Netplay not synced when using start function

[1j01]

I was initializing my game inside the start callback, but this lead to a memory access error when trying to use netplay, or if I defined empty arrays at declaration like let players: Player[] = []; then the game would render but with entities missing on the client player's side. The host would see all the entities as well as inputs from the client player, but the client's game was broken.
I fixed it by moving initialization out of start and doing it at top level.
Is start just not supported yet for netplay?
This is a very cool feature, by the way! I love how simple you're trying to make it.

Here's my game for reference: https://github.com/1j01/polywogg
If you check out bcbf7aaa43 (before this workaround commit) you can see this issue.

[aduros]

Hi, thanks a lot for reporting this!

I believe start shouldn't be called for joining clients, since calling it after joining would usually clobber the game state they just received from the host. There might actually be a bug causing start to be called when it shouldn't be.

[aduros]

Cool game btw! I love Nidhogg.

[CanyonTurtle]

I ran into a desync issue and fixed it because of this thread. Thanks for documenting this!

My fix: CanyonTurtle/PING@a045f98#diff-d924ca21d81d7d5a59eb9e10d3e2689c4c58a7e28e6ecce6a064701666f5a730

My game:
https://canyonturtle.github.io/PING/

P.S. let me know what you think @aduros of the game! I'm not done with it yet but it's up there.

[JerwuQu]

I built the game from the linked commit and the issue is clearly present.
Here's the built cart.wasm + cart.wat for anyone else interested in taking a look.

To elimiate the case of start being called multiple times, I added some debug logging in the runtime, and at least that is not the issue here.

I checked a wasm2wat conversion of the WASM and there are 26 WASM globals which immediately raises red flags, but since moving init stuff from start into update fixes the issue, I don't see how that could actually affect anything.

I managed to break at the first update in the joining client and it looks like the globals are indeed different:

Not entirely sure this is the issue because the workaround works, but I wouldn't rule it out...

Tricky! Would be nice to test again if we ever fix globals for netplay (is it even possible?)