[PoC] [WIP] throw_legacy_failure declare statement

[Girgias]

This is a proof of concept for letting internal/old functions be able to throw an Exception in case of failure instead of using the legacy behaviour of returning null or false.
This behaviour is controlled by the user on a per-file basis, as with strict_types, using the following declare statement: declare(throw_legacy_failure=1);

This declare statement is not meant to throw an Exception on valid false/null return values, e.g. strpos() returning false when the needle was not found in the string.

A couple of question still remain:

• The name of the declare statement
• Should there be general FailureException or more specialised exceptions per extension.
• The functions which would benefit from this feature.

[deleugpn]

Perhaps following the JSON change on 7.3 and calling it throws_on_error=1 might be worth considering.

[nikic]

I believe it's okay to convert this kind of error condition to an Error unconditionally (I've been doing this occasionally). Basically any error condition that indicates a programming error and no reasonable code should handle locally.

For this kind of directive, I think the area to focus on would be things like I/O errors.

[Girgias]

Okay, what is the best way to convert these? Using zend_throw_exception(zend_ce_error_exception, message, E_ERROR); directly instead or is there a predefined function/macro for throwing ErrorExceptions?

[strongholdmedia]

I personally think these are not strictly programmer errors, and could save us from an awful lot of boilerplate pre-validation code if e.g. the arrays are populated from some file or even a HTTP request.
But surely, anything catchable is of help.

[franzliedke]

I agree with @deleugpn that something along the lines of throws_on_error is far more self-explanatory than the throw_legacy_failure proposal.

👍 for the general idea, in any case. Thank you!

[Girgias]

I agree with @deleugpn that something along the lines of throws_on_error is far more self-explanatory than the throw_legacy_failure proposal.

👍 for the general idea, in any case. Thank you!

I'll be changing it to that when I'll start working on that again.
I'm currently focused (when I got time again) on changing programming error Warnings into proper Errors (/ValueError/TypeError) to see what the scope of this should be. :)

[Girgias]

Closes this: an improved version is located at Girgias#2