Fix pen handling for multiple func_breakable_surf skipping glass breaking effect

[sm90x]

Description

Fix bullet penetration through multiple func_breakable_surf panes.

Two fixes:

• Proper penetration handling for glass on both sides (glass texture and nodraw texture)
• Saves the function penetration handling function from stopping too soon and failing to apply shatter effects to glass

Plus I told the ai to not apply any damage reduction with these, but I am not sure what the behavior in the original is. (easily removable if incorrect)

The original fix had some crash happen, I went through the firing range and it didn't crash. I also tested the double shatter multiple times, nothing bad happened.

disclaimer: These changes were authored by AI. I reviewed it best I could, but it's the first time I look at Source(tm) code. The solution makes sense I guess.

AI assisted in finding cause of bug.

Toolchain

• Linux GCC 10 Sniper 3.0

Linked Issues

• fixes Bullet penetrating multiple func_breakable_surf should break them all #1685

[sm90x]

Thanks! I'll see if I can solve that, or at least make it more like ognt. In all versions very shallow angles fail to draw the decal.

[sunzenshen]

While AI generated comments are useful for understanding what the introduced changes were, I would reconsider:

• Is the code is clean enough to be understandable? Perhaps variable names could be adjusted?
• Do the comments just repeat what is happening in the code? It may be better to only keep comments that explain the why of the code.

Generally repeating what the code does in the comments eventually leads to a "the map is not the territory" mismatch, and it's a pain for collaborators that aren't using AI to keep comments up to date. And collaborators that are comfortable with AI can just query an agent to research what the code actually does. Excessively verbose comments also potentially confuse humans and can cause AI agents go into a contradiction resolving rabbit hole, so I would learn from such comments but then try to make them more succinct.

[sunzenshen]

Also I would generally keep the Description field on topic about what this PR does, and to move the methodology explanations and general commentary in the comments section.

[Rainyan]

Here's a test map with 1 unit width windows to the right, and 2 unit windows in the left rows:

ntre_pen_test_map.zip

And some screenshots of the test.

Shooting 1 bullet with the weapon_zr68s at setang 0 90 0; eyeangles:

OGNT

Penetrates eleven 1 unit panes. So I guess 1*11 < MAX_PENETRATION_DEPTH holds, assuming OGNT used the same constant? This is what we would want.

NTRE branch at commit cbcc3b0 (latest commit before this PR)

Penetrates 1 pane, but then the bullet does appear to traverse through the rest? Or at least the bullet decal is drawn.

This PR at commit 1bd5807 (current latest)

Goes through them all. Which is a bit unexpected, considering the issues mentioned above with the bullet decal.

Arguably going through all of the panes is still better than what we currently have in the master branch logic, but ideally we'd want the parity behavior.

[sm90x]

I've been testing and trying this, I noticed the panes in the dev map don't break if you shoot from the other side, only one glass breaks at a time. Is that a map thing or the fix is doing weird things?

happens in ognt huh

[Rainyan]

I can't reproduce only 1 pane breaking at a time in OGNT. For NTRE master branch, I am seeing only 1 pane break, regardless of which side it's shot from.

However, I'm also seeing the strange discrepancy in OGNT for the amount of panes penetrated, based on which side of the pane the shot comes from:

In the above screenshot, the left side has the bullet enter from the glass texture side, resulting in 11 panes of 1 unit thickness penetrated (at least for the ZR68S). Whereas penetrating from the nodraw side in the right-hand side row (which has been flipped around) penetrates only 6 panes.

The func_breakable_surf are built such that there is only 1 surface with the glass texture applied, and all others are nodraw. Here's a map with the right-most row of panes flipped around to expose the nodraw side:

To my understanding this is the standard way to build a func_breakable_surf, and attempting to apply more than 1 glass surface makes the entity not spawn in the level at all.

I have to assume this is a bug in OGNT, and the intention was to penetrate using the glass surface side's penetration value, so that's probably what we'd wanna go with. Although replicating this weird sidedness would be parity, so I suppose that is also an alternative option.

[sm90x]

Alright after reading this better, the ai code was truly garbo and I rewrote most of it, but I still used it's core fix of faking a fraction value.

Two big issues before the fix:

• As suggested above, the nodraw was changing expected behavior of the glass. Since it has no defined value for penetration resistance, the code defaulted to concrete pen resistance on any nodraw side of glass (i verified with devmsgs, the material value when shooting from the nodraw side logs as concrete). This caused the reduced penetration.

  • Fixed by forcing the glass pen resistance after determining it's a shatter-able glass surface. Note, I am not sure if I should have taken into account more materials, and there is also shatter-able tiles which I have no idea what that applies to (ep 1?)
  • It checks also for material == CHAR_TEX_CONCRETE so we can get the same behavior from the nodraw side.

• Because the glass pane disappears when turned into shards, the value of penetrationTrace.fraction after TestPenetrationTrace() is 1.0f and the function shortcuts out because that's the "open air nothing to collide with" case check

  • Fixed by setting the fraction value of the trace to a synthetic value 2.3f, without shortcutting the effect works as expected
  • I used 2.3 because that is the minimum I found to be on parity with OGNT, results in 11 glass panes breaking with ZRC
  • (I accept suggestions to methods to test damage if this is a concern, I tried spawning a target but it didn't work out so well)

Only thing I am not happy with is the bullet holes sometimes refuse to draw when breaking through multiple panes, I couldn't figure that one out.

--

screens (all ntre screns are with fix):

rise test (ntre):

break limit test:
OGNT

NTRE

Six with hole
OGNT

NTRE

Ten with no hole:
OGNT

NTRE

[Rainyan]

Thanks for continuing to work on this.

This looks good to me for the most part, however I left some notes for things that we should fix before merging. Most importantly the NEO preprocessor guards.

---

On another note, I noticed that for some guns we don't quite get the same breakable_surf penetration as OGNT. I don't know if it's because OGNT does something completely different for glass penetration, or whether our penetration values are inaccurate, but in either case I think this is good enough for merge and we can revisit later if needed. But for reference, I tested all of the OGNT guns against the 1 unit pane test map, and here are the results:

Using position: setpos 477.541809 -1622.499756 186.031250;setang 0.000000 90.000000 0.000000;

With the test map: https://github.com/user-attachments/files/25472359/ntre_pen_test_map.zip

All guns penetrate 11 panes, except:

weapon	n panes broken
aa13	5
supa7	8 (For normal shot. Slug round penetrates 11 panes)
srm_s	9
kyla	9
m41l	Untested (gun is disabled in current OGNT)
smac	Untested (gun is disabled in current OGNT)

But again, perhaps out of scope for this PR, so I'm not requesting any more changes for the glass pen values.

[sm90x]

retested seems ok

[Rainyan]

Looks good to me. Thank you!

[sm90x]

Same, thank you for the map and debugging, it was a huge help.