[nikic]

Followup to #5500, which makes sure all array returns are initialized, so the default "null" value does not get returned.

cc @cmb69 @remicollet

[remicollet]

I'm not a big fan with such BC break (and hard to find one)

All people using is_array(...) will have to use count(...) which doesn't work on older version (as false is not countable).

P.S. yes BC break are possible in 8, and yes this seems a better API, btw, if we can avoid it...

[cmb69]

I'm not sure that the BC break is actually a thing in practise. I think the case that enchant_broker_describe() returns null means there is a broken setup (why having enchant without any back-end?). And even enchant_broker_list_dicts() returning null looks rather unlikely. Furthermore, the PHP manual does not even mention the null return; instead it says the function return an array, and false in case of failure (wereas the failure case is not even specified, but no back-ends/dictionaries don't look like a failure to me).

And those who are looking for PHP 7 and 8 compatibility, could test for empty() or just cast/convert to return value to bool.

So I'm slightly in favor of this change, but am fine if we don't change the behavior.

[nikic]

I would like to add why I think having array rather than ?array return here is valuable: It allows you to write code like this:

foreach (enchant_broker_list_dicts($broker) as $dict) {
    // ...
}

while right now, at least if you want to handle these special cases (I don't know if they actually happen in practice), you'd have to write something like this:

$dicts = enchant_broker_list_dicts($broker);
if ($dicts) {
    foreach ($dicts as $dict) {
        // ...
    }
}

As we have the opportunity to change this now for PHP 8, I think we should take it.

[nikic]

Anyone else have an opinion on this? @kocsismate @Girgias maybe.

[kocsismate]

I can 100% relate to Christoph's reasoning. Not having null documented as a return type + the fact that it's an unlikely case is worth the BC break in order for a good cause :)

[Girgias]

Basing on the other work we did to remove false/null returns to more sensible defaults (emtpy strings, empty arrays, error) I would agree that breaking BC here seems pretty reasonable due to the various other cases doing the same.