Test config option rule: null -> hardcoded default, false -> disabled

[pabzm]

An unspoken (as far as I know) rule of Roundcube config options is that setting an option to null makes Roundcube use the (hardcoded) default, while setting it to false disables it.

This PR introduces a test that checks that behaviour.

[pabzm]

@johndoh I'd be interested if you have any comments, too! (Can't request a formal review from you.)

[alecpl]

I'm not sure you are not mixing two things. This is indeed how set() and get() methods work. However, loading options from a config file will override the defaults. Which also is as intended.

[pabzm]

Hm, I'm not sure we misunderstand. To be precise: by "hardcoded" I mean the second parameter to get(), not an entry in defaults.inc.php.

I just verified by testing that:

• if a config option's value is set to null in the config file, get() returns its second parameter,
• if a config option's value is set to false in the config file, get() returns false.

"The config file" being defaults.inc.php, or (overriding settings from the former) config.inc.php.

@alecpl Do you agree that this is correct and wanted behaviour?

[pabzm]

@alecpl I'd like your feedback on my last comment here

[alecpl]

One could argue that using array_key_exists() would be better than isset(), but this is how it always been so I'd keep that. So, I agree this is correct behavior.

[pabzm]

One could argue that using array_key_exists() would be better than isset()

I guess you mean this line? https://github.com/roundcube/roundcubemail/blob/master/program/lib/Roundcube/rcube_config.php#L377

That is unchanged since 0ac4160 (from 2012). But I'd be happy to change it if you confirm.

So, I agree this is correct behavior.

Then this PR could be merged in order to actually check this behaviour, I guess? (I'll do it in a few days unless you object.)