|
|
|
[
Permlink
| « Hide
]
Dessie Linden added a comment - 07/Mar/09 02:57 PM
Vector - Thank you for logging this bug and including the excellent snapshot. When you get the chance, please identify the SL Viewer version you're running and add that information to the Affects Version/s field. If you're running a recent RC viewer, please also let us know whether or not shadows appeared correctly in prior Release Candidates. Thanks!
Hey Dessie, thanks for looking at this. Sorry, I didn't see the "source code" option under Release version, I think that's as close as I can get, because this is the trunk, not even RC yet. My understanding is it's the basis of what's likely to be in 1.23.
To my knowledge, Shadows will be new in 1.23 and doesn't exist even in the RC 1.22 yet. Those who are working on the trunk might want to know about it. Cheers. Just to add that this is indeed an issue with the Release Candidate 1.23.2.
Another possible contributing factor is that the combination of animations and height parameters for the avatar on the right (with only prim shadows) causes him tio hover a few centimeters above the ground, while the female avatar on the left always has her feet grounded. Note however, that this lack of body shadows continues when flying. Partially (see attached shadow.jpg) confirmed on Linux too, 1.23.2 RC
By the way it happens even without playing animations. Second Life 1.23.2 (120719) May 18 2009 12:02:26 (Developer) I've seen this before, though I'm getting slightly different symptoms right now... For me right now, the shadows develop holes when I have on clothing that has alpha on it, and the transparent places end up not getting shadowed. Rebaking seems to work around it (at least until the next sun/moon update).
It's possible what I'm seeing isn't quite the same problem, though. Second Life 1.23.3 (122255) May 30 2009 13:13:41 (Second Life Release Candidate)
Release Notes Built with MSVC version 1400 You are at 237346.7, 253664.9, 26.1 in Dreams located at sim2573.agni.lindenlab.com (216.82.18.70:12035) CPU: AMD (Unknown model) (2599 MHz) libcurl Version: libcurl/7.18.1 OpenSSL/0.9.8j zlib/1.2.3 Are you seeing the same issue with 1.23.4 ? I want to rule out any self-built viewer issues. That will help the graphics team handle this better.
Yes it does (and also in snowglobe).
However, it doesn't only happen with lots of prims attached. See picture "no prims no body". My friend there is not wearing any attachments whatsoever, but an important body part is missing in the shadow. When he takes off his shirt, his shadow is complete again. Texture rebake doesn't help in my case... In case of missing body parts when wearing prims, taking off a shirt (or something) for a moment shadow is complete, before it gets back to the state where parts are missing. Thanks, I've moved it into the general post-1.23 pool
The above is wrong, see correction below. To add to Adeon's comment.. Skin does not leave shadows fine.. A dark colored skin for example won't cast a shadow.
@Evo: Thanks, hadn't tested a darker skin. So apparently the final mesh is looked at as monochrome, and 50% gray or darker gets counted as transparent.
Occurs on all Just another one who complains about it.
Second Life 1.23.5 (136274) Oct 14 2009 13:31:00 (Second Life Release Candidate) Built with GCC version 40102 Sie befinden sich in 267430.3, 271588.0, 25.1 in Berkman located at sim968.agni.lindenlab.com (216.82.13.118:13000) CPU: Intel(R) Core(TM)2 Duo CPU E8400 @ 3.00GHz libcurl Version: libcurl/7.16.4 OpenSSL/0.9.7c zlib/1.2.3 c-ares/1.4.0 I recently upgraded my graphics card to an Nvidia 275 GTX from an 8800. I definitely don't think this is a hardware issue, and is definitely an issue with the alpha sorting. With it, I've also noticed that if you rebake textures, you will briefly be able to see your full mesh's shadow until something happens causing it to no longer make a shadow. Really hurts the immersion that shadows create when you turn into a shadow ghost...
I'm also consistently experiencing this on all versions of snowglobe, including the trunk 1.2 versions. Stats: Built with MSVC version 1400 You are at 263841.9, 279219.6, 57.1 in Eder D Uru located at sim1063.agni.lindenlab.com (216.82.13.213:13002) CPU: AMD (Unknown model) (2999 MHz) Just to the comment that this occurs on all known viewers, Kirsten's S18 has fixed it. There are other bugs, but the basic alpha sorting problems have been licked, so it's solvable.
From what I see, if there's any layer of the body texture that has been uploaded with transparency, the avatar body will not cast a shadow. This may include skin textures that are fully opaque but are saved with an alpha channel (that is fully opaque), or maybe just textures that use transparency.
@Vector: I reproduced this in Kristen's S18205's. Can you confirm the bug's gone for you in S18205? Remember that it will always display correctly before the skin has 100% loaded in.
Updated name of Jira title to accurately reflect the specific cause of the bug.
See attached screenshot: Reproduced bug by creating shades of greyscale skins and seeing which cast shadows and which do not. I'm curious if anyone has looked at the Hardware Options in Kristen's viewer, it includes the "RenderUseFBO and Render Deferred" UI options and produces a completely different result; very soft shadow outline (practically a blob like the feet make under the normal viewer), but it works as expected for all I can tell. The postprocessing in Kirstin's is rather more radical and everything looks very soft a toned in comparison to non Deferred rendering. I normally use Snowglobe and LL viewers including RC, which change the lighting slightly under deferred rendering and exhibit the same problems as describe by Adeon and others. BTW Kirsten's FPS drops to a 1/6 eg from 36 to 6 under the same conditions when deferred is switched on on my laptop and curious errors are listed too, see fragment from the log below.
_________________________________________________________ 2009-10-31T02:39:04Z INFO: display_stats: FPS: 37.50 2009-10-31T02:39:07Z INFO: LLAppViewer::idle: Kills on unknown objects: 8 2009-10-31T02:39:12Z INFO: LLAppViewer::idle: Kills on unknown objects: 8 2009-10-31T02:39:14Z INFO: display_stats: FPS: 39.30 2009-10-31T02:39:15Z INFO: LLViewerShaderMgr::setShaders: ~~~~~~~~~~~~~~~~~~ Loading Shaders: ~~~~~~~~~~~~~~~~~~ 2009-10-31T02:39:15Z DEBUG: LLShaderMgr::loadShaderFile: Loading shader file: windlight/atmosphericsVarsV.glsl class 2 2009-10-31T02:39:15Z DEBUG: LLShaderMgr::loadShaderFile: Looking in C:\Program Files (x86)\Kirstens S18-1 (211)\app_settings\shaders/class2/windlight/atmosphericsVarsV.glsl 2009-10-31T02:39:15Z INFO: LLShaderMgr::loadShaderFile: Loading file: shaders/class2/windlight/atmosphericsVarsV.glsl (Want class 2) 2009-10-31T02:39:15Z DEBUG: LLShaderMgr::loadShaderFile: Loading shader file: windlight/atmosphericsHelpersV.glsl class 2 .............. (edited out).................... 0(8) : warning C7532: global type sampler2DRect requires "#version 140" or later 2009-10-31T02:39:16Z DEBUG: LLGLSLShader::mapUniform: Uniform diffuseMap is at location 0 _________________________________________________________ CPU: Intel Pentium III/Pentium III Xeon (0.25 micron process) with external L2 cache (2527 MHz) Rewrote jira entry based on newest information and included repro.
Thank you Adeon for paying such close attention to this!
Don't pictures 3 and 5 suggest coloration of a clothing layer is also at issue? Yes - most colors when viewed in black and white are dark grey, so they aren't past the 50% grey mark.
I don't know the exact reason for why, but from testing different things, for whatever reason the final baked mesh is converted to black and white, and then anything lighter than 50% grey leaves a shadow. So, for example, Obscure's dark green shirt turns into something like 25% grey (1/4 white 3/4 black) when viewed in black and white, which just is darker than middle grey, so that part of his mesh has no shadows. Very odd that this happens, but that's what the patterns seem to show. Now if anyone can actually take a look at the code to see WHY that's happening, that would be wonderful. PS: I compiled the render-pipeline branch, and this bug doesn't exist. But, that is 6 months old, so I can't tell wither this is "Already fixed" or that was just before the bug was introduced. |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||