I still reproduce the bug on master.

I believe this is fixed?

I still reproduce it on Mesa radeonsi driver.

could you update the root post to make it explicit which hardware you use? This seems to be important for that bug.

[edit]
I added the GPU-Radeon tag

I've been on Nvidia when I reported the issue.

Interesting. I'd be surprised if only intel would not have it, so there must be something else causing the issue.

Oh, nevermind! This hasn't happened to me; I just moved the issue here!

We already have a GPU-AMD label so I replaced GPU-radeon with it (Radeon = AMD). I'm not sure this bug is Radeon-specific though.

To reproduce the bug, one should enable dynamic lighting (r_dynamicLight 1).

The test map is a map done by @FreeSlave, who is known to use J.A.C.K. editor, which only supports a map format upstream NetRadiant doesn't support yet.

It means the map was compiled with tools we don't use ourselves, like @FreeSlave's own branch there: https://github.com/FreeSlave/GtkRadiant/tree/map220_netradiant

So it's also possible the map compiler that was used had a bug and the produced BSP file is buggy, but I don't know how to verify it.

To my eyes this looks much more like an issue with our real time shadows though. Possibly a NaN/DBZ issue?

Our tiled renderer does not support real time shadows yet (and no one is working on them). So it's very unlikely a bug is in real time shadow: they're expected to not be rendered at all.

The forward renderer which had real time shadow code (feature I never seen working correctly anyway) is now deprecated and disabled by default. See also:

• drop forward renderer, focus on tiled one #236

This looks like the same issue as the incorrect dynamic light in one of the basement rooms in https://users.unvanquished.net/~sweet/pkg/map-hq-beta28_0+0.dpk.

Removing the AMD label because this is either an engine or map bug.

It apparently comes from here: https://github.com/Unvanquished/Unvanquished/blob/6b8fdb8f6030ce6a8266200855e2d2729178f521/src/cgame/cg_ents.cpp#L308.

Apparently this is indeed the same issue as on hq-beta28, and is caused by dynamic lights.

I've looked into it and apparently in both cases the lights have way too colours and large radius:
hq-beta28:

While tremtest has 2 similar lights, but with radius 300

For comparison, dynamic lights produced by the pulse rifle have colour {1.49, 1.49, 1.49} and radius 25.

Here's how this looks on hq-beta28: