Something that is documented, and that works the same in every released version of PHP (and hhvm) can't really be described as unexpected, or buggy. It's surprising at the most ...

I think this needs an RFC to resolve it, so I'm going to label it as such ... at the very least it requires an internals discussion.

Agreed. I'll start a discussion on internals later today. Thanks for the feedback!

sorry guys, I didn't see this PR before I sent my one, but I have alternative approach to fix this.
#2384

I don't think it's a thing that needs fixing (although it is wtfaf). At most, I'd say deprecate this method and introduce a new one, but even that feels heavy handed given how positively ancient this function is.

@sgolemon, thanks for the feedback.

If you check the implementation, you'll see that this change wouldn't be limited to the array_fill function but to all arrays where the first key is negative. So, in the scope of this implementation, deprecating a function/introducing a new one wouldn't really make sense.

@pmmaga I am wondering how it will influence existing code which relies on previous version of array indexes?

@andrewnester, yes, it would introduce a BC break in those cases.

Looks like the latest changes broke the build completely...

It did indeed. I didn't have the time to fix this today so I added the [WIP] to the title until I get it right again.

Ok, I've managed to fix the issues and it's all green now.

This definitely should not target PHP 7.2 as it breaks backward compatibility.

'Documented but unexpected' doesn't make any sense to me?

Changing documented behaviour is a clear BC break so this should target 8.0 IMO

I'd rather see an extra param to array_fill then change the default in 8.0, FWIW

see https://wiki.php.net/rfc/releaseprocess
it's pretty clear this cannot go into 7.2

I find the current behaviour of array_fill unintuitive and quite useless. But with that said, this is a completely unnecessary BC break. Given that this behaviour has been known since PHP 4.2, I don't see why it is pressing enough to need to change it ASAP in the next minor version. So I voted -1 for it targeting PHP 7.2, but I'd be +1 on targeting the next major version instead.

I have updated the RFC to target 8.0 and updated this PR to throw a E_DEPRECATED notice on the cases where the behavior will change but keep it for now.

Following the vote on https://wiki.php.net/rfc/negative_array_index and given that the deprecation notice was rejected, I'm closing this PR.

For reference: this was implemented in PHP 8.0 as #3772.