Automatic live 9x9 double-elimination tournaments have McMahon bars (!?) about online-go.com

[dexonsmith]

Doing a bisection through 1991 archive pages at 25 tournaments per page:

• 1000: https://online-go.com/tournament/61206 good (no negative points, despite presence of DDKs)
• 500: https://online-go.com/tournament/87805 bad (negative points)
• 750: https://online-go.com/tournament/74347 good
• 612: skip (not seeing negative points, but not confident tournament had DDKs)
• 611: https://online-go.com/tournament/81966 good
• 550: https://online-go.com/tournament/85204 good
• 525: https://online-go.com/tournament/86478 bad
• 537: https://online-go.com/tournament/85824 bad
• 543: https://online-go.com/tournament/85557 bad
• 547: https://online-go.com/tournament/85377 bad
• 548: https://online-go.com/tournament/85338 bad
• 548: https://online-go.com/tournament/85327 bad
• 548: https://online-go.com/tournament/85308 bad
• 548: https://online-go.com/tournament/85303 bad
• 548: https://online-go.com/tournament/85279 good
• 548: https://online-go.com/tournament/85299 bad
• 548: https://online-go.com/tournament/85295 skip

That means the first "bad" one I can find was "Live 9x9 Double Elimination Tournament 2022-01-25 14:30", https://online-go.com/tournament/85299.

That's pretty close to when two backend changes were committed related to setting initial points in McMahon bars:

• f9b40a50841de263e8120c40e0a0c19a32dcc9d4 (committed January 21, 2022)
• 07b0d6eaf16d619d175852d63393b876d156c797 (committed January 26, 2022)

After staring for a while, I can't see how they triggered the bug. I actually found those changes before doing the bisection, since they were obviously touching related code, but decided they couldn't be the problem. It looks to me like the bug was always there: if a double elimination tournament somehow had McMahon bars set, they'd be used.

[anoek]

Interesting, thanks for finding and (presumably) fixing, i'll do some testing with it and get it deployed tomorrow if it all looks good.

[anoek]

ogs #1959 has been merged and deployed, new tournaments going forward should be good to go

[pdg137]

Note that you can see the negative points at this other API endpoint: https://online-go.com/termination-api/tournament/105862/players

In case it helps here's an ancient one where there don't seem to have been negative points awarded (e.g. to user stargrazerfairy who would have probably been DDK): https://online-go.com/tournament/3565 https://online-go.com/termination-api/tournament/3565/players

But it does have the bars in the settings: https://online-go.com/api/v1/tournaments/3524

More recent ones like https://online-go.com/termination-api/tournament/103270/players (2023) show negative points.

[anoek]

Huh, nothing has changed with those configurations in many many years, odd we're seeing stuff now..

[dexonsmith]

If there are old ones without negative points, then maybe some refactor related to the backend changes a few months ago triggered this.

[dexonsmith]

After staring for a while, I can't see how they triggered the bug. I actually found those changes before doing the bisection, since they were obviously touching related code, but decided they couldn't be the problem. It looks to me like the bug was always there: if a double elimination tournament somehow had McMahon bars set, they'd be used.

Just to clarify: it looks to me like the bug was always there, from looking at the code. But the bisection has convincing evidence that somehow those two commits changed the behaviour; before those commits, McMahon bars in D.E. tournaments were ignored; after those commits, they caused negative points.

[dexonsmith]

@anoek, looking at other automatic tournaments (not just "live 9x9"), I'm also seeing McMahon bars also in:

• https://online-go.com/tournament/116084 (Blitz 19x19 Elimination)
• https://online-go.com/tournament/116030 (Live 19x19 Swiss)
• https://online-go.com/tournament/115877 (Blitz 13x13 Swiss)

Combined with "Live 9x9", that's the full set of live/blitz.

For automatic correspondence (I didn't find any in the first two pages of archive, but...):

• The simultaneous round robins won't hit the issue since they don't share a code path, but it might be nice to remove bogus McMahon bars from them too if they're there.
• The simultaneous McMahons should have bars, presumably.

[dexonsmith]

Just to clarify: it looks to me like the bug was always there, from looking at the code.

Update: I figured out how f9b40a50841de263e8120c40e0a0c19a32dcc9d4 (January 21, 2022) triggered the bug. I still think it's worth cleaning up configurations to remove bars where they don't make sense, but maybe a backend change is also in order.