Default assert.exception to 1

[morrisonlevi]

As discussed on the mailing list, this changes the default of assert.exception from 0 to 1, meaning exceptions will now throw by default. To restore the old value set assert.exception=0 in an INI file.

I opted to leave existing tests as written and instead just changed the --INI-- section to have the old value.

[nikic]

Shouldn't we also set warning=0?

[morrisonlevi]

I think not. Exception takes priority over warning, so it doesn't issue a warning and then throw; it only throws. This way reverting to PHP < 8 behavior is only assert.exception=0.

[nikic]

It seems pretty odd from a documentation perspective to have assert.warning=1 without actually enabling warnings. Ideally this would have been assert.mode=exception or so, because really only one of assert.bail, assert.exception and assert.warning makes sense at a time. But given things as they are, I would set it to zero.

[GrahamCampbell]

Doesn't this change need to go through the RFC process?

EDIT: It looks like something related was started here, but never completed.

[kallesommernielsen]

@GrahamCampbell, there was no object on internals so no

[GrahamCampbell]

Isn't that what the purpose of the RFC system is to establish?

[morrisonlevi]

@GrahamCampbell Do you have a specific concern about the change itself?

[GrahamCampbell]

I think discussion over the default value assert.warning is perhaps needed. The RFC process is usually good at teasing out all the fine details, since lots of expert (in different areas) eyes see it.

[morrisonlevi]

IMO we should leave it as 1 or On; this way reverting to the previous behavior is just toggling assert.exception to 0.

This area needs overhauled but nobody cared to do it in time for PHP 8.0. For example, assert_options should get formally deprecated or removed (has been discouraged in documentation for all of PHP 7) and many of these settings don't make sense to me.

[kallesommernielsen]

@GrahamCampbell well yes and no, but not everything is going through the rfc system if there is a consensus like there was with this. It is all very subjective to a case by case basis.

Edit: If we go with the RFC process for the warning part here, then that part will not make it for 8.0, something I would argue that is related to this.

If your argument for using the RFC issue is that more expert eyes see it, then perhaps those expert eyes should be reading internals actual in the first place and I put their concern there, as it’s meant for.

[nikic]

Okay, I don't care particularly strongly about the assert.warning part. This LGTM.