ternary reference assignment should be legal

[divinity76]

Description

I wanted to do

foreach((is_array($value) ? $value['images'] : $value->images) as &$image) {
// process $image
}

but that silently fail (!) because foreach will create an internal copy and iterate the copy (see `&$image);
So instead I tried:

$iter = is_array($value) ? (&$value['images']) : (&$value->images);
foreach($iter as &$image) {
// process $image
}

but that fail because ternary reference assignment is illegal (?)
So instead I have to do

if(is_array($value)){
    $iter = &$value['images'];
} else {
    $iter = &$value->images;
}
foreach($iter as &$image) {
// process $image
}

which works.
Now it feels like I'm fighting the language itself.

[iluuu1994]

Ternaries are not special in that regard. The same occurs when iterating by-ref over an array returned from a function by-value. The ref operation itself does not fail, it's just performed it on a temporary copy of the array, hence not being observable.

[divinity76]

I see, but

$iter = is_array($value) ? (&$value['images']) : (&$value->images);

fails with

Parse error: syntax error, unexpected token "&" in /in/2l3o7 on line 3

[iluuu1994]

Right, we don't have a syntax that does what you need, i.e. fetches a "var" in write context BP_VAR_W, transforms the value to a reference if it isn't already one, and then returns it. I would personally rather see less support for references rather than more, but my feeling aside, this is still a valid feature request. This would need an RFC though.

A warning for by-ref foreach when the array refers to a non-indirect, non-ref value might be possible too. Then again, it's not technically correct that it's fully side-effect free (https://3v4l.org/bnMCa#v8.4.5). Honestly, I would just leave things as is. Another warning also opens more possibility for error handler issues (#12805). But I'm open to hear what others have to say.

[bwoebi]

The primary issue is where references "decay". Currently references in ternaries decay before they're assigned. But they don't really have to.

It would be perfectly fine to have $a ? $b : $c return a reference (if the input is a reference). The only place where it actually needs to decay is when it's ultimately used for another operation.

I think that's rather a shortcoming of the language.

foreach(is_array($value) ? $value['images'] : $value->images as &$image) {
// process $image
}

should just work in my opinion. Like, if you iterate by reference, you explicitly opt in into wanting a reference to the iterated array.

I've been myself annoyed in the past by it and would have liked it to just work.

[iluuu1994]

Reference decaying is not the only problem though. Nested expressions also need to gain the understanding that they are being fetched by reference, omitting warnings, creating offsets, etc.

[bwoebi]

@iluuu1994 Yeah, ?: and ?? would essentially need to become part of the (delayed) "variable" handling in compiler, which determines which fetch_type to use. But I think it's not as complicated as you try to describe it.

[ndossche]

This very likely needs an RFC

[github-actions]

There has not been any recent activity in this feature request. It will automatically be closed in 14 days if no further action is taken. Please see https://github.com/probot/stale#is-closing-stale-issues-really-a-good-idea to understand why we auto-close stale feature requests.