Fix the type of the options of the RedisArray constructor

[mlocati]

The options may have any type: for example, connect_timeout and read_timeout are numbers, lazy_connect and consistent are booleans, function and distributor are callables, previous is an array.

See the phpredis documentation of RedisArray:
https://github.com/phpredis/phpredis/blob/develop/arrays.md
and the phpredis stub of the constructor:
https://github.com/phpredis/phpredis/blob/develop/redis_array.stub.php

[phpstan-bot]

You've opened the pull request against the latest branch 2.3.x. PHPStan 2.3 is not going to be released for months. If your code is relevant on 2.2.x and you want it to be released sooner, please rebase your pull request and change its target to 2.2.x.

[mlocati]

please rebase your pull request and change its target to 2.2.x

Done.

[staabm]

@SanderMuller please review

[SanderMuller]

I am not a maintainer, so the merge call is @staabm's. I verified the change rather than read it, and it does what it says.

The type claim holds. The arrays.md you linked documents six different value types. previous is an array. function and distributor are callables. lazy_connect, consistent and autorehash are booleans. connect_timeout and read_timeout are doubles, retry_interval is an int, and algorithm and auth are strings. array<string, string> cannot describe that.

Base against your head, on a file that passes one option of every documented type:

base   Parameter #2 $options of class RedisArray constructor expects array<string, string>,
       array<string, bool|(Closure)|float|int|list<string>|string> given.
head   no error

One thing I would add while you are in there. The parameter is still not nullable, so the documented default is a false positive:

new RedisArray($hosts, null);

head   Parameter #2 $options of class RedisArray constructor expects array<string, mixed>, null given.

redis_array.stub.php declares __construct(string|array $name_or_hosts, ?array $options = null). Changing the entry to '?array<string, mixed>' takes that file to zero errors, and the same shape is already used two hundred lines above for Redis::copy:

'Redis::copy' => [..., 'options='=>'?array<string, mixed>'],

Your call whether that belongs here or in a separate PR.

Checked and found fine: stubs/Redis.stub declares class RedisArray {} with no constructor, so nothing there contradicts the map, and there is no other RedisArray::__construct entry.

Gate on your head: full suite green, 21440 tests. The signature map suite is green too, 2844 tests and 64036 assertions. No performance statement is needed beyond the obvious: this is a data entry read through the signature map, with no runtime path.

CI: 31 of the 32 red checks also fail on other open 2.2.x pull requests, so they are the base. The one that does not, Result cache E2E tests (result-cache-relative-path), failed on could not download checksum from .../bashunit/releases/download/0.44.0/checksum, which is infrastructure rather than your change.

[mlocati]

One thing I would add while you are in there. The parameter is still not nullable, so the documented default is a false positive:

new RedisArray($hosts, null);

Fixed, thanks!.

[SanderMuller]

Verified on 672a020ec. The repro that failed twice on the base now reports nothing. The three cases that should still be rejected still are: a non-array second argument, an array<int, string>, and an int for the first parameter.

Gate on your head is green. Full suite 21440 tests, signature map 2844 tests and 64036 assertions. No performance note beyond the obvious, since this is a data entry read through the signature map with no runtime path. There is no test in the diff because a functionMap entry has nowhere to put one, so I checked the behaviour with a file instead.

Your six red checks that other open 2.2.x pull requests do not share are the base, not you. Lint fails on Parse error: tests/PHPStan/Rules/DeadCode/data/dynamic-name-always-terminating.php:20, and Name Collision Detector on Bug15244\Foo is defined 2 times. Both files are on 2.2.x already and neither is in your one-line diff.

[staabm]

thank you!