Beta testing 2.12

[Frost_Salamander]

Getting a bunch of warnings on map start, I think related to subtitles?  It happens when I put the AI entity 'atdm:ai_townsfolk_female' in my map.

 

WARNING:Couldn't load SRT file                                                 
'subtitles/lady02//tdm_ai_lady_alertdown_to_idle_heard08.srt' for sound        
'sound/voices/lady02/tdm_ai_lady_alertdown_to_idle_heard08.ogg' according to   
decl 'tdm_ai_lady02'                                                           
WARNING:Couldn't load SRT file                                                 
'subtitles/lady02//tdm_ai_lady_alertdown_to_idle_seen_evidence04.srt' for sound
 'sound/voices/lady02/tdm_ai_lady_alertdown_to_idle_seen_evidence04.ogg'       
according to decl 'tdm_ai_lady02'                                              
WARNING:Couldn't load SRT file 'subtitles/lady02//tdm_ai_lady_idle09.srt' for  
sound 'sound/voices/lady02/tdm_ai_lady_idle09.ogg' according to decl           
'tdm_ai_lady02'                                                                
WARNING:Couldn't load SRT file 'subtitles/lady02//tdm_ai_lady_idle11.srt' for  
sound 'sound/voices/lady02/tdm_ai_lady_idle11.ogg' according to decl           
'tdm_ai_lady02'  

 

[freyk]

On 1/28/2024 at 6:37 PM, nbohr1more said:

( Would be cool if we had an automated tester / screenshot comparison tool like Dolphin Emulator uses to automatically test if engine updates affect visuals in multiple games. )

How about using TDM  automation framework (and maybe pcem/qemu)?

More info see:

https://forums.thedarkmod.com/index.php?/topic/19828-automation-features-and-discussion/

Edited January 31 by freyk

[nbohr1more]

2 hours ago, stgatilov said:

I committed a fix in svn rev 10636.

Let's hope it does not break something else in player movement.

Fix confirmed. No other issues seen thus far ( swimming works, climbing, no hanging in the air when jumping, etc).

[Dragofer]

The new subtitle audio locator widget is nifty but it only seems to update every nth frame, so it looks a little bit laggy. Can we make it per-frame or smoothen it out?

[datiswous]

1 hour ago, freyk said:

How about using TDM  automation framework (and maybe pcem/qemu)?

Yeah I guess you could make a script that loads missions, then teleports to different area's and makes screenshots (actually I made a similar script for envshot a while back), but then you have to review all those screenshots?

Maybe just a script that changes calc in something else and don't look back (maybe this is dumb)?

[stgatilov]

8 hours ago, datiswous said:

Yeah I guess you could make a script that loads missions, then teleports to different area's and makes screenshots (actually I made a similar script for envshot a while back), but then you have to review all those screenshots?

Yeah, I wanted to implement some kind of testing framework on top of automation but did not.

[stgatilov]

11 hours ago, Frost_Salamander said:

Getting a bunch of warnings on map start, I think related to subtitles?  It happens when I put the AI entity 'atdm:ai_townsfolk_female' in my map.

I deleted double-slashes in svn rev 16957.
I think they were causing the issue.

I believe it only happens on Linux, Windows side somehow reduces the double-slash to single directory separator.

[stgatilov]

12 hours ago, Daft Mugi said:

I found the setting that causes it. The glitch happens when anti-aliasing (r_multiSamples) is enabled.

EDIT: Bisecting shows that rev 10384 is the first commit to introduce this glitch.

Antialiasing mostly affects some backend logic, while the commit changed data structures in frontend. It is very hard to believe they are linked in any way.

Maybe the bug is not reliably reproducible?
Maybe antialiasing is an important condition when it happens, but otherwise it is timing-specific, so accelerating frontend made it pop up. Or maybe antialiasing is just one way to slow gown backend and using maximum soft shadow quality or r_fboResolution 2 would make it happen too.

Maybe the particular behavior is driver-dependent (e.g. NVIDIA masks the error away).
Which GPU do you have? Which drivers do you use?
Which CPU do you have BTW?

I cannot reproduce it myself, even on Linux VM.
Does it happen in the very specific viewpos and goes away if you move away slightly?
Maybe someone else can reproduce it?

[nbohr1more]

3 hours ago, stgatilov said:

I cannot reproduce it myself, even on Linux VM.

Does it happen in the very specific viewpos and goes away if you move away slightly?
Maybe someone else can reproduce it?

Also unable to reproduce, Linux Mint 21.3 Virginia. Nvidia Geforce RTX 2060 latest non-free 535 drivers.

[Frost_Salamander]

10 hours ago, stgatilov said:

I believe it only happens on Linux, Windows side somehow reduces the double-slash to single directory separator.

I was actually using Windows - so maybe not?

[nbohr1more]

I don't think this counts as a bug but maybe it's something to look at...

Nvidia GPU, Intel CPU

Linux: Scroll of Remembrance - Shadowmaps ( size 1440 ) 70+ FPS / Stencil 58+ FPS

Windows: Shadowmaps ( size 1440 ) 52+ FPS / Stencil 30+ FPS

I am guessing we can mostly attribute this to Windows CPU usage bloat. Can any other dual-booters confirm?

Edit:

I went back to check my Windows security "Exploit Protections" area for "TheDarkModx64.exe" to see if it still existed and still had most of the protections disabled and found some new protections were added. I disabled "Heap Protection" and:

Shadowmaps 58+ ( mostly 60+ ) / Stencil 38+ ( mostly 40+ )

( All the FPS are at the start of the mission facing the manor, eg worst case scenario ).

[lowenz]

Windows or some other programs in background?

[Daft Mugi]

On 2/1/2024 at 2:24 AM, stgatilov said:

Antialiasing mostly affects some backend logic, while the commit changed data structures in frontend. It is very hard to believe they are linked in any way.

Yeah, I don't know what to say. Rev 10383 works fine and 10384 does not.

On 2/1/2024 at 2:24 AM, stgatilov said:

Maybe the bug is not reliably reproducible?

It is always reproducible for me, meaning that every time I try to reproduce it, I can reproduce it.

On 2/1/2024 at 2:24 AM, stgatilov said:

Maybe antialiasing is an important condition when it happens, but otherwise it is timing-specific, so accelerating frontend made it pop up. Or maybe antialiasing is just one way to slow gown backend and using maximum soft shadow quality or r_fboResolution 2 would make it happen too.

After some more testing, I found that r_multiSamples set at 2 or 4 has the glitch most often at that viewpos. When set to 8 or 16, it is still possible but less likely.

Setting r_fboResolution 2 made no difference.

If com_maxFPS is set to something lower, such as 30, the glitched frames display for a longer period of time.

On 2/1/2024 at 2:24 AM, stgatilov said:

Maybe the particular behavior is driver-dependent (e.g. NVIDIA masks the error away).
Which GPU do you have? Which drivers do you use?
Which CPU do you have BTW?

Yeah, I also wonder if it driver dependent or GPU dependent. I discovered the glitch while using graphics drivers Mesa 22.0.5, so I updated my OS and graphics drivers to Mesa 23.3.2 today as a test, and the problem is still there. It's strange that rev 10384 exposed this problem.

My machine specs and settings:

• Linux, Ubuntu 22.04
• AMD Ryzen 9 5900X
• AMD Radeon RX 6700 XT (Mesa 22.0.5 & Mesa 23.3.2 both exhibit the glitch)
• com_fixedTic 1
• com_maxFPS 60
• Anti-Aliasing 2x (r_multiSamples 2)

On 2/1/2024 at 2:24 AM, stgatilov said:

Does it happen in the very specific viewpos and goes away if you move away slightly?

The video I shared shows the viewpos. Even at the same viewpos, the glitched frame will pop in and out. I've only seen this glitch happen in this single room in Accountant 1. If I move around the room at that viewpos, I can find other view positions that also expose the glitch.

Here's another viewpos for Anti-Aliasing 4x, 8x, and 16x. It requires moving the mouse slightly to get into the exact viewpos to expose the glitch. (1.7 -166.4 0.0) is not precise enough.

-1460.29 1425.48 172.25 1.7 -166.4 0.0

Glitched frame (Shows the curtain texture, covering most of the frame. Also, a hint of the red carpet texture?):

Good frame (Same viewpos and Anti-Aliasing settings, when the glitched frame disappeared for a moment.):

 

(Before and after rev 10384) I wonder if this might also be related to:

 

(Before and after rev 10384) Also, sometimes on the loading screen with Anti-Aliasing enabled, I get a glitched background, which does not show with Anti-Aliasing disabled. See left side of screenshot:

 

I don't know if the last two examples (distorted first frame on launch & loading screen distorted background) are related, but I thought I'd mention them just in case it helps track down the issue.

[thebigh]

In case you're still tracking and fixing light leaks, I found the following in Poets and Peasants:

 

 

The first picture is dev16814-10408. The second picture is dev16809-10394. Something must have changed between those two devs.

[Daft Mugi]

Regarding worldspawn "forceAllShadowsBehindOpaque" set to "1", it doesn't seem to get restored on game load.

To reproduce (works with any fixed mission using "forceAllShadowsBehindOpaque"):

1. Add "forceAllShadowsBehindOpaque" "1" to the worldspawn in the map file.
2. From main menu, "Start this Mission" or "Restart Mission".
3. setviewpos
4. Notice that the light leak is fixed (not present).
5. Save game.
6. Load game.
7. Notice that the light leak is present (not fixed).

Example

1. From main menu, "Start this Mission" or "Restart Mission".
2. Run setviewpos.
3. Save game.

4. Load game.

[thebigh]

9 hours ago, Daft Mugi said:

Glitched frame (Shows the curtain texture, covering most of the frame. Also, a hint of the red carpet texture?):

 

I'm seeing the same thing in Dragons Claw, a big chunk of skybox right here:

 

[Baal]

Playing Tears of St. Lucia, I noticed really bad framestutter in the church. It seems to happen around the builder guards. The game starts to micro freeze at a short interval (every second roughly). I don't know if this a beta or a mission issue.

[stgatilov]

10 hours ago, thebigh said:

I'm seeing the same thing in Dragons Claw, a big chunk of skybox right here:

I can reproduce this one.
Enable noclip and execute: "setviewpos 207 -1212 -408".
Tiny movement by Y and you get proper view, but here exactly at visportal it gets closed.

@Daft Mugi, does it look like the issue you have?
I bet it does not.

[stgatilov]

I looked closer at the problem in Dragon's Claw, and this is caused by some kind of precision issue in dmap.

The .proc file contains imprecise BSP plane:

/* node 18377 */ ( 0.0399680398 1 0 1202.8780517578 ) -38 18378

And also there is a visportal stored there:

/* iap 62 */ 4 37 50 ( -96 -1200 -488 ) ( 304 -1216 -488 ) ( 304 -1216 300 ) ( -96 -1200 300 )

The game loads the visportal polygon and computes portal plane from it, which is:

plane	(a=-0.0399680398, b=-0.999200940, c=0.00000000, d=-1202.87805)

As you see, the plane normals are different by 0.2%.
The visportal plane computed from polygon is almost perfectly unit, but the BSP plane is not unit (length slightly greater than unit). This makes the game think it is on the left of visportal in one place, but on the right of visportal in another place.

I tried to re-dmap but the data is the same.
I'll try to see what is going on.
Indeed, making the portal axis-aligned would have most likely fixed the issue.

[Daft Mugi]

32 minutes ago, stgatilov said:

@Daft Mugi, does it look like the issue you have?
I bet it does not.

It's a different issue.

[Daft Mugi]

14 hours ago, Daft Mugi said:

Regarding worldspawn "forceAllShadowsBehindOpaque" set to "1", it doesn't seem to get restored on game load.

@stgatilov Thank you for fixing this!

[stgatilov]

Ok, here is the source of the goddamn precision issue:

static void WriteFloat( idFile *f, float v )
{
	if ( idMath::Fabs(v - idMath::Round(v)) < 0.001 ) {
		f->WriteFloatString( "%i ", (int)idMath::Round(v) );
	}
	else {
		f->WriteFloatString( "%f ", v );
	}
}

I guess I'll never stop finding more weird tolerance stuff in this codebase

Anyway, I'll fix this after 2.12: never know the consequences.

UPDATE: Here is issue in bugtracker 6480

[stgatilov]

beta212-06 is released.

[nbohr1more]

New fresnel mod:

https://www.moddb.com/mods/the-dark-mod/addons/fresnel-mod-212-beta-3

Probably not needed. I don't think anything in the previous base pk4 changed but I built it just in case.

[JackFarmer]

I cannot load saves/quicksave in one of my project maps anymore. When starting quickload/load the mission loads a tiny bit, but crashes to desktop then.

Can't say whether this was already a problem with the previous beta versions.

Checked all other WIPs and it works fine there.

qconsole.log Darkmod.cfg