The drop aligns perfectly with the wmf.21 train (ref T281162). However, this is only a metric issue as such not blocking the train.
Description
Details
Related Objects
Event Timeline
This is fallout from https://gerrit.wikimedia.org/r/c/mediawiki/core/+/713520 (/cc @DannyS712). I forgot that we hardcoded these file paths in blameStartupRegistry.php.
Patch incoming...
Change 716090 had a related patch set uploaded (by DannyS712; author: DannyS712):
[mediawiki/extensions/WikimediaMaintenance@master] Update blameStartupRegistry for core changes
Change 716084 had a related patch set uploaded (by Krinkle; author: Krinkle):
[mediawiki/extensions/WikimediaMaintenance@master] blameStartupRegistry: Call StartupModule::getScript instead of hardcoding
Oops, I didn't see your comment about sending a patch, sorry - I sent https://gerrit.wikimedia.org/r/c/mediawiki/extensions/WikimediaMaintenance/+/716090
For people who may not have seen the commits, please note that once its fixed, we still can probably expect a drop in the startup size, just much smaller
Change 716090 abandoned by DannyS712:
[mediawiki/extensions/WikimediaMaintenance@master] Update blameStartupRegistry for core changes
Reason:
other patch was better
Change 716092 had a related patch set uploaded (by Krinkle; author: Krinkle):
[mediawiki/extensions/WikimediaMaintenance@wmf/1.37.0-wmf.21] blameStartupRegistry: Call StartupModule::getScript instead of hardcoding
Change 716084 merged by jenkins-bot:
[mediawiki/extensions/WikimediaMaintenance@master] blameStartupRegistry: Call StartupModule::getScript instead of hardcoding
Change 716092 merged by jenkins-bot:
[mediawiki/extensions/WikimediaMaintenance@wmf/1.37.0-wmf.21] blameStartupRegistry: Call StartupModule::getScript instead of hardcoding
Yep. I've been keeping a tally in a spreadsheet:
-422 B Subtotal [wmf.20 > wmf.21] (compressed/transfer sizes)
In actuality (dashboard), for enwiki which is on wmf.21 as of yesterday, the difference was only a fraction of that. The decrease was 50.06kB > 49.95 kB overall, and that's uncompressed so we'd expect a bigger difference here, not smaller.
The decrease was mostly due to NavTiming having one less bundle. startup_js actually remained equal around ~7.1 kB. I suspect this is due to my blameStartupRegistry patch changing the measurement to include a few previously-missing bytes for the embedded variables (P17134).