[VOTE] Undefined Variable Error Promotion View as Markdown

Le lundi 14 mars 2022, 18:18:55 CET Mark Randall a écrit :

I have started the vote for promoting undefined variable access to throw
an Error exception.

The vote will run for 2 weeks until March 28th 2022.

https://wiki.php.net/rfc/undefined_variable_error_promotion

This does not mention properties, are they concerned by the change? What about array keys?

Côme

Hello Mark and the list,

Le mar. 15 mars 2022 à 21:36, Mark Randall marandall@php.net a écrit :

Who knows, maybe PHP 9 can be the version that removes remove all
warnings from the zend engine.

This is not far from the reason why I voted "no".

I am not against the fact this warning becomes an error per se. I am just
not very fan of cherry-picking an individual kind of problem (read:
notice/warning/error/...) and changing it without a more global frame.

While it may not be a good practice. It's not that we cannot recover from
that, it is more that we don't want to. This doesn't match really the
notion of "error" which is not very much documented, apart from
https://www.php.net/manual/en/errorfunc.constants.php:

"E_ERROR: Fatal run-time errors. These indicate errors that can not be
recovered from, such as a memory allocation problem. Execution of the
script is halted."

Currently E_WARNING is documented as:
"Run-time warnings (non-fatal errors). Execution of the script is not
halted."

This describes how the core reacts, not when it is supposed to happen.

This could potentially be extended to:
"Run-time warnings (non-fatal errors) that are likely to denote bad
programming practice (like the use of something undefined) that are usually
solved with code changes. Execution of the script is not halted."

But then this wouldn't be true anymore if this RFC passes as it would
result in:
Undefined variables: Error
Undefined index: Warning
Undefined property: Warning

IMHO, this leads to something more inconsistent than what we have now.

I know other RFCs could target the other aspects. But there is no way to
vote "yes, provided that all other RFCs under the same umbrella would pass
for the same version".
This is probably a limitation of our current RFC process, but in order to
maintain as much consistency as possible, I see no other option than voting
"no" even though I am +1 on the principle.

Kind regards,
Patrick

Am 16.03.2022 um 10:00 schrieb Mark Randall marandall@php.net:

I am not against the fact this warning becomes an error per se. I am just
not very fan of cherry-picking an individual kind of problem (read:
notice/warning/error/...) and changing it without a more global frame.

I think we should try to hit all of them, ideally in time for PHP 9.

That was the premise behind: https://externals.io/message/116918

It will likely take several RFCs until we've got it in the state we want it, some of those are more contentious than others (such as undefined array indexes).

I think undefined property access should throw too, and will likely have supermajority support, but let's do it in a separate RFC.

What I don't think we want is the entire package being blocked because a particular item (i.e. array keys) is contentious.

This is actually exactly what Patrick is complaining about: No global plan.

I know I'm on the "losing" side of this discussion but sometimes it feels like one of the main arguments is "we already changed this and that to exceptions, now we have to do it in this case too". Which, to me, fits the first paragraph of https://en.wikipedia.org/wiki/Salami_slicing_tactics

Maybe we should ask ourselves the question: Why would the entire package be blocked? Just because it is too big or maybe there are subtleties which have not been properly resolved?

• Chris

Am 16.03.2022 um 11:39 schrieb Mark Randall marandall@php.net:

Maybe we should ask ourselves the question: Why would the entire package be blocked? Just because it is too big or maybe there are subtleties which have not been properly resolved?

It's politics and the practicalities of getting things done.

A nice description of salami slicing tactics ;-)

An all-or-nothing approach will inevitably end up with nothing, because those who are opposed on principle need only rally around attacking the weakest element, without the need to address any of the other parts which might individually be much stronger.

That's not to say that the weakest elements shouldn't be debated, they should.

But we shouldn't sabotage ourselves, and bring about a state of permanent project paralysis, by mandating that semi-related weaker changes are bundled with stronger ones and must be voted on as a whole.

It would just open the door to the constant use of a poison pill argument.

The problem is you are saying "the weakest should be debated but not right now, let's move forward anyway" and you are also stipulating that the changes are only semi-related.

By doing that you're dismissing Patrick's notion that they are more closely related than you acknowledge because he's saying having inconsistent handling of undefined stuff is also a WTF.

And at the same time I really hope once we start discussing undefined array indices you won't bring up consistency with undefined variables (and possibly object properties) as the reason to change it. That would feel weird given that you want to treat them as separate issues right now ;-)

• Chris