Search results

Event Feast of Boris Stomach Capacity: [min(1,path(none)+path(Teetotaller))*(1-class(AWOL))*15] Looks like someone misspelled "Teetotaler".

For what it's worth, I don't think the GOAWAY fix has been backported to Java 21 yet. I think Java 22 might have it, 23 definitely does.

As others have mentioned, you need at least Java 17 to build and run Mafia. We are planning to bump that minimum version to Java 21 (the latest long-term support release) in the near future (next month or two, I'd guess), so it may make sense to get ahead of the curve. To the best of my...

I suspect that may be the case -- it sounds like we need to make a clear distinction between JS that runs in the browser, as opposed to JS that runs in Mafia's Rhino engine to execute various actions including generating assets served by the relay server (which is what most mentions of "JS" as a...

This should be fixed with r28100.

I will defer to @Veracity here, but I'll try to address some of the comments / questions. I expected both of you would have received an alert when I mentioned both of you directly in #3. https://kolmafia.us/account/preferences has an option for receiving notifications (via the bell icon in the...

https://github.com/kolmafia/kolmafia/pull/2482 seems to fix the problem on my end.

I can reproduce readily. A manual bisect indicates that this broke as a result of r28091, which finished replacing the json library that we use. @ikzann We may want to revert the change if we can't figure out a fix-forward.

Specifically, I mention the passive skill cache because recalculateAdjustments(true, ...) creates a new DebugModifiers which in turn, among other things, forcibly refreshes the passive skill cache.

I'm noting that in this test, "unbrekable umbrella" is misspelled. That doesn't seem sufficient to cause test breakage, but might be a small step in the right direction, given the latest observations. I'm also noting that Crimbo Training: Passenger Greeter does provide +1 exp in the modtrace...

Indeed, this is the newer version of nginx correlated with the GOAWAYs, as I thought it might be. That debug log is fascinating, though. It almost looks like we're running into a timeout, since it's just over an hour since you initiated the first request. This might actually be AWS ALB once...

21 brings new preview features in the form of string templates and unnamed patterns / variables, and finalizes record patterns and virtual threads (both introduced in Java 19). By upgrading from 17, we also get better Unicode handling, and a couple of other minor library changes that would...

Bumping this thread -- as far as I'm aware, we are now able to build with Java 21. Spotless is now on 6.23.3, the task errors are now resolved, we're on Gradle 8.8. Most warnings are gone (there's still a warning for lib/darrylbu/util/MenuScroller.java since Object.finalize is marked for...

The command I specifically listed should open a log, make a single (small) request, and then turn off the log, so you can run it both at the start of your session and right after encountering the problem.

I was not going to mention the China angle, but since you mention it: I don't think there's any verification needed to state that Alibaba is a Chinese-owned company headquartered in China, and the vast majority of their employees are also based in China. I am not saying by any means that the...

fastjson2 has me worried because it's hundreds of thousands of lines of Java across thousands of files over thousands of commits, and it has a bunch of features we don't need (read: broad attack surface). Based on the "projects using us" list, I also don't see much adoption by other large...

@fronobulax raised some points in the pull request that I feel warrant more discussion / visibility. @ikzann @Ryo_Sangnoir The obvious place where faster JSON might be relevant is in parsing api.php, but it's also not clear to me that JSON handling is a bottleneck at this time. It's also not...

Hm. For the next time this happens again, mind grabbing a debug log at the start of your session, then again right after it occurs? Specifically, debug on; text game.php; debug off should be enough to grab some response headers, which'll tell us which version of nginx is issuing the GOAWAY...

KoL is apparently using both nginx/1.18.0 and nginx/1.8.1, per the returned headers (the AWSALB cookie pins you to a particular backend). Perhaps they've recently started a migration, which would explain why this started happening again. (HTTP/2 support in nginx was added in 1.9.5.)...

That said, r26388 implemented the nominal "reset the HttpClient every X requests" (which was later refined in r27798). I would be a little surprised if we're running into the headers-too-big problem. I'd be more inclined to believe that something isn't actually using the ResettingHttpClient.