Commands get sent to the wrong bot?

7:01 AM] liviuc: I just noticed some funky behavior, not sure if it's related to a bug or not. Here goes:
[7:02 AM] liviuc: if your bot initializes itself using a different player name than your GitHub acc (e.g. my code actually names the bot "z0d" instead of my "liviuchircu" GitHub id), the Halite platform will only tolerate this in some 90% of your matches(edited)
[7:04 AM] liviuc: In the remaining 10%, the bot kit will misbehave, apparently causing the bot to actually play some other players' ships(edited)
[7:05 AM] liviuc: This happens when your username's index matches the bot's fake name index in some "sorted by name" player array. This is just an assumption, based on lots of replays and analysis
[7:08 AM] liviuc: Some bad replays (say ~10% chance, both "liviu" and "z0d" would be the 4th player sorted by name)
---------------
https://halite.io/play/?game_id=2413897&replay_class=0&replay_name=replay-20171108-112733%2B0000--4084977309-264-176-1510140355
https://halite.io/play/?game_id=2413826&replay_class=0&replay_name=replay-20171108-112528%2B0000--4003060868-312-208-1510140273
https://halite.io/play/?game_id=2411370&replay_class=0&replay_name=replay-20171108-101133%2B0000--3747793101-336-224-1510135723

Some working replays (90% average case, "liviu" and "z0d" indexes differ in the name-sorted player array, and the platform tolerates me providing the "z0d" name)
--------
https://halite.io/play/?game_id=2409653&replay_class=0&replay_name=replay-20171108-091633%2B0000--549507420-312-208-1510132525
https://halite.io/play/?game_id=2408420&replay_class=0&replay_name=replay-20171108-084100%2B0000--2764482809-240-160-1510130445(edited)
[8:15 AM] lidavidm: @liviuc I will file an issue to track this, thanks for reporting. So just to clarify, there is a chance that you accidentally control another player's ships?
[8:16 AM] liviuc: I cannot tell for sure this is what's happening, but the bot behavior is completely insane, so this has to be it
[8:17 AM] liviuc: When I get home, I will re-upload it with the GitHub id, and will let you know if it solves the problem
[8:17 AM] liviuc: So let's just keep the issue open for now, and I will update it
[8:17 AM] lidavidm: Ok. And what you're seeing in those replays is your bot is unexpectedly behaving completely insane?
[8:17 AM] liviuc: Exactly, all I'm doing is navigate(), and it skips turns, and rams into random directions / planets
[8:18 AM] lidavidm: That is indeed weird, especially as the engine only uses the player ID for moves, and the environment actually overrides the name you provide, hmm

[liviuchircu]

Thanks, @lidavidm. I made a mistake where my per-turn self-monitoring for timeouts actually included the blocking reading of the result of game.update_map(). So any time the 2/4 bots took more than 2s/turn combined, my game state would be empty (no ships, no planets, no players!), since the update_map() call would get brutally interrupted by an OS SIGALARM. My command queue would remain untouched from the previous turn, and I would send it again to the game engine (hence why the strange crashes into planets).

The ML bot was good at reproducing this, since it likes to … take its time!

Fun fact: looking back, I now understand why this bug was always happening when I was playing against Java bots and their massive arsenal of libraries XD

Thank you, Liviu

[lidavidm]

I do not want to lead you astray, so it is not anything about the ML bot in particular, just that it happens to reproduce the issue reliably.