Valid array access on a non-empty-string yields a non-empty-string

[robchett]

Fix for #6482

Unsure if TSingleLetter provides more info than TNonEmptyString as it is technically a TNonEmptyString of size 1.
TSingleLetter isn't exposed to userland (I don't think) so should be providing more information than is available currently.

[orklah]

Thanks for the PR!

We don't have size tracking on TNonEmptyString so TSingleLetter could still provide more information.

The snippet you added in tests is not currently failing: https://psalm.dev/r/7b2a4579da

So I'm not sure what you tried to fix. In fact, it seems #6482 was fixed in an other way. The original issue was that explode could still return false with the snippet, but it doesn't anymore: https://psalm.dev/r/1ff0613ad1?phpVersion=7.1

Can you give more information about what issue you were facing?

[psalm-github-bot]

I found these snippets:

https://psalm.dev/r/7b2a4579da

<?php
                    /**
                     * @param non-empty-string $a
                     * @return non-empty-string
                     */
                    function bar (string $a): string {
                        return $a[0];
                    }

Psalm output (using commit 958eb5a):

No issues!

https://psalm.dev/r/1ff0613ad1

<?php

/** @var non-empty-string $value */
$value;

$delimiter = $value[0];

/**
 * @psalm-trace $delimiter
 */

$nonFalse = explode($delimiter, '');

/**
 * @psalm-trace $nonFalse
 */

Psalm output (using commit 958eb5a):

INFO: Trace - 12:1 - $delimiter: string

INFO: Trace - 16:4 - $nonFalse: non-empty-list<lowercase-string>

INFO: UnusedVariable - 12:1 - $nonFalse is never referenced or the value is not used

[robchett]

Honestly, I was just trying to help with an issue marked as Good first issue + Help wanted because I love Psalm and wanted to pay a little back.

When I ran the original snippet from #6482 and took the issue to be that $delimiter was a string not a non-empty-string in the trace. You are right however that the return type is correctly picked up as a non-empty-string. Because TSingleLetter is by definition a TNonEmptyString?

Here baz is reporting what looks like an error in the trace (string doesn't fulfil non-empty-string) but is passing correctly? https://psalm.dev/r/8d8391db91

Is this because TSingleLetter is traced as a string when it could be better traced as non-empty-string?

[psalm-github-bot]

I found these snippets:

https://psalm.dev/r/8d8391db91

<?php

/**
 * @param non-empty-string $a
 * @return non-empty-string
 */
function bar (string $a): string {
    /** @var non-empty-string $v */
    $v = $a[0];
    /** @psalm-trace $v */
    return $v;
}

/**
 * @param non-empty-string $a
 * @return non-empty-string
 */
function baz (string $a): string {
    $v = $a[0];
    /** @psalm-trace $v */
    return $v;
}

Psalm output (using commit 76ebf54):

INFO: Trace - 11:5 - $v: non-empty-string

INFO: Trace - 21:5 - $v: string

[orklah]

My bad, I didn't notice that issue has been fixed, I should have closed this one earlier.

Your contributions are very welcome :)

The type displaying is the one that is returned by getId:

psalm/src/Psalm/Type/Atomic/TNonEmptyString.php

Line 18 in 69658e8

return 'non-empty-string';

For TNonEmptyString, we return non-empty-string but TSingleLetter doesn't rewrite this method because it doesn't have an alias, it's only an internal type:

psalm/src/Psalm/Type/Atomic/TSingleLetter.php

Line 12 in 69658e8

}

I wish TSingleLetter would inherit TNonEmptyString but it's not currently the case (you could probably try to do that and see if CI is ok with that :) ). That way, TSingleLetter would be displayed as a non-empty-string too (but would be more precise still)

[orklah]

Can you fix the code style issue? Otherwise, seems good like that :)

[orklah]

Thanks!