Sorry for the static, this was my mistake: I had a stale extensions/Wikibase/vendor, once I deleted it everything worked as expected.

This looks really good already, but I run into an error when trying to start the module locally:

Checking existence of old default messages...[6290d0b253938b7c4432853c] [no req]   TypeError: Cannot access offset of type Wikimedia\Timestamp\TimestampFormat in isset or empty
Backtrace:
from /home/adamw/wikimedia/mediawiki-core/extensions/Wikibase/vendor/wikimedia/timestamp/src/ConvertibleTimestamp.php(312)
#0 /home/adamw/wikimedia/mediawiki-core/extensions/Wikibase/vendor/wikimedia/timestamp/src/ConvertibleTimestamp.php(285): Wikimedia\Timestamp\ConvertibleTimestamp->getTimestamp(Wikimedia\Timestamp\TimestampFormat)
#1 /home/adamw/wikimedia/mediawiki-core/includes/GlobalFunctions.php(1078): Wikimedia\Timestamp\ConvertibleTimestamp::convert(Wikimedia\Timestamp\TimestampFormat, string)

I've updated mw-core and Wikibase with no change to this glitch.

Adding Wikibase module as the prerequisite to a more complete WIT local development environment.

Bug: T420384

I'm unsure whether this comment is just left over, or if it's explaining what the mw-cli equivalent looks like as documentation?

Yes thanks, this is a bigger problem that the hostname is hardcoded in many places.

You should already be able to get the port from the environment variable ${MW_DOCKER_PORT}, but it doesn't need to happen in this patch.

This line can be removed, I imagine it was stray copy-pasta from the CentralAuth module but we don't use the variable in this file.

Sean Leong (5586b92d) at 18 Mar 14:16

... and 1 more commit

Adding Wikibase module as the prerequisite to a more complete WIT local development environment.

Bug: T420384

Sean Leong (b672c354) at 18 Mar 09:42

Adam Wight (ce41d22c) at 16 Mar 07:40

Adam Wight (1b3c4329) at 13 Mar 08:36

Adam Wight (cedd0b22) at 12 Mar 22:48

Adam Wight (6ab44b94) at 12 Mar 22:47

Thiemo Kreuz (3cdf2095) at 12 Mar 18:20

Thiemo Kreuz (4631911e) at 12 Mar 18:20

... and 3 more commits

This makes it possible to observe performance of this algorithm independently from the rest of the ref calculations, and to disable it separately if we choose.

The current hypothesis is that this module accounts for the largest share of processing time, since effort increases with the square of ref count.

Thiemo Kreuz (0aad2847) at 12 Mar 10:58

... and 1 more commit

Thiemo Kreuz (bdae96ba) at 12 Mar 10:58