SYNC ERRORS ARE REAL

[Sefidel]

"I know what a sync error looks like."

that is exactly what im talking about. In order to admit that sync errors exists, you cant assume that you know what they look like. additionally, if you look at your replay and it seems fine, that is meaningless. The only way to make the comparison is if you take multiple peoples replays and make comparisions. The only way this can be solved is if it is resolved for real.

I request intervention on behalf of all spring players. Not only does the misconception of spring sync errors need to be crushed, but the problem needs to be fixed. I volunteer to test whatever is neccassary, and im confident others will as well.

[Mav]

I volunteer to test whatever is neccassary, and im confident others will as well.

I volunteer you to play LLTAComplex with 11 other players. I've said this several times now throughout this discussion. I even said that it would be "a challenge" for you to find the "sync error".

I've done my work to disprove that a true game-altering sync error exists by spending several hours watching several replays now. You have done nothing except try and disprove my statements with no evidence. Prove me wrong. You'll find that I'm a reasonable guy and will admit my ignorance if found guilty of ignorance. I'd post my personal replays, but from what I've seen of the forums thus far, knowledge of someone's game identity leads to the perpetual flinging-of-poo as they are judged for being "noob".

Good luck. I have nothing else to add to this discussion until someone plays LLTAComplex and finds the error.

Once again, discussion of "real" sync errors verses "fake" sync errors is pointless. Fix the error and solve the problem for good.

[Sefidel]

You are the exact problem Mav, by saying that you "know" what the errors are, other people read you saying that and think they are experts at sync errors.

Maybe you can determine some errors, but the fact that you dont really know is the problem. I am sure that some sync errors are real and some are fake, thats not the problem. The problem is everyone who thinks they are an expert at it.

It is always better to assume that all sync errors are real. I played 3 games yesterday with REAL sync errors, and none with fake. I refuse to get into an argument about this.

Tomorrow ( Wednesday the 6th) , at 6 PM Pacific Time, I will be hosting a game of llta complex. please all join to test this issue. Afterwards, assuming sync errors ,I will have each of you upload your replays.

The link below will provide you with timezone conversion.

http://www.timezoneconverter.com/cgi-bin/tzc.tzc

[lurker]

Everyone should have the same replay, but it'll run differently on different computers if it desynced.

[AF]

Lets say that a solar collector was built with a location that was 0.00000001 out from the rest of the players.

This is a desync, sure the game is not affected by the change but it is a change nonetheless, it does not make the desync any less real than one where a krogoth dies in one game but not in another..

[Argh]

Lets say that a solar collector was built with a location that was 0.00000001 out from the rest of the players.

Uh, wait a minute... aren't things like that prevented by holding the accuracy of things within floating-point norms? I mean, we don't really need perfect accuracy- 4-5 significant digits (i.e., should be safe on all platforms, all compiles) should be fine. I really doubt that's a major issue, but meh, maybe desyncs are due to RNG stuff or rounding problems where on one platform the result of X - Y = 1.0000001, and on another, it's 1.00000005... and on both platforms, rounding to 1.00000 would have been safe.

[imbaczek]

what you suggest would lead to incorrect behavior rather quick. we instead rely on standarized floating point operations - the IEEE 754 standard, to be precise. the problem is that some implementations aren't really standard and give slightly different results for the same computation. note: this is not, by far, the main source of sync errors in spring.

[Argh]

Yeah, I figured you guys were following IEEE. Why would cutting significant digits be a serious problem?

[Auswaschbar]

Yeah, I figured you guys were following IEEE. Why would cutting significant digits be a serious problem?

What if its already done?

[Argh]

OK, so basically there's nowhere useful to go with that.

So, where are most of the sync errors, then? OpenGL stuff? Big events causing CPU timeout long enough that gamestates branch?

[Auswaschbar]

• uninitialised memory
• bugs in simulation which make it depend on the players system
• bugs which make synced stuff depend on unsynced

[Tobi]

Yeah, I figured you guys were following IEEE. Why would cutting significant digits be a serious problem?

Not a problem, not a solution either. There will always be numbers very close together, one which would be rounded up and one rounded down, hence you can never entirely get rid of this kind of desyncs by introducing more rounding. In other words there's not really a good reason to round more then hardware already does (since it's finite precision).

Also to add to Auswaschbar's post:

• memory corruption

So, where are most of the sync errors, then? OpenGL stuff? Big events causing CPU timeout long enough that gamestates branch?

Neither. Timeouts have nothing to do with desyncs; even if you're 10 minutes behind, engine keeps executing stuff in the correct order, so you will not desync because of this.

[Sefidel]

Dare i Say it? The most obvious solution, and probably the most difficult, A RESYNC option.

[BrainDamage]

that's not a solution, if the desync is consistent ( see programming error ) you'll desync, resync, desync .. in a loop, and all the "theoretical" resync solutions that has been discussed create a big strain on certain parts ( either network, cpu, etc ) so the game would be no less ruined

[Sefidel]

it would have dual functionality though, with a resync you could have ingame joins. Im just throwing the idea out, not likely it will happen.

[Mav]

Tomorrow ( Wednesday the 6th) , at 6 PM Pacific Time, I will be hosting a game of llta complex. please all join to test this issue. Afterwards, assuming sync errors ,I will have each of you upload your replays.

Anything ever happen with this? I've tried a few times to get games going, but the n00b factor has been high this week and I can't get people of of their damn DSD games. Seriously, last night there were FOUR 8v8 DSD games going, and that was all.

If someone can convince a group to do LLTA and upload a couple replays I'd be eternally grateful.

[JohannesH]

Its not just LLTA. Theres many replays on adune now where these happen, check bluebend, escarpment, and bb games (among others) to see fake desyncs.

Escarpment is pretty new map too, so its not just old maps but (almost?) anything with features/water.

[Mav]

My replay... HOPEFULLY more coming...

[Mav]

Sefi was supposed to attach a replay... he left early.

[imbaczek]

i specced some of the yesterday's 2v2 tournament and indeed on bluebend the game reported a sync error but apparently sim stayed in sync; hence i can, albeit reluclantly, declare this issue as confirmed.

i hope we'll manage to investigate this and figure out if it's mod lua or something else.