Should `$a instanceof $b instanceof $c` be permitted in implementation/specification?

Question

Should `$a instanceof $b instanceof $c` be permitted in implementation/specification?

TysonAndre opened this issue 7 years ago · 3 comments

$x instanceof $y instanceof $z is a nonsense expression, but is successfully parsed in PHP 5.6-7.3 (see https://3v4l.org/QWYZJ)

This contradicts what is mentioned in https://secure.php.net/manual/en/language.operators.precedence.php (the website says it's non-associative. However, other non-associative expressions do what would make sense: They cause Parse errors (e.g. 2 == 3 != 4)

Either of the following would make sense:

Mention that this is non-associative in php-langspec (to reflect php.net) (It seems like that's probably a consequence of the current CFG, so maybe nothing needs to be done)

Additionally, try to make that a syntax error in php 8.x or 7.4, to reflect the documentation on php.net: https://secure.php.net/manual/en/language.operators.precedence.php
Update the specification to allow it, and update php.net (it seems like it's currently parsed as ($x instanceof $y) instanceof $z according to ast\parse_code()) - I'm opposed to that, and it seems like the spec in this repo forbids that

I'm not quite sure what repo to file this in, or the process to request the change to php implementation's syntax, or if this has been mentioned elsewhere.

Answer 1 · 2018-12-05T10:15:39.000Z

Nice catch! I don't think that allowing this was intentional. The operator is marked non-associative in the PHP grammar, but because it's not an ordinary binary operator (with expressions on both sides), that does not end up actually mattering.

If we want to change this, then an issue should also be filed on bugs.php.net, as changes generally always go from the implementation to the spec, no the other way around. I also suspect that this is going to be ugly to fix.

Answer 2 · 2019-03-15T08:56:50.000Z

For now, I've marked instanceof as left-associative in the spec, in order to reflect current reality: 15aa452

Answer 3 · 2019-04-19T02:56:41.000Z

I think a better solution is to just note that the Zend implementation is non-conforming. This has the benefit of keeping the parse error as acceptable, while also letting other parsers choose if they want to implement the deviation for compatibility purposes.

Since there are already multiple implementations that do not have this bug, it makes more sense to go through the RFC process at this point rather than unilaterally make them in violation of the specification too.