Your assertion that attempts to fetch more than one diff at a time will fail is demonstrably untrue. I was using bulk queries of 20 diffs at a time, and I fully understood the cache issue. The cache issue would affect the query only if one of the given diffs were not in the cache.

There was an analysis of usage. What did the analysis show for rvdiffto, which was also deprecated? It looks like that ApiAction was called 15640856 times, which would translate to 15640856 * X calls to action=compare.

You said on mediawiki-api that you would rethink if anyone had "major objections".

Also, looking through the comment history, there are already objections in threads where the deprecation was suggested:
https://phabricator.wikimedia.org/T31223
https://phabricator.wikimedia.org/T164106

The objections mirror my own.

Your assertion that attempts to fetch more than one diff at a time will fail is demonstrably untrue. I was using bulk queries of 20 diffs at a time, and I fully understood the cache issue. The cache issue would affect the query only if one of the given diffs were not in the cache.

You're right, I misread the code and when I tested it on my own wiki I ran into T216554 without realizing it.

OTOH, it's not terribly often that diffs are already cached unless you're hitting the same revisions multiple times. Why would you be doing that?

You said on mediawiki-api that you would rethink if anyone had "major objections".

I said that almost two years ago, yes. You're a bit late.

And you've yet to raise any major issues, just minor ones to which you're ascribing undue importance.

You could be right.

My concern is that I don't want to cause undue load on Wikipedia or appear to be running a denial-of-service attack.

At this moment, there are 146 edits per minute on English Wikipedia. If you honestly think that hitting the compare endpoint that many times is a minor concern, I will modify my application to work in that manner.

You had at least two objections at the time, and apparently you added the deprecation notice despite not realizing people were using it to make multiple diff queries.

I think that alone could be reason to remove the deprecation notice then. People weren't exactly scrambling to add a deprecation notice to the endpoint.

But again, I'm saying all this under the presumption that hitting compare a large number of times has a non-negligible amount of overhead compared to the endpoints that were deprecated.

@Elephanthunter: Please read https://www.mediawiki.org/wiki/Bug_management/Phabricator_etiquette and stick to technical arguments instead of getting personal in the initial version of your last comment. Thanks a lot.

The Action API on Wikimedia sites is currently handling around 407000 requests per minute (around 150000 of those just for enwiki). An extra 146 isn't likely to even be noticed.

Just follow the usual etiquette recommendation to send your requests serially rather than in parallel and you should be fine.

@Anomie Those are some honestly impressive stats. Pretty cool. It looks like the API change isn't a problem then.