CHANGELOG

I’m editing this post rather than reposting, so this is what changed and when. If you’ve read it before, skip to the first date you haven’t seen. Everything below the changelog is the current state of things.

25 September — “ghost shadows” identified, fault 7 retraction, two new faults

Found what “ghostly shadows” is, and there’s a workaround. In Fullscreen (Windowed) mode, setting a Resolution below your display’s native resolution breaks shadow and light rendering — lit surfaces stop catching light and shadows appear with nothing casting them. In Windowed mode at the same resolution it’s correct. Isolated with one variable, Render Scale held at 100%. New fault 18, and it likely explains why three of us gave three different Shadow Quality thresholds for this — we were on different resolutions and the Shadow Quality number was incidental.

New fault 17: /console gxRestart crashes the client during precipitation. Confirmed by walking out of rain (rebuild works), back into it (crash), repeatedly. Fog is clean — it’s falling precipitation, and it crashes indoors too, so it tracks the zone’s weather state rather than particles in view. This breaks the workaround this whole thread relies on.

Retracting the fault 7 fix. I called it fixed on 24 September and that was half right. Volumetric fog now draws at every Shadow Quality setting — that part is genuinely fixed. God rays only draw at Good and above, which is inverted from the original fault where Low was the only setting that worked. Full sweep in the fault 7 section.

Fault 5 and fault 10 confirmed fixed properly. I spent part of the day convinced fault 5 had returned and I was wrong — that was a measurement error on my part, described in trap 6 below. Water: applied the trigger deliberately at every Shadow Quality setting, cannot reproduce.

The exact-zero test I gave everyone is retired. Nothing on this build measures exact zero any more, so “exact zeros versus merely dark” now returns a false negative on foliage that is unambiguously black. Replacement test in trap 6.

Fault 1, three new measurements: a corruption boundary with the same grass tufts on both sides of one frame (red drops to 36%, green to 67%, blue doesn’t move); severity scaling inversely with light level on one stone surface with a shadow edge; and the cast modulating in real time with a plant’s sway animation, periodic at ~3.5 s. Details in the fault 1 section.

Optional GPU Features: changing it and then rebuilding the renderer crashes the client; before the rebuild, rain renders as diagonal bands. It also does not fix the color corruption, which is what I was testing it for. Possible bearing on fault 3 — see that section.