Use has_config() in get_config() to prevent warnings on null values

[dougaxe1]

In get_config(), use has_config() to test for key existence instead of equivalent isset() which throws a warning for null values.

A null value can occur if the option is present, but without a value. Certain config values, such as --url, default to null causing has_config() to be true, while throwing a warning in get_config(). This change provides consistency between has_config() and get_config().

Fixes #5353.

[dougaxe1]

I was unsure whether the test should go in test-configurator.php, or another file -- I don't see any tests for has/get_config currently. I am also unsure whether it is safe for downstream tests to run $runner->init_config();, please provide feedback and I can yank the test, or amend.

[danielbachhuber]

I was unsure whether the test should go in test-configurator.php, or another file -- I don't see any tests for has/get_config currently. I am also unsure whether it is safe for downstream tests to run $runner->init_config();, please provide feedback and I can yank the test, or amend.

I think what you have is fine. The tests pass 😁

I'm having a little bit of a hard time grokking the nature of the change, though. Can you spell it out for me?

Also, I'd be curious to hear how you ran across this.

There's probably technically a back compat break here, but it's probably fine to make in practice.

[dougaxe1]

Hi @danielbachhuber!

We had written a script for both single and multisite installs which allowed overriding the site URL with WP-CLI:

if ( defined( 'WP_CLI' ) && WP_CLI && WP_CLI::has_config( 'url' ) ) {
	$siteurl = WP_CLI::get_config( 'url' );
}

When tested on a single-site, we quickly saw: Warning: Unknown config option 'url'. Because for url = null, array_key_exists = true but isset = false.

After reviewing the code I felt the two functions should behave consistently. I could technically live with both using isset(), but array_key_exists() is my expectation.

In #5353, I noted that it is perhaps a different bug in that config['url'] exists on all requests regardless of the --url flag existing, but this seems like a simpler fix.

[danielbachhuber]

@dougaxe1 Thanks for the explanation! Seems reasonable 😊