• https://github.com/phpstan/phpstan-src/pull/1791/files#r1025871491

Array access with json_decode mixed problems

• Specifiy json_decode default return type phpstan-src#1283
• [Types] Add JsonDecodeDynamicReturnTypeExtension phpstan-nette#89

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

PHPStan on PHPStan finds error while developing PHPStan 😄

[cs278]

Recently stumbled upon the issue described in #8388. Which either needs me to ignore the error or chain a bunch of isset() and is_array() calls together.

My understanding is that objects not implementing ArrayAccess cause fatal errors even inside an isset() (the same also applies to null coalesce): https://3v4l.org/QgpEg

Which is a fun PHP "feature" I'll be honest I'd totally forgotten about, so it's great that PHPStan provides protection from this particular footgun. Are there any other cases I'm not aware of?

However, PHPStan is not consistent in doing this which IMO is worse than the PHP footgun, see these examples: https://phpstan.org/r/14363150-f65d-4748-a9df-1bf733c20cf6

So I'd definitely support a change to add a separate check to determine if a fatal error is possible and making this behaviour consistent, perhaps isOffsetAccessLegal().

Is it possible in the meantime to make PHPStan's behaviour consistent?

[rajyan]

In your example, 'object' is too wide and PHPStan can only say it's isOffsetAccessible->maybe (which wouldn't throw an error in isset nor coalesce).

FYI PHPStan can handle ArrayAccess objects correctly, if the class is "final"
https://phpstan.org/r/cf338470-beef-4d50-b57a-5f44bccd002c

The difficult part with the current implementation is that NonArrayObject and bool are both isOffsetAccessible->no and treated the same way, but in isset or coalesce this behavior is different. The purpose of this feature request is to separate these ->no
the idea is from #8388 (comment)