#perf_prof_branch

1 messages Β· Page 24 of 1

heady plinth
#

yeah. pure guess ofc, but it does seem oddly related

whole cloud
#

Ah I didn't get to that channel yet

latent sigil
#

It doesn't trigger for me. I am pressing shift + (num - ) and typing sframe, nothing happens. My fps is actually above 30 fps but there are instances of those micro stutters.

whole cloud
#

need to run profiling binary, not performance (the default one)

raven delta
#

Ran stable and had several players revert back to it (I stayed on performance)

Some people still had bad desync, one person was about 30-60 seconds in the past!
Is this just unfortunate timing where 3 players in different countries have suddenly had their internet connections become unreliable, or is it an arma issue?

cedar turret
#

I've spotted a couple of blocking bugs in Windows 7 for version v2.20.152984. I think the main one is that the new version tries to use a Steam API that no longer works with the W7 Steam client (which has been unsupported since last year, although it still works). Up to v2.19 everything worked fine regarding Steam. So basically now I see no servers, no multiplayer, and no other Steam-related functions. Manually connecting to an official server by IP, it seems to connect but blackscreens for ever. I've registered both bugs in the feedback tracker, with codes T192972 and T192977. For the first one the key log lines are the launcher.log file showing [ERROR] 5: SteamLayer: SteamAPI_Init failed. Unable to initialize SteamWorksSDK. and the game .rpt file which is littered with Unable to initialize Steam API. Seems to be the way the game talks to Steam in the new version, but it should work as W7 is still supported as per the changelog.... I wonder If in the meantime I could download the beta for v2.19 in another folder and patch the launcher executable in the v2.20 folder?

whole cloud
#

I had a big event on saturday, no issues at all 🀷

whole cloud
#

Also please use line breaks, that text blob is hard to read

cedar turret
#

Well that's interesting. The v2.19 launcher worked fine in April. Now the v2.20 launcher shows "Update unavailable, Steam is not running" after fetching the news. Sorry about the wall of text.

whole cloud
#

If the new Steam API is not supported by the steam client, then there's nothing I can do to fix that.
Only way is you upgrading to a steam client that does support it

cedar turret
#

But theoretically v2.19 already used the Steam API, didn't it? Maybe it's another little thing that prevents the game from communicating to Steam.πŸ€”

whole cloud
#

yes

cedar turret
#

Here is the launcher log for the v2.19 version:

Launcher.log:2025-04-20 22:58:21,948 [ INFO]   1: LauncherViewModel: - Initializing connection to Steam client...
Launcher.log:2025-04-20 22:58:22,036 [ERROR]   5: SteamLayer: d:\Bis\Source\Stable\Futura\El\Interfaces\iEval.hpp(38) : Assertion failed 'n == 0'
Launcher.log:2025-04-20 22:58:22,057 [ INFO]   5: Steam.Wrap: SteamAPIWrap::Init
Launcher.log:2025-04-20 22:58:22,560 [ INFO]   5: Steam.Wrap: SteamAPIWrap::InitServices
Launcher.log:2025-04-20 22:58:22,561 [ INFO]   5: Launcher.Steam.LauncherSteamFacade: Steam init result: Online

And here the v2.20 version:

Launcher.log:2025-06-29 00:50:20,875 [ INFO]   1: LauncherViewModel: - Initializing connection to Steam client...
Launcher.log:2025-06-29 00:50:20,994 [ERROR]  13: SteamLayer: d:\Bis\Source\Stable\Futura\El\Interfaces\iEval.hpp(38) : Assertion failed 'n == 0'
Launcher.log:2025-06-29 00:50:21,029 [ INFO]  13: Steam.Wrap: SteamAPIWrap::Init
Launcher.log:2025-06-29 00:50:21,112 [ERROR]  13: SteamLayer: SteamAPI_Init failed. Unable to initialize SteamWorksSDK.
Launcher.log:2025-06-29 00:50:21,113 [ INFO]  13: Launcher.Steam.LauncherSteamFacade: Steam init result: APIInitFailed
Launcher.log:2025-06-29 00:50:21,271 [ INFO]  13: SteamDlcManager: Steam is offline
empty goblet
cedar turret
#

I will see if I can update the client as well (although I surely had it frozen for a reason).

empty goblet
#

it's bit more complicated as you need keep the old STEAM web client from around jan-april 2024
so \steam\bin\cef\cef.win7x64from the older

#

but i didn't investigate much more after getting it to work on vm w7 snapshot and old PC ...

latent sigil
cedar turret
#

Would a Steam connection problem cause the other problem of not being able to enter driver or gunner as well? That seems to happen in pretty much every mission (I tried a couple in single player).

whole cloud
#

But its weird that the game launches at all, if it cannot connect to steam

cedar turret
#

There is a Steam, just maybe not the kind of Steam it expects πŸ™„

empty goblet
# whole cloud But its weird that the game launches at all, if it cannot connect to steam

i suspect it's just the libraries used by the game are too new for the too old version of steam client
too old version of steam client == steam verifies account login / game ownership, game starts (singleplayer) yet it fails to work with auth on server join (multiplayer)
i actually tested it , ie downloaded n versions of steam client until it MP worked on legacy OS

whole cloud
#

I know that, and you know that I know because I told you that before..
I don't get what this reply is supposed to accomplish.

uncut wigeon
#

I talked you before about those I know what does. (and to not have those on at all with the perf branch) the names of what they are... are pretty clear.
I was mainly inquiring into any known visual issues like I was having if it was known or something new it wasnt always that way couple months ago, for me this is recently new...

sharp hornet
#

@whole cloud I’ve been having a unique issue since the last update. On the profiling build when I get into an aircraft with a helmet mounted display, frames drop to <1 FPS upon engine startup. This is only occurring in profiling, not dev or stable builds, and only in first person. Has anybody else had this issue?

whole cloud
sharp hornet
uncut wigeon
#

actually would be a good thing to get pinned to the top cuz discord search sucks for finding on how to do that profiling capture.

whole cloud
uncut wigeon
heavy vortex
#

Holy shit, that page is a lot bigger than it used to be

outer moat
floral socket
#

i have a problem on the profiling build with seemingly random crahses, i dont know if this is the right channel but this is what my rpt says:
Fault time: 2025/06/29 09:24:10
Fault address: 5DCEEC56 01:0003DC56ll C:\WINDOWS\system32\XAudio2_7.dll
file: WarlordsRedux2 (__CUR_MP)
world: Malden
Prev. code bytes: 0F 83 8A 00 00 00 0F 1F 40 00 41 0F BF 4C 54 FC
Fault code bytes: 41 0F BF 04 54 49 83 C0 02 2B C1 66 0F 6E C1 41

note: Minidump has been generated into the file C:\Users\Gigachad\AppData\Local\Arma 3\Arma3_x64_2025-06-29_07-19-13.mdmp

light cargo
light cargo
light cargo
light cargo
light cargo
spiral pond
glass lagoon
patent sky
fiery crescent
#

Having a sporadic dedicated server crash on 3 separate hardware. Happens on the latest prof and latest Main ver. I sent dump zip to Dedmen. Does the auto crash reporting working on server side or client only or both?

patent sky
fiery crescent
spiral pond
#

is the AR crash reporter fully custom, or uses some 3rd party tool to collect the reports (ie sentry), or is it some Windows service again?

fickle geyser
#

internal tool.

fiery crescent
glass lagoon
#

thought that makes sense

whole cloud
#

And there you are wrong sir :harold:
Please write more messages. Its fun to read them

2025-07-01_17-55-53.png
knotty wraith
#

why does a player from germany have 14 mbps traffic with the server? if other players have about 1.2 mbps on average

photo_2025-07-01_20-32-20.jpg photo_2025-07-01_20-32-25.jpg
#

the second player from Germany - standard traffic

photo_2025-07-01_20-37-36.jpg
thin wyvern
#

ADSL/VDLS vs optic

knotty wraith
thin wyvern
#

Ask them, RU player/s, if they have optic or cable

knotty wraith
uncut wigeon
#

cuz maybe they flat out on the backbone of where the server is or even the hosts local's network....

light cargo
#

would it be possible to make crashreporter automatic for clients?

fiery crescent
thin wyvern
#

I play last 5 years on RU server...Ural66

inland dew
#

offtopic

uncut wigeon
#

just a single gateway inbetween you throttling would do it too... also prob better off taking the convo into #hardware_vs_arma

knotty wraith
light cargo
uncut wigeon
#

actually its shown on all clients on the player list always has.

inland dew
uncut wigeon
#

or maybe #arma3_config isnt a bug thats purely a client<>server quality issue if everything is default

inland dew
heady plinth
#

there is no speed limit in germany

uncut wigeon
#

you should see what its like when you running the server on the same rig as your client ^_^ but again offtopic. Im dropping off this now.

inland dew
knotty wraith
#

we checked - this is the speed for everyone who entered the server, then it gradually drops to 1 Mbit per second

uncut wigeon
#

again offtopic ... its not specific to this branch. and how to adjust that would be best to discuss in arm3_config.

thin wyvern
#

Now you just have better information...there's no problem with perf/prof exe πŸ‘‹

analog acorn
uncut wigeon
analog acorn
inland dew
#

guys, just let it be.
enough clutter already for nothing perf/prof related

empty goblet
whole cloud
# latent sigil

HUH!
Very curious.

There should be a parallel for, running jobs over all your threads.
But instead of doing that, it does... nothing at all for 19ms, and then... does all the jobs in only the main thread.
Wha the kaboozle indeed.

Can you send me a RPT too to go along with this data?

whole cloud
whole cloud
whole cloud
whole cloud
whole cloud
whole cloud
whole cloud
whole cloud
outer moat
floral socket
fiery crescent
glass lagoon
#

however, there is sandboxing and whole a lot so in my case there could be other causes rather than game's implementation, but for the record the steam perf branch might not be detected by your servers on clients' behalf, in specific conditions probably

empty goblet
patent sky
uncut wigeon
#

the profiling UI? see pinned msg.

sharp hornet
#

@whole cloud Getting home from work tomorrow, do you still need more data for the Helo HMD issue? Anxious to get back to profiling, vanilla gives me an unrecoverable black screen 60-70% of the time when I try to boot, have to restart my PC.

uncut wigeon
#

Does BattlEye mess with running the profiling UI? because I can not even find any diag_* functions at all and trying to do anything from the wiki page to open it keeps giving me a script error.

cloud sky
uncut wigeon
#

yeah Im using that installed via steam branch and it says experimental profiling branch 2.20..... so Im like totally stumped

#

like main screen bottom right showing the icon and version

#

so its def running the right one and installed.

image.png
#

so Im totally stumped. following the wiki page it just does not see the function. BE on/off regardless.

image.png
charred pagoda
uncut wigeon
#

no that wasnt the issue steam should renamed it but I figured why it didnt cuz some reason I had a dead arma3_64.exe still running in the BG so steam could not select it... talk about one oddball quirk

#

literally just noticed in task manager I had one running still when I wasnt IG

#

and even oddly enough it flat out was saying its the profiling version when it wasnt.

charred pagoda
uncut wigeon
#

sorry its 3am I meant from the launcher when ya pick which one to run

charred pagoda
uncut wigeon
#

Finally, use the dropdown list to select the branch profiling - Performance Profiling Build

#

nvm again its late that was a combo errors. but I figured it out.

charred pagoda
uncut wigeon
#

still looking at the photo its clearly saying its running profiling.

charred pagoda
#

Profiling exe is arma3profiling_x64.exe

uncut wigeon
#

welp now its been renamed.

charred pagoda
uncut wigeon
#

I figured it out but do appreciate the help.

charred pagoda
#

Ok, renaming it to arma3.exe makes sense

uncut wigeon
#

guessing then to use the launcher easily ya just then use the pulldown and pick the 32 bit version and that makes it simple to keep'm seperated but still use the launcher to run it if needed

#

ie arma3.exe vs arma3_64.exe

#

also just exited and again the exe didnt shut down still showing up in task manager brb gonna reboot my rig.

whole cloud
uncut wigeon
#

Dedmen not gonna ping ya over this but is it normal since I started using the profiling branch that my arma executable is not closing and shutting down even after leaving the game like currently Im looking at two of them running in task manager.

whole cloud
#

Can you do a right-click and create memory dump and send me that? I can look why its not closing

uncut wigeon
#

gonna be a min I just started a verify on my install just to eliminate that possiblity

uncut wigeon
#

just got done. launching now want it sent in a DM?

whole cloud
#

yes, pack in zip or 7z

inland dew
#

I had also 7 players wondering about black screen out of nowhere yesterday, just like that, at random moments (no alt+tab, no full screen)

#

don't think they were on perf/prof, since that were players that just bought Arma and our server was the 1st they've connected to

#

our server also isn't on perf/prof

uncut wigeon
#

Ugh. file size exceeds limits...

inland dew
#

so they just terminated arma process and I couldn't get anything useful from them, since they knew barely anything to begin with

inland dew
uncut wigeon
#

again thanks for the help and taking the time, its time for me to put the brain in sleep mode.

inland dew
#

alt+f4 = arma closes, but exe continues to run πŸ€”
current perf

patent sky
tribal pasture
#

Gotta mine them coins

restive turtle
#

When u close arma, steam continues to detect it running lol

#

(Cuz exe is running(

inland dew
#

no

restive turtle
#

Its been a thing for me personally for atleast 4+ years

#

Also yes

#

200% I've had this happening for years lol

fickle geyser
#

at least when I work on a gamemode for SGD.

#

I think I've sent lke 10+ crash reports last weekend legallydistinctkekw

empty goblet
whole cloud
inland dew
#

i was just playing (online), perf, not prof

#

then I had to go and pressed alt+f4

#

as usual

#

actually, I tried again and arma exe always continues to run, each time I alt+f4, although arma itself closes, as expected

inland dew
#

ok, when I launch arma and alt+f4 from its main menu, arma exe is insta gone
when i connect to a server, but then alt+f4, without even spawning a single time, arma exe takes several seconds to disappear from task manager

#

arma exe was still running, after alt+f4, is after i spent several hrs playing online

#

ok, I see.
spent several minutes flying in a jet over altis (online) and then alt+f4.
it took some time for arma exe to vanish from task manager

#

it's like the longer you're in mp, the longer it takes for arma exe to vanish from task manager πŸ€”

cloud sky
#

Deallocating memory or something?

inland dew
#

exiting the server and then the game (no alt+f4), feels ~same as alt+f4

unreal sparrow
#

I mean, given that A3 blocks alt-f4 when the main menu is not open it would not surprise me if it just exits the same way in both circumstances.

silk pewter
#

when the main menu is not open
(when no menu is open*)

unreal sparrow
#

Ye

empty goblet
#

from dev branch Added: Multithreaded loading of BISIGNs already in profiling or will be ?

light cargo
restive turtle
#

Steam still detects it running

#

But its literally always been like that for me

hoary jolt
#

There appears to STILL be a syncing issue between perf on client vs stable branch on server for smoke and fire particles not showing up. I am using perf builds conncted to stable server. In the escape mission, smoke and fire particles do not show or takes like a while to be "enabled". I can see the fire glow and destroyed vehicle husks, but cannot see smoke or fire client side. Stable to stable no issue. It appears to happen MORE with RHS loaded but it happens in SOGPF Escape missions with no other mods as well.

tawdry gazelle
whole cloud
orchid mesa
#

Have had a few comments from friends running the current prof build, apparently going first person in RHS jets is causing significant stuttering

#

possibly something to do with the jet targeting huds and the recent multithreading changes to MFD target filtering?

whole cloud
#

yes, others reported it too

#

Haven't quite figured out why yet

orchid mesa
#

rgr

fiery crescent
whole cloud
heady plinth
#

i see dedmen is stressed today

#

hope there is enough ice in amsterdam office

patent sky
#

No ice, AC tho

fiery crescent
whole cloud
fiery crescent
whole cloud
#

!ban 138342923312824320 0 software piracy

solemn tartanBOT
whole cloud
#

Goodbye sir. Don't ever try to waste my time again. If you want to play our games then you gotta pay for them. For fucks sake

heady plinth
#

what a twist

quaint flame
#

That escalated quickly

charred pagoda
#

did he send own logs from non-legal copy of the game? πŸ’€

ripe willow
#

I just dont get it why pirate a game with this price:
If you go on street and look for loose change you can find more money then this.

Snimka_zaslona_2025-07-03_133510.png
silk pewter
#

(even working can bring this much money, so I heard!)

tawdry gazelle
#

Just being curious: how do you tell it's pirated through log? πŸ€”

silk pewter
#

(the CRC of the logs did not match Dedmen's birthday)

autumn timber
empty goblet
empty goblet
whole cloud
#

The annoying thing is that I cannot repro the problem at all :/

Arma3RetailProfile_DX11_x64_VLRRowklxp.png
#

The one capture I've seen so far, looked like the OS didn't have enough threads. And when waking up the worker threads to start the parallel execution, it somehow lead to the main thread being suspended.
Probably the main thread was suspended, so the OS could run the worker thread on the same CPU core.
The worker thread then started spinning at 100% load, and didn't let go of the core until the quantum expired.

#

But with proper cpuCount set, that would only happen if otherwise the system is very busy and other processes take up the other cores.
Which.. does seem a bit of a problem indeed.

(proper cpuCount in this case also means -enableHT off, because having it on, could cause exactly this)

#

The problem is that we spin up the workers before the actual work is ready to be worked....
Turning that off is easy but also means we loose valuable time on main thread, because the worker start will be delayed.

I could just spin up only half the workers, that way we immediately get some multithreaded work, but don't bog down the whole system. I shall try that

#

Also instead of spinning forever waiting for work, I shall add a time limit.
If someone told them to spin up for upcoming work, but the work doesn't come, they should just shut back down.

That should turn this ~20 millisecond freeze down to a couple microseconds

autumn timber
whole cloud
#

MFD drawing in general could use a bit of work πŸ˜„
We have multithreaded rendering, but its not used there. But thats so much effort πŸ’€

Arma3RetailProfile_DX11_x64_jyCFfAIX6I.png
orchid mesa
#

is cpuCount necessary if using cpuAffinity?

whole cloud
#

no, cpuAffinity also sets the count

orchid mesa
#

gotcha

whole cloud
#

In the last months/half year there have been quite a few crashes where objects had no shape/model, I wonder where thats coming from.

Like theres a soldier crew member in a vehicle, and he has no model, I don't even know how thats possible πŸ€”

is that a engine messup, or are some mods somehow doing that.
The first big one was some sign object in cup, that was actually missconfigured and really did not have a shape.
And it ripped up every code it came in contact with

patent sky
#

Maybe culled due to being in a vehicle(not visible)?

whole cloud
#

Not just visible shape. No shape at all, not even loadable

vivid rune
#

Something like the "pilot" in the drone?

whole cloud
#

Don't think so, I think they do have a model, its just a invisible one with nothing in it

#

Like if you set model="" or a path to a non-existent p3d in the objects config.
That would be one without shape

whole cloud
#

Game crashes when starting without mods, fix coming...

analog acorn
#

issues caused by not having mods instead of the opposite is a new one

whole cloud
#

Next crash I'm looking at, a physx smoke grenade, without a model. Why does this happen so much. Or are these all modders working on their broken mods

#

Scifi mods loaded, that must be it πŸ˜„ Probably just broken mod

heady plinth
#

i blame addonbuilder

whole cloud
#

ajo. I guess broken mod then

notepad_kL69w2pkq6.png
restive turtle
visual tapir
#

oof

#

I'll get perf branch running today again

vivid rune
#

Or an attempt for being fps friendly.

whole cloud
#

We reverted it, we'll repush in about 5070min

inland dew
#

barely touching any bush/fence, in armored vehicle -> you insta die...

whole cloud
#

πŸ€” Uh, you mean in the last update? the one that we instantly pulled back?

#

Ah yes, thats correct βœ…

patent sky
#

Did i revert to the wrong one? πŸ˜‚

whole cloud
#

Good that we pulled that back πŸ˜„ It was supposed to be 026, but it actually built 025 and I didn't check

#

025 had the breaking it again, and 026 had fixing it again

inland dew
#

the one that Steam downloaded

#

025

#

is 026 out for download already?

whole cloud
#

no

patent sky
#

soonβ„’

heady plinth
#

perf branch, the new dev branch

void badger
#

Read up, dedmen reverted

spiral pond
patent sky
#

V6 now on steam :D

whole cloud
rain moth
#
  • Added: -keysFolder= command line parameter
    😍
empty goblet
#

i do wonder, could crash reporter collect some info / snapshot about frozen process ? like if manually invoked CrashReporter.exe -frozen %PID% , unless beter to make freezereporter.exe 😁

light cargo
empty goblet
light cargo
#

ah ok

autumn timber
#

What would be the use-case for that? You'd want to have an external keys directory, outside of the game directory (because it's dockerized without external mounts) or something else?

analog acorn
#

Handy for running multiple servers on the same key configuration and managing them all at the same time, or for quickswapping a server between key configurations by just changing which folder it looks at

light cargo
#

if only there was a cmdline for allowing the mods' shipped keys folder

#

that would remove managing keys on a server because it'd be dynamically loaded based on modlist

analog acorn
#

(as long as the server and client are running exactly the same mods)

light cargo
#

yes sure of course

#

the idea is to make keys work out of the box without needing to go digging each mod and stuff

#

qol to facilitate hosting signed servers

empty goblet
#

1.54.133998 profiling & performance v10 server and client, windows, linux

Dropbox
Arma3

Shared with Dropbox

whole cloud
whole cloud
whole cloud
#

Small bit more effort, but a lot easier now

empty goblet
whole cloud
#

-keysFolder=@modfolder/keys;@othermodfolder/keys; this should work

empty goblet
# tawdry gazelle mimalloc has updated to v2.2.4 and fixed several memory release problems. Any fe...

some might missed it, there is option to define params via environmental variable for mimalloc set MIMALLOC_ALLOW_LARGE_OS_PAGES=1 set MIMALLOC_ARENA_EAGER_COMMIT=1 set MIMALLOC_EAGER_COMMIT_DELAY=1 set MIMALLOC_PURGE_DECOMMITS=1 set MIMALLOC_PURGE_DELAY=100 set MIMALLOC_RESERVE_HUGE_OS_PAGES=8 set MIMALLOC_SHOW_ERRORS=1 set MIMALLOC_SHOW_STATS=1 set those in batch before the start of engine binary (client or server) ,
MIMALLOC_RESERVE_HUGE_OS_PAGES shall work as older locked pages library, in GB
if you have plenty of memory feel free to toss like 16,24,32 there ...
so those who like to benchmark for more FPS, here you have something to play with
atm. not sure about the errors / stats part
note: the documentation https://microsoft.github.io/mimalloc/environment.html

rain moth
charred pagoda
#

2.2.4 still not whitelisted :sad:
-# p.s. is it ok that we discuss mallocs / performance improvements in this channel?

#

-missionsFolder=path/to/tvt
-missionsFolder=path/to/pve
for multi-instance servers will be amazing

light cargo
rain moth
#

I’m tempted to pull stats from my steam workshop metadata mirror πŸ˜„

heavy vortex
#

A lot of hosts automatically pull keys, so there must be a fair amount of consistency.

knotty wraith
#

there is an old problem - you create a local weaponbox... but at the same time the global wehicle simply moves it when it is run over, sometimes at the expense of the physicist quite decently

fast hornet
#

what even is the reason to not load bikeys by default? I can't really think of a scenario where the server would load a mod but then not allow clients with that mod to connect?

#

since most mods come with the bikey included anyway that would make things easier i guess

cloud sky
heavy vortex
empty goblet
light cargo
uncut linden
#

Thank you πŸ™ That made things better for my server

storm crag
sullen crow
#

@whole cloud How many cores the new update can utilize on a dedicated server? Currently we're seeing full load on the 6 cores and 12 threads on Ryzen 3600 (with HC), alot of AI on the mission. Asking because we're considering what CPU to upgrade to.

floral socket
# whole cloud
  • Fixed: Possible crash when switching in/out of thermal imaging
    is this out yet? i have this issue
heavy vortex
#

I think that update got hotfixed because that change is in 153034 now.

#

153034 is current perf branch on steam.

floral socket
#

yeah im on 153034 but i still have that issue

hallow yew
#

These the debug traps you mention?

21:54:33 Streaming anim a3\anims_f\data\anim\sdr\mov\knl\wlk\ras\rfl\amovpknlmwlksraswrfldfr.rtm
21:54:33 Anim stream offset 0 = 2814

heavy vortex
#

dedmen is probably gonna want your mdmps then.

floral socket
heavy vortex
#

well, that's a problem

#

Do you have a reliable replication case?

floral socket
#

note that it only happens on vehicle thermals (drones and turrets), the ENVG's are fine

#

and only started happening after july 03 update on profiling build, had no problems with it before

#

i will see if the guy i talk to who also has this issue has .mdmps, but either way for both of us the .rpt does not show anything in the last line, sometimes being delayed by a few minutes from the crash

floral socket
#

what ive done: verified game files, the other guy verified, reinstalled arma, verified again
ran the game without mods
the mods i had installed before was JSRS Soundmod 2020
have not been able to replicate it in editor, but ive only tried a few times and that was in the virtual reality map with no units spawned

heavy vortex
#

Does the time of the last message there match up with the crash?

floral socket
floral socket
#

i attached the .rpt of the crash in each message

spiral pond
#

notable things

  • no crash info in rpt
  • startup params: -exThreads=7 -enableHT -malloc=system -hugePages -enableHT -noPause -noPauseAudio
  • bunch of Client: Unhandled user message Type_402
  • bunch of Client: Unhandled user message Type_395
  • ...: Unknown action ListRightVehicleDisplay/ListLeftVehicleDisplay..
  • Ref to nonnetwork object 829127: shell_tracer_yellow.p3d Sh_120mm_APFSDS_Tracer_Yellow // so probably not scripted if happens with vanilla
  • Observer B Alpha 1-1:1 (derangriffsteiner) REMOTE in cargo of B Alpha 1-1:1 (derangriffsteiner) REMOTE; message was repeated in last 60 sec: 9666
  • Road not found - (7077.414551, 91.761566, 7302.433594) // noticed this too in our logs. is that a new log? i can only recall it with ``Road not found ""` which still exists also - both would benefit of more info πŸ™
  • Sound: Error: File: a3\sounds_f_orange\vehicles\air\uav_01\uav_01_land_hit_04.wss not found !!! - from JSRS or vanilla A3?
  • Sound: Error: File: .ogg not found !!! // certainly from JSRS but still interesting - soundX = ""; entry?
#

@floral socket this was locally hosted MP?

empty goblet
floral socket
floral socket
weak panther
empty goblet
floral socket
# spiral pond notable things - no crash info in rpt - startup params: `-exThreads=7 -enableHT ...
  • ...: Unknown action ListRightVehicleDisplay/ListLeftVehicleDisplay..
    either warlords specific or the gps / info right and left tabs while in a vehicle, im guessing
  • Ref to nonnetwork object 829127: shell_tracer_yellow.p3d Sh_120mm_APFSDS_Tracer_Yellow // so probably not scripted if happens with vanilla
    happened right after i shot with a 120mm as in the video, desync?
  • Observer B Alpha 1-1:1 (derangriffsteiner) REMOTE in cargo of B Alpha 1-1:1 (derangriffsteiner) REMOTE; message was repeated in last 60 sec: 9666
    the guy sitting in back of one of my transports, probably warlords specific
floral socket
# floral socket https://cdn.discordapp.com/attachments/1219689481964097557/1391335204408266803/A...

note the timestamp of the video (clipped 2s after crash) and how there is nothing in the .rpt about it (i still get no .mdmps)
https://cdn.discordapp.com/attachments/1219689481964097557/1391341063528120390/image.png?ex=686b8ad7&is=686a3957&hm=a967b308d697e553d756b46b2a5e8c5a7bae0affaa625b2c0b89890f3ce727a5&
https://cdn.discordapp.com/attachments/1219689481964097557/1391340907290165389/image.png?ex=686b8ab2&is=686a3932&hm=a614df1de31c1e18b54d137b207e61420221eb3187d9995c8ceb43f0c67c372f&

fast hornet
sullen crow
fast hornet
#

HC are extra, but they are very rarely needed with a modern CPU anyway.. yes, server.

empty goblet
#

so, i guess thanks to @my name is jeff
when the profiling branch client crashes (performance and profiling binaries),
if it's while using SYSTEM malloc it seems to fails to produce dump files πŸ›
while with mimalloc 2.2.3 and tbb4 it produces crashdumps fine

#

so now, hopefully we have the thermals crashdump tracked

floral socket
#

(the videos were all using Windows allocator (system) malloc)

restive turtle
#

@dedmen

Is there any optimizations that have happened with waypoint calculation for AI and such.

#

or is that off the table for compatibility reasons

whole cloud
whole cloud
uncut wigeon
#

Cores or Logical Processors?

whole cloud
uncut wigeon
#

yeah why I asked always been fighting for what side of the CPU it should use. since affinity isnt an option without a 3rd party program or using task manager on launch every time.

whole cloud
#

affinity is an option since recently

uncut wigeon
#

so /AFFINITY (hex flag here) ?

charred pagoda
uncut wigeon
#

thats only the main thread is it not?

#

nvm read the wrong one

#

but ty yeah was looking for the syntax. already know the hex setting

pastel folio
#

@empty goblet I am back. reforger perf branch when? ;)
also, long time. Hope all is well!

whole cloud
inland dew
#

so if I want arma to use 8 first cores (0-7), then it should be 80, right?

whole cloud
#

No. I will not explain hexadecimal to you, look it up on the interwebs.

fast hornet
uncut wigeon
inland dew
#

so it's -cpuAffinity=0x00000000000000FF then

#

for 0 to 7, so 8 cores (no HT)

inland dew
#

found a problem
when using cpuAffinity, cpuMainThreadAffinity doesn't work

#

8 p-cores + 16 e-cores (24 cores no HT)
1st screen:
-cpuMainThreadAffinity=0x0000000000000080 (core nr. 7, so 8)
-cpuAffinity=0x00000000000000FF (0-7, so limited to 8 first cores)

2nd screen:
-cpuMainThreadAffinity=0x0000000000000080 (core nr. 7, so 8)
no cpuAffinity used

20250707_154003.jpg 20250707_154824.jpg
fast hornet
#

never tried the mainthread one.. but it should? Did you define a single core for it?

#

nvm πŸ™‚

#

ah .. no. That's expected.

#

When the game's affinity is 0-7 it can't use 8 for anything of course.. it needs to be within the affinity range

inland dew
#

so I need to do 0-6, so 7 cores and 7, so 8th core as main?

fast hornet
#

you don't need cpuMainThreadAffinity at all unless you have a specific reason for it.. but when you specify it i'd expect it needs to be within the range specified by cpuAffinity - because those are the only threads it's assigned to.

#

since how should it do anything on 8 if it only runs on 0-7 πŸ˜„

inland dew
#

0-7, means 8 cores

fast hornet
#

yes, but 8 would be the 9th core.

#

it's always from zero

inland dew
#

that's what I wrote

#

main thread is set to 7, so 8th core

fast hornet
#

you wrote "core nr. 8".

inland dew
#

so I'm still right

#

cpuMainThreadAffinity doesn't work, when used with cpuAffinity

#

affinity 0-7, main thread 7 - main thread affinity doesn't work

#

main thread 7 only (no cpu affinity) - works

whole cloud
inland dew
#

yes, with H

#

not missing letter in the command

#

only typing the text here I missed H

whole cloud
#

checked RPT? it logs when affinity couldn't be applied for some reason

quaint flame
#

I feel like we need a thread just for affinity

#

#community_wiki message I made some suggestions to improve the docs. Please add to the list if something else could be added. (E.g. I didn't know it logs to rpt)

inland dew
whole cloud
#

mh looks good 🀷

inland dew
#

but doesn't work

orchid mesa
#

CPUCount : 9 ? πŸ€”

inland dew
#

I'm not using cpuCount at all

#

so yeah, cpuMainThreadAffinity doesn't work, when used with cpuAffinity...

tardy adder
#

anyone else realized that Steam doesnt show the current server anymore since the recent update from performance to stable. (talking about friendslist) only reason im posting here since this has been happening since the prof update to stable and ive only used prof

opal hound
#

Yes it's done that for a while on perf (though I prefer it this way πŸ˜†)

empty goblet
opal hound
#

Going to stab a guess that it was related to the SteamAPI update? I don't remember when that was

inland dew
tardy adder
opal hound
#

1FF is 0-8, but it's mostly irrelevant

uncut wigeon
#

sadly I regret opening this bottle in the discord real issue is MS has basically zero documentation on this. and anything they do have is based on 32bit API

inland dew
restive turtle
# whole cloud No, more details of what you mean?

Generally, I mean when using a lot of way points/ having a lot of AI Moving does a good amount of fps drop although its fine when they are idle (I was wondering if there's room for optimization there)

#

I don't have specifics its more so a topic suggestion or conversation

#

Probs wasting time sorry notlikemeowcry

whole cloud
#

would need a repro

quaint flame
#

@whole cloud Do both affinity startup parameters log to the rpt if setting could not be applied? Or does only cpuMainThreadAffinity do that? (Updating the wiki right now)

whole cloud
#

main thread only logs to RPT when it failed

quaint flame
#

Rgr

whole cloud
#

the overall affinity logs when success and failed

quaint flame
#

ok

idle wren
#

Can confirm the crashes are still happening when going into thermals on the profiling branch

whole cloud
#

Just press submit, and I'll find it πŸ«‚

whole cloud
#

600 crash reports since thursday πŸ’€
previous prof had 700 over 10 days.

Either alot more people played on the weekend, or something is not good πŸ˜„

#

ajo I see the thermals crash

empty goblet
#

shows how the TI optics are popular 🀣

void badger
#

Was going to follow up with "been running stable client and perf server with no crashes" and just found out my server segfaulted earlier after about a 60 hour uptime

empty goblet
void badger
#

I'm a chump running Linux perf so no dumps here 😩

whole cloud
#

Only 126 of them are the thermals though
45 XAudio
22 game exit
102 0xDEAD error messages, I won't even look at them
9 amd graphics driver

#

The storage space used for all things script is limited to 3gb.
There are quite alot of people running out of space in that and the game force crashing itself meowsweats

#

Also guys, if you are a modder, and you are F'ing up your own config while modding, and the game throws an error and kills itself.
Please do not submit that crash, that stuff is really useless for me πŸ₯Ί

There are soooo many crashes of broken configs or preprocessor errors

#

I made a fix attempt for a crash at game exit, that was just a guess hoping it would fix it.
Well since that version the crashes are gone so βœ…
Having this info is amazing, that was a great idea

heavy vortex
#

Server auto-submits, right

inland dew
#

@whole cloud @empty goblet disabled 16 e-cores in the bios, so now that i have only 8 p-cores (hyper-threading off), so all cores are of same type (non hybrid architecture) - main thread 7 + cpu 0-7 works πŸ€”

whole cloud
whole cloud
#

But server is quite stable, compared to client 🀣

firefox_2025-07-07_22-29-51.png
heavy vortex
#

Are you sure that's not just fewer servers than clients :P

whole cloud
#

And considering client also has a dark number of players who don't click submit

heavy vortex
#

I should probably force a crash just to check what this submit UI looks like

#

It's not the old windows one, right?

whole cloud
#

its the same as reforger

#

its probably easier to look at it in reforger

heavy vortex
#

It would if I had any inclination to install reforger :P

#

but I was waiting until it crashed less

inland dew
modest geode
dense light
inland dew
#

with only 8 p-cores enabled (no ht, no e-cores)...

  • tried to do cpu affinity 6 cores (0-5) and main thread affinity core 4 (3) - cpu affinity doesn't work and main thread affinity doesn't work
  • tried to do cpu affinity 4 cores (0-3) and main thread affinity core 2 (1) - cpu affinity doesn't work and main thread affinity doesn't work
#

with only 8 p-cores enabled (no ht, no e-cores)...

  • tried to do cpu affinity 8 cores (0-7) and main thread affinity any core I want - works (cpu affinity + main thread affinity)
#

πŸ€·β€β™‚οΈ πŸ€”

heady plinth
#

how much do you have to f' up your config for the game to crash? i didn't even know the game can crash from config stuff πŸ˜„

#

guess i will blame addonbuilder again, which just compiles whatever you feed it

modest geode
inland dew
modest geode
inland dew
modest geode
#

Ah, that's just not needed at all, this is what one of my servers use -cpuAffinity=0xFE -cpuMainThreadAffinity=0x2 That's core index 1-7 affinity and core 1 as main
returns this in rpt

22:03:14 Process affinity forced to 0xFE (requested 0xFE)
22:03:14 Process has custom affinity mask FE/FFF
inland dew
modest geode
inland dew
#

so 2-8 cores + 2nd core being main?

#

what's your cpu model?

modest geode
modest geode
inland dew
#

so non-hybrid cpu

modest geode
#

Kinda weird, mine is returning 22:03:14 Process has custom affinity mask FE/FFF FE/FFF would be 11111110/11111111
While yours return 16:07:00 Process has custom affinity mask 1FF/FFFFFF 1FF/FFFFFF would be 111111111/111111111111111111111111
I can only assume that "/FFFFFF" would be total detected cores?

inland dew
#

just ignore that

inland dew
#

it's FF, for 0-7

modest geode
#

You still getting /FFFFFF though? That would be 24 cores total

inland dew
#

yes, 24 total

#

8p+16e 14900ks ht off

modest geode
#

Try moving the affinity declaration to before -malloc? This is the order i have it in
{self.path}\arma3serverprofiling_x64.exe" -port={self.port} -cpuAffinity={cpuAffinity} -cpuMainThreadAffinity={mainThread} -malloc=mimalloc_v224_20250703
I can't remember if I put it in that order for a reason or if it just happened to be that way

inland dew
#

it is before anything

modest geode
#

have you tried with both profiling exe and normal one?

inland dew
#

i just have the check mark for malloc in the launcher - not using command

modest geode
#

Gotcha

indigo anvil
#

I've just switched to mimalloc_v224_20250703 along with -maxmem=8192, however running the mission seems to run away with all the memory it can find (22G) . Any tips on what can go wrong?

inland dew
#

maxmem doesn't work

#

never has

indigo anvil
#

lovely. Any other way to constrain server process mem hunger?

modest geode
indigo anvil
modest geode
modest geode
indigo anvil
indigo anvil
heavy vortex
#

We don't currently compile config so Arma killing itself is how we find config errors.

light cargo
#

tbh, crash report should be no opt out for clients on profiling branch too

floral socket
tawdry gazelle
#

Hi, anyone would like to test if mimalloc has any performance difference on different compilers?
One is Intel C++ Compiler, another is LLVM.

floral socket
tawdry gazelle
hoary jolt
hoary jolt
tawdry gazelle
tawdry gazelle
hoary jolt
tawdry gazelle
magic elm
tawdry gazelle
whole cloud
heady plinth
#

i see. pboproject would instantly complain and not even let you finish πŸ˜„

whole cloud
#

I saw a couple mimalloc crashes, but they were 223, not the new 224

tawdry gazelle
#

Oh poor alignment...meowsweats

whole cloud
#

Well my crashes are crashes inside mimalloc.dll
In that one, I see no mention of it

tawdry gazelle
#

yep. they say crash occurs since they switched from mimalloc v223 to v224, but mdmp gives no mention of mimalloc.dll, that's why I can't tell if it's a crash by arma exe or mimalloc.

floral socket
charred pagoda
tawdry gazelle
charred pagoda
whole cloud
#

Anyone know which mod is "USAF_GBU38"?

#

Ah USAF Missilebox, I even have that

quaint flame
whole cloud
#

No RHS uses RHS prefix

charred pagoda
whole cloud
#

Someone crashed every player on the server by dropping one of those πŸ˜„

charred pagoda
#

by explosion fx?

whole cloud
#

the bomb being spawned itself, no explosion

quaint flame
#

Well..it was effective at least...in its own way.

whole cloud
#

lol I think, it has an Init eventhandler, that deletes itself.
Object created, init event runs, deletes itself, PostInit event.... oh no what happened

heady plinth
#

dedmen about to start fixing other peoples mods now

whole cloud
#

engine bug though.

They actually delete the object by calling setPos
πŸ˜„
When we create the object, we don't set the reference counter and leave it at 0.
setPos takes a reference to the object (tbh its stupid code that shouldn't be there and wastes performance.. I'll remove that), and once it lets go of the reference again, it goes to 0. That means we should delete the object now πŸ˜„

lethal trench
#

mods about to break with fix to setPos πŸ₯·

heady plinth
#

"you cant change that, my mod depends on the object being deleted by setPos!"

autumn timber
#

xkcd: workflow

slender lagoon
heavy galleon
#

I had to use a delay because setPos in fired event didn't work.
Is it because of the same reason?

whole cloud
#

did your game crash? no? then no

#

Function call gives parameter 0x00007ff693647940
Then inside function it crashes, because parameter was 0xfafd7ff693647940
?
Mh I wonder what CPU that user has..

12th Gen Intel(R) Core(TM) i9-12900K
Ah yep. Well good luck with your warranty

#

Can someone tell the optre guys to fix that.
I got 9 crashes because of that

notepad_2025-07-08_16-25-00.png
tulip arrow
# tawdry gazelle Hi, anyone would like to test if mimalloc has any performance difference on diff...

intel 12600k (e-cores disabled)
64gb ddr4 "3500"
nvidia 4060ti 16gb (undervolting max 0,9v)

windows11

MIMALLOC_ALLOW_LARGE_OS_PAGES 1
MIMALLOC_ARENA_EAGER_COMMIT 1
MIMALLOC_PURGE_DELAY 100

TBB_MALLOC_USE_HUGE_PAGES 1

process lasso Arma3_x64.exe priority above normal

Arma3_x64.exe -hugePages -limitFPS=1000

video 1920x1080 "standard"
difficulty "veteran"

https://i.postimg.cc/WbNM2gXX/mimalloc-v224-20250703-1.jpg
https://i.postimg.cc/yYxkQPf4/mimalloc-v224-20250703-2.jpg
https://i.postimg.cc/DZ5w5JLw/mimalloc-v224-20250703-3.jpg
https://i.postimg.cc/K8LcwsBV/mimalloc-v224-A-1.jpg
https://i.postimg.cc/s24jDQn4/mimalloc-v224-A-2.jpg
https://i.postimg.cc/yd6VJ1t4/mimalloc-v224-A-3.jpg
https://i.postimg.cc/XJ2n9M6Z/mimalloc-v224-B-1.jpg
https://i.postimg.cc/XNzVJ0gm/mimalloc-v224-B-2.jpg
https://i.postimg.cc/rFRqRYnm/mimalloc-v224-B-3.jpg
https://i.postimg.cc/43sGM4J7/rpmalloc-v145-1.jpg
https://i.postimg.cc/zBfrzdHq/rpmalloc-v145-2.jpg
https://i.postimg.cc/2ysRxrc5/rpmalloc-v145-3.jpg
https://i.postimg.cc/1R2QSrmZ/tbbmalloc-v202170-1.jpg
https://i.postimg.cc/zXh1GHvH/tbbmalloc-v202170-2.jpg
https://i.postimg.cc/DwK3Jf4j/tbbmalloc-v202170-3.jpg
https://i.postimg.cc/CxHTjyjF/tbbmalloc-v202170-4.jpg
https://i.postimg.cc/6QYkmdjg/tbbmalloc-v202170-5.jpg
https://i.postimg.cc/tTcLX9Hz/tbbmalloc-v202170-6.jpg

whole cloud
heady plinth
#

this mesh needs to be Component03

grafik.png
#

no clue why. i suspect some ancient dark magic

#

there is no Component02.

whole cloud
#

meowsweats
Well I had crash it trying to access 41 when only two were present (but I couldn't see the numbers of the two)

heady plinth
#

the issue did not make my game crash, though. i just noticed the rpt spam and it confused me so i investigated

#

there might still be some issue with the mods uniforms ofc

#

this is how the named selections need to look. 42 is the last named entry before the components starts (41 if we start counting at 0) - so you might still be onto something here :p

grafik.png
unborn urchin
#

why? dunno

unborn urchin
heady plinth
#

amazing

#

maybe a2 geo had another mesh? or old opfor and this is legacy code? πŸ˜„

unborn urchin
#

a2 mesh was different yeah it was the 'long' pointy geo that people were using

#

cant remeber if it had an extra component tho

whole cloud
timid veldt
#

Where can I report a problem? I'm using the Profiling but my FPS runs at 50+ but then will just drop down to 7- and then I lag big time. Thanks

whole cloud
whole cloud
#

First crash with SimpleVM πŸ˜„

The code

if (!argument.IsNull())
  argument.GetType(); // Crash! Argument was null

The users CPU: 12th Gen Intel(R) Core(TM) i7-12700H
Yeah.. Good luck with your warranty

empty goblet
whole cloud
#

I assume the famous intel degradation story

empty goblet
# whole cloud I assume the famous intel degradation story

right ... 12th and 13th are literally like lotto, there is chance for win but most of time it's loss ... because there is so much what could go wrong
(tho users who have 13th gen told me recent firmware UEFI and drivers updates fixed theirs crashes)

#

it's also interesting to hear that several people stopped crashing with 2.2.4 mimalloc (vs 2.2.3)

whole cloud
#

We had alot of tbbmalloc crashes in latest prof too, because of the thermal imaging stuff

empty goblet
fast hornet
#

guess my 10+ 13900 server crashes trained your eye for intel degradation πŸ˜‚

empty goblet
fast hornet
#

haven't heard about xeons degrading so far, only the consumer chips.. but maybe it's just because of limited press exposure

empty goblet
#

it wasn't actually degrade with those initial 13th gen xeons, those were bugged from get go it was just hard to get the errors to start appear

#

not-our partner but one of my friends in game server hosting told me they replaced all of those early 13th gen xeons w/o mercy after being unable to figure it out

charred pagoda
fast hornet
#

our A3 server runs on a dedicated 13900 machine and i (and as a result dedmen) was chasing weird random crashes and errors for months until it finally completely died.
Hoster replaced it with a new machine and everything's working perfectly again. (just hope the second 13900 survives for longer...)

empty goblet
fast hornet
empty goblet
#

i think i've some 13th gen Xeons too, let me check

#

actually we don't , uff 🀣 guess someone listened to my scream while ordering replacements

heady plinth
#

oh, I just remembered that my old pc had a 12gen cpu and it was randomly crashing since always

#

hard blue screen of death

whole cloud
empty goblet
heady plinth
#

more like off topic, because it was crashing with anything at random.. except arma :p

empty goblet
# whole cloud it eroded my trust in intel

for many it was final nail to coffin and then AMD came with X3D for anyone with both AM4 and AM5 sockets , so for gaming or good parallel compute it was simple choice

whole cloud
#

Another crash fix prof tomorrow, hopefully this time I won't introduce another big new one πŸ˜„

fast hornet
empty goblet
quartz rampart
whole cloud
#

Then I'll still blame it on intel and overclocking πŸ˜„

quartz rampart
#

just wanted to clear that πŸ˜„

empty goblet
heavy vortex
empty goblet
fast hornet
#

DDR5 speed and stability was a shitshow for everyone at the start to be fair.. modules nowadays are a lot more reliable with tighter timings, lower voltages etc. compared to 2-3 years ago

carmine stump
#

dedmen fix AI NOT stopping when inside a jet even if you 1-6

empty goblet
carmine stump
#

pretty much what I already said. tell an AI from your group to get inside a jet as pilot, tell him to stop and he wont. jet will be rolling regardless

#

this is bad in cti when you want to have a jet in standby at base.

inland dew
#

that's the place for that. definitely not here

carmine stump
#

they will never fix it there.

#

dedmen can probably fix it in a day.

inland dew
#

also need to see if there is already a ticket on feedback tracker for that or if no, create one and describe that in the ticket

carmine stump
#

I wont, just leaving it here hoping itΒ΄s a quick fix from him. if not, ok.

inland dew
heady plinth
#

any ai behavior thing is never a quick fix

analog acorn
thin wyvern
patent dome
uncut wigeon
#

I just wanted to thank Dedmen and everyone else that was involved in that convo the other day because I just finally got to double check. and Process has custom affinity mask FFFF0000/FFFFFFFF got a victory!

tawdry gazelle
whole cloud
whole cloud
whole cloud
hoary jolt
whole cloud
whole cloud
patent sky
#

Soonβ„’ no more crashes, right? πŸ˜‚

quaint flame
#

Seems the crash reporter already pays off.

feral harness
inland dew
#

but fix NOW!

indigo anvil
tawdry gazelle
ripe willow
# whole cloud

-Fixed: When a land waypoint was given above water the helicopter would attempt to land on water. Isnt that expected when you set waypoint above water ?

whole cloud
ripe willow
whole cloud
#

yes

idle wren
whole cloud
#

The scope/muzzle specific resolution/zooming parameters

empty goblet
whole cloud
#

diwako'nt

hexed verge
#

pls

#

Found a weird one. I was trying with last profiling version to connect to my unit's server. I got this error as our server runs the "small" cdlcs. (as a side note those flder are symlinked to the steam supplied addon folders inside the arma directory)

I figured i somehow corrupted something and just deleted the folders and let steam verify and redownload them. just for it to happen again

I switched off the profiling branch and tried connecting again, and it worked.

dwarden asked me to verify on current profiling version, so i did, this screenshot is from the current version.

The server is running the 64bit linux binary from the stable branch.

image.png
whole cloud
#

Prof has that change for bikey loading, but if you are not running profiling server that obv doesn't apply.
Its a bikey side only thing, so it'd only apply on the server there

#

When did you last try? If its the bikey thing then v6 would've been it

hexed verge
#

i tried just now, screenshot is like 4 minutes old

whole cloud
#

I mean when did it last work, previous prof build

hexed verge
#

one week ago. it started happening when i tried playing this sunday.

#

was not palying much muliplayer as i was busy irl and used othe free time to do some modding

whole cloud
#

so v6, last thursday.
Which added -keysFolder argument

#

But if you don't set the argument, then it wouldn't do anything.

I assume setting it -keysFolder=!keys on clientside might break like this?

fast hornet
#

btw. just a quick question inbetween: If you don't explicitely negate the default key folder it always looks in there as well, correct?

whole cloud
#

yes

hexed verge
whole cloud
#

yeah, and verify that you don't have -keysFolder in start parameters

#

Also maybe RPT says this?

WARNING! No bikey's were found in 'keys' folder. Please check, there should always be atleast one key for the base game.

hexed verge
#

checked my latest rpt with the mismatch messages and there is no "-keysFolder" param anywhere as i use A3S to launch arma.

also nothing of the sort of the warning message inside my rpt

#

stable: works
v5: works
v6: does not work

whole cloud
#

The question then is, how did I mess something up, that when rechecking the code is not messed up, and that so far no-one else had issues with

hexed verge
#

maybe it is some weird compounded issue that only happens with a specific setup

#

these are the start up params minus the -modones

"E:\Program Files (x86)\Steam\steamapps\common\Arma 3\arma3_x64.exe" -debug "G:\GIT\diwako_cbrn\diwako_cbrn.vr\mission.sqm"
-noPauseAudio
-limitFPS=144 -noPause -hugepages -nosplash -world=empty -connect=138.201.249.238 -port=2302 -password=****

"G:\GIT\diwako_cbrn\diwako_cbrn.vr\mission.sqm" is to launch into a mission directly

whole cloud
#

If you remove the -connect and direct loading of mission?

#

maybe the connect triggers too soon?

hexed verge
#

will try

fast hornet
#

if it helps, i also use a batch with -connect= to start the client and it works fine with v6/v7

hexed verge
#

so far
-connect without mission: works
no -connect and no mission: works
-connect with mission: does not work
no -connect but mission: does not work

funny how that is... so it is again a me issue....

whole cloud
#

But I don't get how a mission would do that 😒

hexed verge
#

neither do i, i just want to launch into 3den :V

#

also if you set the path to some none existent sqm file it also does not work

#

strange enoguh tho it seems to be only an issue with ebo's?

fast hornet
#

-keysFolder works like a charm btw.. would be cool if the game would just recurse through subfolders as well so i could just point it at the mod folders. Since i generate those anyway.
How it is now i need to generate it for every single mod since the folder name is not standardized.. i've seen key, Key, keys, Keys and server_key so far meowfacepalm

light cargo
#

or make automatic loading keys folder from -mods list

fast hornet
#

yeah that would also be a nice option

light cargo
fast hornet
#

the key folder should not contain any mods that are not loaded anyway though?

light cargo
#

i'm referring to doing recursive folders search in say !workshop

fast hornet
#

ah no, that would be a bad idea

#

i'm talking about pointing it at each of the mod folders

light cargo
#

imo taking advantage of the mod list is better

#

it's mostly standardized as the folder name in the game afais

#

isn't that how the launcher detects signed mods?

#

pbo + bisign + keys

#

i'm not sure though

fast hornet
#

the key file is server side only. The client calculates a key/hash from the pbo files .. otherwise you could just use modified pbo files with a correct key

light cargo
#

ah

fast hornet
#

or well, not a hash.. i guess it reads the signature and that must match the key on the server

light cargo
#

i imagine launcher just processes the pbos and bisigns

#

this subsystem is like gpg sort of?

fast hornet
#

never looked into what arma actually does.. but yeah something like that. You sign the files with your private key, the server gets the public key and then you can check if it's actually corrupted somehow or not

light cargo
#

the bisign is the sig equivalent

hoary jolt
# whole cloud

WEIRD! I was testing something else in ARMA, DXVK with my 7800X3D iGPU, and then this came online and updated. After the update, I GAINED FPS on the iGPU. Before I was running 2K resolution at 50% scaling ~ 55 - 60 FPS, and now I can run it at 100% if I turn down HDR, and OFF PP, AA and AF, same FPS!! It is totally playable and the only issue was streaming of assets as it will hitch coming from DRAM, not VRAM, and smoke at 100% is laggy.

whole cloud
#

Well then thank your OS automatic updates. Because there was no performance in thid

hoary jolt
slow dust
#

Do people usually also run profiling branch on the server as well? Or just on clients?

void badger
#

Perf binary server is a great choice

#

And still benefits clients running it in some ways

whole cloud
#

Server binaries get the same Bugfixes and performance improvements.

charred pagoda
empty goblet
empty goblet
hoary jolt
# empty goblet the last segments feels like plea for more optimized smoke particles πŸ™‚

It's not but good idea and you can verify for academic curiosity. πŸ˜„ It's something I've noticed while playing around with the iGPU that smoke particles at 100% greatly affects performance (I have this setting maxed out for immersion), dropping it to 50% see high FPS gains, while TEXTURE quality barely affects FPS so long as we set the iGPU to GAME mode, not UDMA in the BIOS. Lowering PP, AA, AF and HDR settings also secure high gains so that iGPU is actually playable. Guessing it is hitting specific HW limitations - TMU or Shaders.

whole cloud
#

This is looking pretty great. Now you just gotta keep it that low

firefox_4rE0U7tC8S.png
#

And most of them are just 0xdead forced crashes, not even real ones
Some server hoster getting their server.cfg wrong a dozen times in a row 🀣

K62zRiVrNd.png
patent sky
whole cloud
#

I would prefer to exclude them after we receive them.
Sometimes there might be 0xdead's that are interesting or that we might want to prevent

whole cloud
#

3 servers are still running prof v5, and crashing on already fixed stuff.

Last week:
Wed: 76 crashes (21 0xDEAD)
Thu: 81 crashes (21 0xDEAD)
This week:
Wed: 49 (22 0xDEAD)
Thu: 30 (20 0xDEAD)

So yesterday we had 10 "real" crashes total.
Out of those 10:
6 were from one server who's own extension "bandicam" was crashing?
2 of them don't have any logs so 🀷
1 is still running v5 and ran into a already fixed crash
1 is some crash inside the graphics drivers 🀷

whole cloud
#

Today so far
7 0xDEAD (All from the same server with a preprocessor error in a config)
4 real ones

1 proton without crash files
1 with... doesn't make sense, but its an i9-14900K so..
1 crash on exit, I've seen that before, something wrong on our end but don't know where.
1 server crash in particle simulation meowsweats

tawdry gazelle
#

@whole cloud Hi, I see Arma 3 allocating memory through either aligned or non-aligned malloc interface.
I am just being curious: in what case or for what purpose does it allocate through aligned malloc? And what for non-aligned malloc?

tawdry gazelle
whole cloud
tawdry gazelle
#

I recently read an article about memory allocator types and implementation and wondered if forcing all non-aligned allocation to be aligned would benefit performance. Then I came up that aligning itself also consumes performance, so I gave up. x_

whole cloud
#

I think we do depend on the allocator giving us a minimum alignment of 8 anyway

#

Oh also our page-size allocators are aligned to 4096.
They are generally used for pool allocators or filecache

tawdry gazelle
whole cloud
whole cloud
#

I have alot of crashes at game exit, while its trying to unload models.
And its crashing because models aren't present.

And indeed..

6:57:37 Warning Message: Cannot open object zeus_games_pcu\pcu_untucked.p3d

I can add checks to this, but checking for objects without models would be hundreds of checks, and more things to be checked == less performance..

So probably just gonna ignore all the reports there, and hope people realize they should fix their mods

#

Or... I turn "Cannot open object" into a popup error πŸ€” Mh

#

Fixed by force crashing the game earlier βœ…

#

oi. I hope that's just corrupted files? @heady plinth

notepad_zIOH92JSqb.png
honest fulcrum
#

Those booniehat lines look like they maybe meant to use \a3\characters_f\data\ui\icon_h_booniehat_desert_ca instead of the same but with their mod pathing in the front

knotty wraith
#

Can you separate the surface detail and the grass removal?
How is this appropriate?

whole cloud
#

That sounds possible

knotty wraith
whole cloud
#

I certainly can

knotty wraith
#

surface detail and lack of grass feel like two different worlds. I don't understand why they are interconnected

#

Well, it's like a cry from the heart) the problem is old - everyone is used to it

autumn timber
#

Especially in the context of people turning their grass off to get an unfair advantage.
However, now that I think of it, people will be able to turn the grass off without being punished by having to look at 2005 terrain 😦
Not that I'm against adding that feature! It's just unfortunate that we'll have that situation as a side-effect

knotty wraith
#

playing pve from grass only minus fps and you are seen by the ai, whoever has the opportunity removes the grass, but the detail is reduced
for pve projects grass does not make sense until the ai takes grass into account

heady plinth
#

no issues on my end

whole cloud
#

maybe a 3rd party mod trying to use ws paths but getting them wrong

heady plinth
#

rpt log is fine as well

#

whatever is going on, it's not us πŸ˜„

#

the paths look ok to me. can't find anything obviously wrong right now

#

also they havent changed in 4 years now, so no clue

whole cloud
#

Thanks for checking πŸ«‚

void badger
whole cloud
#

90+% of the code accessing the model, never checks for the model potentially not existing.
Because in real use, you would also never have an entity without a model, that's just not something that happens so its not checked for.

But when the model is not there, the starting code just lets it continue on anyway, thats where the issue is

spiral pond
whole cloud
#

yes but I don't want to look now

tawdry gazelle
spiral pond
#

You could share with @heady plinth , so he can figure out the source and contact the author.
Less crashes = good, less complaints "on WS" = good

whole cloud
whole cloud
heady plinth
#

meh, if they have issues, they can write us

empty goblet
whole cloud
#

Its for AI visibility

empty goblet
olive silo
#

Ran into something odd but no logs or things like that unfortunately for it. Server (latest profiling) got stuck at 15 to 17 FPS for minutes at a time (25 players or so) and it looked like the server was only using one thread, even though it has 16 available. It fixed itself eventually and fps jumped back up to 300+. Anyone else run into this?

heavy vortex
#

most likely it was working extremely hard at something that's not threaded. Unscheduled scripts maybe.

uneven bluff
#

Does it make sense to try to combine setVariable when sending over the network?
As in, is it worth manully combining a bunch of setVariables on the same object into a single hashmap, serialize, and then deserialize on the client end, or are these kinds of optimizations already done automatically by the game?

#

^ This is in context of a script that likely sets a bunch of variables in a single frame.

heavy vortex
#

If you had a lot of setVariables that you were setting at the same time, maybe?

#

Otherwise each one is a separate network message.

uneven bluff
heavy vortex
#

Messages are combined into packets but there's still per-message overhead.

#

Whether that's significant compared to running extra SQF is another matter.

#

The per-message cost should be much lower on perf branch.

nocturne obsidian
# whole cloud Its for AI visibility

I noticed that regardless of how much memory a player has (from 8GB to 32GB) or how powerful their graphics card is, the game always starts to lag when the number of particles approaches the engine's limit. I think this is an internal engine issue. Where exactly is the performance bottleneck in the engine?

restive turtle
#

This is just my experience with engines that are the age of around this time

#

I dont actually know anything lol

restive turtle
#

At the absolute limit

heavy vortex
#

I suppose it might be doing something daft like grinding through the entire particle list to find the oldest one each time it creates a new one. Not seen a similar issue personally though.

nocturne obsidian
restive turtle
#

but i run 7800X3D and 4090 (both Overclocked)

orchid mesa
#

Not completely sure if it's profiling related, however I've been experiencing reproducible desync roughly since the current build was posted.
Start up arma -> join server -> after ~15 minutes my connection suddenly destabilizes with severe rubberbanding until eventual kick for >2000 ping
however I can immediately reconnect without desync, and it doesn't happen again. appears to only happen on first server join after starting the game

nocturne obsidian
restive turtle
#

I use Murr

nocturne obsidian
#

Then it's probably a problem with the computer.

heavy vortex
#

My PC is trash, but I never saw a need to turn up particles and I heard there's a hard count limit that isn't changed by the setting.

#

If that's how it works then there might be a detectable tipping point. Like 20 smoke grenades is fine but after that it goes rapidly downhill.

limpid fern
#

Hey, yesterday I played an operation but we had a very strange issue were every 3-4 seconds there would be a major stutter. I checked the rpt file and found this error that was being spammed every 2-5 seconds.

ThingX 22707732b00# 172506: dummyweapon.p3d WeaponHolderSimulated was falling under terrain! Position corrected to ground [6202.79, 27.40, 5241.59]. Disable simulation to keep object under terrain.

Is it possible that several WeaponHolderSimulated objects constantly falling through the map could cause this issue?

The issue was happening on the performance branch so i thought id ask this in here

heavy vortex
#

Those are normal. Not sure if they have much perf cost.

#

2-5 seconds sounds like it's just one of them, but you can check the IDs.

limpid fern
#

there were several objectsIDs that were falling under the map according to the rpt

analog acorn
#

Were you perhaps using certain radars with very long range? e.g. some mods with ship-based radars, RHS strategic SAM radars.
There's a thing with radar, which I don't think got fixed yet because it's complicated, where they force dynamic generation of filler terrain if they search outside terrain bounds, and this tends to manifest as a stutter every few seconds if their coverage extends a long way out of bounds.

limpid fern
#

the issue could be resolved by turning down the view distance

carmine stump
heavy vortex
#

Maybe -enableHT should be removed. Lots of servers using it for archaic cargo-cult reasons, often with violently bad results in 2.20.

#

If you really want to use more threads then there's -cpuCount with a lot more control.

fast hornet
#

just documenting it properly might be enough.. i believe before Dedmen or someone else mentioned it here maybe 1-2 years ago noone knew that all it does is doubling the amounts of threads. Even the BI wiki still does not reflect that.

heavy vortex
#

Uh, I don't think it'll double the number of threads unless all your cores are SMT.

#

I don't think documenting it properly is enough because there so much old/bad config out there.

fast hornet
#

Meh.
I agree with people that run the client/game being mostly clueless, for them it could make sense to prevent some ancient wrong steam community guide hurting their performance.
Running a server requires keeping up with technical changes.. if you don't do that, your fault. Imho.

hoary jolt
nocturne obsidian
# whole cloud 16 byte probably, no one would do bit

In Arma 3, functions like addForce and other PhysX-related implementations sometimes don't work properly when applied to units. Occasionally, the unit may fly up into the air and then get stuck mid-air, or show other strange behaviors. (Note: I'm not applying excessive forceβ€”just enough to possibly lift them 30–40 meters.) Is this normal, or am I using them incorrectly?

feral harness
whole cloud
whole cloud
heady plinth
#

would be nice if this issue could be fixed :x for some reason the game keeps failing to keep them above surface. they jump up and clip back into the surface, jump up again, etc.

empty plover
#

well its been a while, so a friendly reminder to @whole cloud πŸ™‚ would be awesome to get the loading optimized

whole cloud
spiral pond
#

possible to recreate the engine check in sqf with PX EH to help with repros?

heady plinth
#

it's a similar thing to the vehicle tire clipping which i was showing a video of long time ago

#

sadly it seems to be hard to reliably repro for some, but for others it happens a lot

spiral pond
#

curious if its happens more frequently with low(er) cell size

whole cloud
#

😾

2025-07-14_11-15-29.png
whole cloud
#

weekend summary
106 reports
-80 0xdead

26 "real" crashes

5: Two servers that pirated the game crashing because of it. May your milk be spoiled!
1: Graphics driver crash, I think thats the same user as the last couple times
2: XAudio
3: Crash that was already fixed in v7 but they didn't update
1: blobdoggoshruggoogly
2: Missing log files
1: Bug in SVG terrain export causing crash at exit, fixed.

The rest are all servers crashing at shutdown. Clearly something is buggered but I don't know what. But crash at shutdown isn't that big of an issue anyway

nocturne obsidian
# whole cloud if I had a repro of it I could look

I conducted extensive testing and found that if an object is killed while in flight, it will immediately stop physical simulation and freeze mid-air instead of falling. I'm okay with that.
However, there's another issue β€” please take a look at my video:
https://drive.google.com/file/d/1UzSJCWHAAf8Lac3CPLbhRMAqxsyQ1Uag/view?usp=drive_link

After being thrown, sometimes characters snap back to their original position.
Note: this is not caused by having too many simulated objects β€” even with only a dozen units, if I test several times, one or two of them may experience this bug. It seems to happen randomly.

In the video, I used a large number of units only because the more characters there are, the more frequently the problem appears β€” which makes it easier to demonstrate.

Google Docs
202507141827.mp4
nocturne obsidian
#

_explosionPos = getPos c1;

{

// find vector to ground zero
_ovec = (_explosionPos vectorFromTo getpos _x);

if (_x isKindOf "CAManBase") then
{        

        _x addForce [_ovec vectorMultiply 7000, [0,0,0]];
        
        _x awake true;

        [_x] spawn {
            params ["_x"];

             sleep 1;
             
             // _x setDammage 1;
        
        };
};

} forEach nearestObjects [_explosionPos, ["AllVehicles"], 500];

whole cloud
#

I also finally went through all the linux proton crashes.
ALL of them were pirated game...

indigo anvil
whole cloud
#

no

heady plinth
#

should build a filter that automatically discards all reports from pirated game

quaint flame
#

But if you can detect pirated games then why not block them from being used

heady plinth
#

then pirates will know that you know they pirated it, i guess

#

they could improve upon their piratery

limpid fern
knotty wraith
#

Will long range air defense beyond the map create problems?

potent sparrow
cloud sky
#

It is/was called FADE:

*Here's an excerpt from the Wiki page of it and what happens in Arma 2. "FADE will begin by gradually decreasing the accuracy of the player's weapons, making it very difficult to hit a target. FADE might reverse the left/right controls of vehicles, or make them randomly start and stop moving. Eventually it will turn the player into a bird and display a message saying "Good birds do not fly away from this game, you have only yourself to blame"" *

inland dew
#

what does any of this have to do with perf/prof...

empty plover
#

it just hit me, people with pirated arma can use prof to turn it into "non pirated" game?

whole cloud
whole cloud
whole cloud
empty plover
# whole cloud no

oh ok, i wonder whys that, do you still have the protection going in the exe or what

whole cloud
#

keep wondering

empty plover
#

lol, yeah i understand you cant say that... i forgot these things are secret

inland dew
# whole cloud You didn't read the context it seems

i did. but if everybody starts to write/discuss here what's done against cheaters in other games, it has nothing to do with perf/prof section, other than just cluttering it with nothing relevant to it.
that's the only point

feral harness
hoary jolt
whole cloud
#

I am aware

lethal forum
#

hey guys uus
mimalloc still needed with the latest patch?

#

got to say im impressed its so smooth

opal hound
#

Does anyone else experience freezing when initially opening the server browser? I'm not sure if it's because I localhost for testing sometimes and it has put a bunch of interfaces into my recents list which then don't respond. I also tried clearing the cache list a while back but that didn't help
What's the best way to profile a full freeze? (Stopped responding)

heavy vortex
#

You can generate a crash dump from a frozen process using task manager. May or may not be useful to Dedmen.

foggy vine
#

So... how would I send 10 GBs of a full crash dump?

#

Also, got error 500 in the crash reporter logs

inland dew
#

upload somewhere and then link

foggy vine
#

Good point

whole cloud
whole cloud
#

and often the full dump isn't actually needed

foggy vine
#

Safe to just send it here, right?

light cargo
#

eww rar

nocturne obsidian
whole cloud
foggy vine
whole cloud
foggy vine
#

... What gives you that idea?

#

Fixed

feral harness
whole cloud
# foggy vine Fixed

blobdoggoshruggoogly Windows multimedia api
The final crash at the end, I cannot see what that belongs too, its out of range for any of the loaded DLL's. But FreeTrackClient is close to it.
The full dump was not needed here

devenv_fM4JPSZzIh.png
opal hound
foggy vine
#

Thanks

whole cloud
nocturne obsidian
# feral harness even better, means you should be able to make a repro for us

I finally managed to reproduce the issue I encountered yesterday:
https://drive.google.com/file/d/1uM2ckP06acFd6JptuoXMwj_vGGRDuGc5/view?usp=sharing

However, I was able to resolve it using my own method.
I believe (please correct me if I’m wrong) that when the frame rate drops, the engine automatically puts some PhysX objects to sleep.
So after using addForce, I immediately attached a thread to the object that continuously runs _object awake true;, and that solved the problem.

Google Docs
202507151848.mp4
feral harness
nocturne obsidian
# feral harness still need that repro

To reproduce the issue where the object moves back to its original position, it's very simple:
Just create an object and name it c1, then place a few hundred units within a 500-meter radius around it.
After that, run the code I sent you β€” that's exactly how I tested it.

feral harness
nocturne obsidian
feral harness
#

is it physX chair?

#

as in when you place chair above ground in editor, does it fall down on preview?

nocturne obsidian
# feral harness is it physX chair?

The chair was just used to provide a reference position β€” I didn’t apply the addForce function to it.
The objects I applied addForce to were the hundreds of infantry units I placed, as shown in my video.

feral harness
#

that wasnt my question though

nocturne obsidian
feral harness
#

physx engine does not normally work on individual objects but on a scene where multiple objects could be involved at once, hence the specific question

#

try removing the chair and just use position coordinates and see if it behaves the same

nocturne obsidian
nocturne obsidian
nocturne obsidian
# feral harness physx engine does not normally work on individual objects but on a scene where m...

Look, the issue I mentioned earlier doesn't really affect my mod development. Whether it's fixed or not doesn't matter to me β€” it might not even be an Arma 3 engine issue, but rather a PhysX-related one.

However, I’ve discovered a new issue that significantly impacts my ability to make mods. Please watch this video:
https://drive.google.com/file/d/1CygBnlwNcEuAncX3PRmRs9v-SryI4g-D/view?usp=sharing

In the video, you'll see that units close to an explosion are completely unaffected by the addForce function (even if I repeatedly execute _object awake true). Meanwhile, units farther away do react correctly.

After thorough testing, I found the root cause: my script applies addForce immediately after the explosion. Units that are close to the explosion don’t respond at all, while those farther away β€” whose code execution is slightly delayed β€” do respond properly.

In other words, if I insert a short delay (sleep) after the explosion, then all units β€” near or far β€” respond correctly to addForce. What’s even stranger is that if I skip the explosion effects entirely and just apply addForce directly, then all units respond as expected without needing any delay.

Please investigate and fix this issue β€” or at the very least, document the cause and provide a recommended workaround on the official Wiki. Thank you!

Google Docs
202507161502.mp4
feral harness
nocturne obsidian
feral harness
nocturne obsidian
inner elbow
#

where to send profiling crashes

patent sky
inner elbow
#

ah yeah

dusky spire
#

I get a β€˜not responding’ freeze and have to kill the process to exit arma, no prompt though.
Does arma still create a report?

whole cloud
#

It creates and sends a report if you get the popup dialog asking if you want to send it.
I think for freezes we do not generate one

final hearth
#

IDK if its of interest but... Rainbows dont render properly on widescreen (21:9), disappears when on the last 1/4th of the screen - Literally unplayable popcat

fiery crescent
final hearth
#

No gold at the end of them either 😭

knotty wraith
swift drift
#

AIThinkOnlyLocal - it completely disables knowsAbout and Targers for AI's?
And in theory it can be changed mid mission too?

So in theory it should be possible to workaround this, by disabling and enabling it inside the mod, when the command is needed to be executed?

heavy vortex
#

Well, your issue there is that knowledge takes time to accumulate.

#

It's not network synchronized. Each machine is doing its own visibility checks.

swift drift
#

So I need to test and experiment then.
Is just that:
Server - mod that affect AI behavior,
HC - all AI units are on it.

  • clients that have nothing to do with it
tribal remnant
#

Dedmen hi, maybe you can give me a clue?
My dedicated server started to hang spontaneously (iirc, never happened before 2.20, but maybe it is just o coincidence).
I tried 2.20, tried profiling branch with -debug, but the server log doesn't contain anything suspicious.
I can't reproduce it playing locally, only on dedicated.
The server process not crashing, and clients can move across the map freely, but no output for console commands at all.
Is there any way to inspect the server deeper, to understand what could cause it?

weak radish
#

Ded has suggested the following in the past. SFRAME

weak radish
tribal remnant
#

and capture a frame
->
no output for console commands at all

heavy vortex
#

Technically the server only needs to receive the commands, not send anything back. Check if something like diag_log works.

#

There are also various #debug commands (which you'd probably need to run before it breaks) but I never convinced any of those to work.

tribal remnant
#

not send anything back
??
The server returns output for all commands that executed in console for the server itself.

Check if something like diag_log works.
Client scripts works, at least it writes to the client log, not sure about diag_log in the console.
For server - nothing works, it hangs.

heavy vortex
#

You're just guessing that from the lack of response though.

#

In theory server could still be capable of processing incoming messages even if it can't send.

fiery crescent
tribal remnant
#

-bandwidthAlg=2
Tested with and without it with no difference.

#

In theory server could still be capable of processing incoming messages even if it can't send
How can you know it, if you have no signs that the server "capable of processing incoming messages"?
No logs, no server units moving, nothing. In theory of course, I can assume that it still doing something πŸ™‚

heavy vortex
#

by sending a debug console command with a diag_log in it :/

fiery crescent
heavy vortex
#

Nah, client just sits there and waits, IIRC.

tribal remnant
#

by sending a debug console command with a diag_log in it
The server does not respond at all. No logs, no output, no units moving. My scripts/loops contains many logging, and all stops.

I think, that if server hangs completely
If the server process hangs completely - yes. In my case I suppose the network part is still doing its job, but it is beyond my knowledge about the game engine.

My own investigation:
I noticed that often the server log ends with

Waypoint conpleted for ***" // My own marker
16:21:38 Road not found - (3790.758545, 31.094397, 12093.906250)

Comes from the new algorithm I wrote not so long ago for civilian vehicles, the idea was to dynamically add a new waypoint when it finishes the current one, to imitate road traffic, so the function for finding new position was called from setWaypointStatements.
I checked it many times, and it always finds positions correctly, but I remembered that sometimes waypoints could be finished momentary, and in theory it could lead to endless cycle.
I changed the algorithm to use WaypointComplete EH instead of setWaypointStatements, and today all works stable during all tests, while before the server could hang in 50%.

heavy vortex
#

Kinda curious what exactly you wrote to do an unscheduled infinite loop there, but that's a topic for #arma3_scripting

heavy vortex
#

Yeah but Dedmen doesn't like this channel cluttered with things that aren't profiling-branch specific. For what it's worth I don't think there's any way for a user to diagnose an unscheduled recursion hang except by staring at their code.

bold raven
#

Can off topic stuff like this be moved/deleted? It is annoying to browse this channel with this stuff cluttering it up

scarlet jolt
#

Are there any current incompatibilities between stable and profiling?

Running a mission right now and everyone on stable crashes on connect to server and everyone on profiling makes it through

scarlet jolt
#

Seems to be a mission file thing specifically, server is running stable, only reason I bring it up here is because whatever the issue is, only clients on profiling are able to connect

#

I built the mission on profiling

opal hound
#

Pass the crash dumps on to dedmen, there are fixes on profiling for crashes but I don't think any of them are 100%-always-crash

scarlet jolt
#

I'll see what I can get from people

heavy vortex
#

That sounds more like a bug on stable that's been fixed in profiling than an incompatibility.

#

It shouldn't matter what version you created the mission on.

scarlet jolt
#

That was what I was thinking

foggy vine
#

One-time occurrence so far, restart fixed it

restive pilot
#

@upbeat breach plz use the relevant channels

empty goblet
tribal remnant
spring zealot
heavy vortex
#

Did you write anything weird recently? :P

uneven bluff
#

Since munitions were touched recently with the arming distance fixes, I'm hoping that there's some interest in this:
Specifically for projectiles that are ShellCore, there should be a way to reference its corresponding networked objects. For missiles, bombs etc, there is a single machine which they are local to. In the scripting, you can find that machine and force the local machine to triggerAmmo, causing the missile/bomb to explode. Or you can delete them, and this propagates throughout the network.

This does not happen with shells. When you generate a shell, there is a new shell on each machine, and there's no way to reference each other. When you call triggerAmmo, it blows up the one on your local machine, and all the other ones will keep on going. As far as I know, there is no way to find all the other shells on the other machines, other than a hacky workaround where I store all fired munitions in a fired event handler in sequence and use the sequence id to find it. It would be nice if shells get the same system as missiles/bombs.

heavy galleon
uneven bluff
heavy galleon
whole cloud
#

while server is hund, is cpu load high or low?

#

Someone found a way to crash their HC at startup, by malformed command line parameters. I'm impressed

knotty wraith
whole cloud
knotty wraith
#

It's just an extra way to remind yourself.
we changed the server to a new 9800x3d processor and the latest Windows server - it's very interesting to know the reasons

#

I just shouted... if you put a boat with a bot outside the map and tell it to sail to a point on the map, it doesn't sail anywhere... the boat must also be inside the map

#

cam_start is the spawn location of the boat, and the task is to drive to the shore. It doesn't work outside the map, script is old and has always worked

_grp Move markerpos "intro_start";
image.png
#

And that's how it works.

image.png
knotty wraith
#

I've already made the air inside the map spawn

FFA_DISTANCE = _objPos distance2d [worldsize/2,0,0];
// 4,6 - heli, other - air
_spawnpoint = 
if (_type in [4,6]) then {
  [random worldsize, random [abs (FFA_DISTANCE -5000) ,abs (FFA_DISTANCE - 6500),abs (FFA_DISTANCE -8000)]]
 } else {
  [random worldsize, random [abs (FFA_DISTANCE -8000) ,abs (FFA_DISTANCE - 10000),abs (FFA_DISTANCE -12000)]]
};

but resp planes are very close if within the map... immediately launch missiles
thoughts come to spawn without missiles, somehow to determine that AI is identified by the player and only then add missiles

tribal remnant
tribal remnant
spring zealot
#

no idea how that got in there

knotty plinth
#

Is it possible autonomous AI has now sort of reduced behaviours and wont work at all if no player is watching at it? Noticed if im not close as player (also with a extern stallite camera) its just not driving, when i watch at it or im close as player it starts moving/driving again). Not sure if something has been changed, didnt notice this before.

plain trout
#

Is it only happening on perf branch?

heavy vortex
#

Pretty sure someone would have noticed on our servers if AI couldn't drive off-camera. Although there could potentially be localhost vs server differences.

#

AI does randomly fail to drive for no good reason at all, but that's nothing new.

olive silo
empty goblet
empty goblet
tribal remnant
# empty goblet it would be still interesting to know if it's bug in the SHK_pos function or our...

If you're interested, I could try to prepare a use-case this week. But I'm afraid it would end up with SHK_pos functions fixing, and I have doubt anybody here interested in it. Because, as I wrote before, often the server log ends with "Road not found", and SHK functions use nearroads, as well as "while" loops. I catched the moment it hangs once - it was the time the civilian vehicle finished its waypoint, when a new waypoint should be generated. Marker just before the SHK_Pos function call was shown, and the one right after it - wasn't.
Taking in account that SHK_pos function used widely in the mod for a long time (and I test that function in console many times using absolutely the same params I got from my markers in the code), I can assume that something happens when this function called at waypoint completion, both in waypointstatement or WaypointComplete event handler, but this guess is appropriate only if these two call options are similar "under the hood".

spiral pond
#

Often the server log ends with "Road not found"
Something seems to have changed with 2.20 though from what i can tell.
Either its better logging, or some other reason I've seen this more than before (plus the other variant that mentions the position)

#

Road not found - (928.849487, 100.501747, 3485.518555)

lethal trench
#

Following up on hoist hook crashes, we have some dumps now, one on prof.

tribal remnant
#

Something seems to have changed with 2.20 though from what i can tell.
Either its better logging, or some other reason I've seen this more than before (plus the other variant that mentions the position)
Road not found - (928.849487, 100.501747, 3485.518555)
I can reproduce it in a couple of hours maximum, if somebody would need. I just need to be told what to do to get all needed info.

knotty plinth
knotty plinth
knotty plinth
empty goblet
knotty plinth
# empty goblet ye some reliable repro would be helpful

Not sure if we can do a repro with vanilla stuff, probably replacing our Harvester with a UGV and replace our other units and vehicles with vanilla stuff.
Would a video work too and show how it happens/what happens exactly and when it happens?

whole cloud
#

Its in pathfinding, when AI wants to stay on roads only, they will walk through the road network.
I don't know how the error happens though, some connection to a road that doesn't exist, don't know where the connections come from

#

Probably need a repro and then look closely at it

spiral pond
#

Alright thanks. We have a test mission for the road network with logging and 2d map visualization you can also run unattended on a DS.
It checks all roads between different type of keypoints. I will run it again with 2.20 (prof) and see what the rpt brings up this time.
Can try to cross reference then the road not found positions with the custom 2d map "collision" event logging.

For road connections we had used another scripted approach; if not mistaken the internal version, has also additional debugs for the road network (2d and 3d).

#

One more question about that rpt log - is that tied to vehicles, or infantry, or any AI?
If possible, please add the vehicle simulation that triggers it - it may help to narrow it down if its not just about vehicles (carX/tankX).

whole cloud
whole cloud
whole cloud
# lethal trench

that has a 0 byte dump πŸ˜„
Seems stack overflow is still not properly fixed
zombi also has a 0 byte, but a couple older ones too

#

I'd recommend fixing some of the script and other errors spamming the RPT to reduce the noise a bit
Getting those errors in CUP models fixed is probably gonna be hard tho

#

No the other dumps in there are also nothing

whole cloud
whole cloud
#

Btw crash rate is magnificent.
There were about a dozen over weekend with the command line bug fixed in next prof, the others are all on-exit crashes.

firefox_2025-07-22_11-30-47.png
unborn urchin
#
  • Fixed: CarHorn type weapons did not work for Tank type vehicles
    holy shit
#

was wanting that one a long time ago

fickle geyser
#

well, shame we won't be able to fix that in all of SOGPF content anymore.

opal hound
#

Anything notable in the server side stuff? 😊

heady plinth
heady plinth
fickle geyser
#

yes, we don't have any patches planned anymore.

#

We could patch that in Nickel Steel indeed.

#

Maybe would be worth taking a look as the scripted horns are tiny bit buggy sometimes.

fickle geyser
plucky field
final hearth
#

Couldn't find that one, cheers!

charred pagoda
#

I think I shouldn't send this report if it happens while normally closing the game, should i?

image.png
floral socket
#

on another note, is it okay to close the game by alt f4? clicking exit takes too long

weak panther
#

Crash Reporter should open if I press Yes, correct? If click No it will open Crash Reporter anyway, I'm not sure, but it doesn't seem to have opened before on prof in No pressed meowhuh
(not from todays update I seeing this for a few days)

image.png
weak panther
proud mist
# whole cloud

@whole cloud
What p3d fatal error change ?
Most modmakers have now more work to do ?

What is the reason to change this ?

Can this be consider to back in error state ?

quaint flame
patent sky
#

and the crash reporter one shouldβ„’ run first

weak panther
patent sky
#

Yeah that sounds right (its still starting in the background), I thought u meant it only shows up once yes/no is pressed on the other one

thin magnet
proud mist
#

@thin magnet he is from mine unit

#

That's something terrible for most of them you trying do something in Arma 3 when acctualy this game is in good shape. Forgive my reaction for that. But with all respect to you creators is it worth it ? With this you destroy most of the community created mods πŸ˜‚

cold lantern
quaint flame
#

(previously it would be a RPT message, followed by a game crash)
That's for dedmen to clarify then. The changelog says otherwise

whole cloud
empty goblet
opal hound
#

trains

empty goblet
opal hound
#

true

empty goblet
#

but UUV and submarine doing 'honk' ping to get range πŸ™‚ ... someone will for sure recreate Red October scene in Arma 3 then

empty goblet
#

and it's as you were afraid

opal hound
#

what does it look like for players?

proud mist
#

Vanilla back days

#

🫑

empty goblet
#

gulps over the mass revert ahead of him

#

ok, maybe not everyone, i just see 1 player at 1 server somehow playing but most of players fails to connect

whole cloud
indigo anvil
whole cloud
whole cloud
whole cloud
whole cloud