I've followed again the test plan (but arc land) and everything works, especially "arc anoid". Arc anoid keeps all my damn infrastructure up and running 24/7 so this patch can be approved.

One last cheeky improvement: add type hint to $arguments so that it's not inferred to be never when it's conditionally checked with is_array.

Okay. Let's drop the argument :)

Added web archive links to the task, because secure isn't loading right now.

In D26037#28106, @aklapper wrote:

After applying this patch, phpstan throws a bunch of additional output here, like Class PhabricatorAuthPasswordQuery extends generic class PhabricatorCursorPagedPolicyAwareQuery but does not specify its types: R and such. Probably that is totally fine. Just saying, as a phpstan newbie.

I personally think having the arguments passed into runWorkflow makes a bit more organizational sense (in that I have a horrible functional programming streak that says use arguments for everything lmao), but it does somewhat leak an abstraction by committing more to a particular argument parsing interface, which looks to have been part of the problem with the previous approach (using an array for arguments).

Thanks (about that unit test, it never worked, see T16098)

It's already accepted but yesssss

I'm a simple person, I see PHPDoc, I approve PHPDoc. We want more. Thanks

Sorry markup

Sorry² tuples

Sorry tuples

Yeah, I like Setup Warnings. The EXPLAIN SELECT * FROM table could be a good trick to have an efficient estimation of the table size, without actually causing a full table scan (e.g. COUNT(*) is bad)

Update description/

P.S. file a task for that broken unit test on your computer :D :D ideally finding if with git bisect you discover a regression

«There's a known problem with 8.2 though» Ouch I don't remember what was this about 🤔 uhm