[r3978] Part of wall disappears : a single line in udmf

[cocka]

There is a bug in this wad:

kommentek.wad

(15.37 KiB) Downloaded 29 times

(map for Hexen in UDMF)

The problem is when you step on the perpetual platform and it starts working the left side of the platform (south wall) disappears as you walk along beside it. I guess it has something to do with that line in the upper right corner of the room, but strangely this glitch won't happen when you convert the whole map into hexen format.

The other problem is when you step on that platform it works weirdly. Sometimes it raises immediately when you cross the east line but in other cases it just clicks and you have to cross the line again so that the platform can start working. You can even get stuck under the platform if you run/jump up and down.

This bug is strange so much the more because the wad doesn't contain any scripts. Just two action specials.

Thanks for your reply.

[ChronoSeth]

What happens when you rebuild the map's nodes?

[cocka]

The result remains the same. Though, in Gzdoom the glitch changes a bit because the wall doesn't disappear, but when you look upwards parts of the wall graphics will be messed up. That's why I think this must be a bug.

[Blue Shadow]

You can even get stuck under the platform if you run/jump up and down.

The problem is with the nodes, as ChronoSeth mentioned. Start the map with [wiki=CVARs:Configuration#gennodes]gennodes[/wiki] enabled and you'll see that issue has gone.

[cocka]

So are you telling me that every time I load the map I should use a console command and reload the map to correct the glitch?

[Blue Shadow]

No. What I was saying is the problem is with the nodes and running it with gennodes enabled is the proof that it is.

There's an alternative: Delete the nodes from the map (use SLADE) and don't build them yourself. ZDoom will do it for you when you start the map. The downside to this, however, is that the map will take longer to load/start the larger it gets.

[cocka]

OK, but the question is: why does zdbsp not work properly? If there is a routine in zdoom for building the nodes perfectly, why does zdbsp not use that routine? And why does it produce glitches by building nodes?

[Enjay]

Just confirming that when I build the nodes manually from the command prompt using my usual settings ( -x -z -w -b -r ) or build them via DB2, the glitch persists but gennodes 1 removes the glitch. As to why this is happening, I have no idea.

[cocka]

but gennodes 1 removes the glitch.

If you "convert" it to hexen format, the glitch will be gone too.

As to why this is happening, I have no idea.

Obviously, this is a bug. I wouldn't have written it here, if there were no problem at all.

I think it has something to do with the vertices using fractional numbers for their coordinates.

[Enjay]

I think it has something to do with the vertices using fractional numbers for their coordinates.

I think you're right. I just used DB2 to move the few vertices that were using fractional coordinates to vertices to whole number coordinates then saved the map. On testing it, the problem had gone.

[cocka]

to move the few vertices that were using fractional coordinates to vertices to whole number coordinates

OK, but this is the circumvention of the problem and the fact remains that it is a bug.

[ChronoSeth]

I'm completely dumbfounded as to why zdoom's internal nodebuilder produces different (and better) results than ZDBSP. Aren't they (almost) exactly the same thing?

[randi]

I have reconfirmed that, aside from off-by-one-bit differences when loading fractional vertex data, the internal node builder and ZDBSP are, indeed, generating absolutely identical nodes. There's just one problem: None of the node formats support node splitters with fractional coordinates, so ZDBSP cannot save its nodes exactly as generated. Oops.

[Xaser]

That could indeed explain why I was never able to get satisfactory results out of ZDBSP for ZDCMP2. If that's fixable, that'd make me a happy Xaser.

[randi]

Fixed in r3980 (ZDoom) and r3981 (ZDSP). I will try to get an updated version of ZDBSP up within 24 hours after I'm satisfied that the introduction of a new node format didn't break the writing of any of the existing formats.