myconid

https://github.com/myconid/0ad/tree/SVN_debug Those errors & the black squares should be fixed in this branch (needs recompile). Let me know if that's the case and I'll commit the changes to SVN.

Sounds like a blending issue. I'll move the code back to git so I have more freedom to test things out, and I'll see what this is about. That's definitely a decal trying to use a model shader instead of a terrain shader. The error should appear only when the problematic decal is trying to draw itself to the screen, so you can probably find out what's causing it.

(I assume you're trying to do only normalmapping? So are you using a material that only does normalmapping or one that also tries to do parallax, ao, specular etc? Create a new material to suit your needs, it's really easy.) (btw the crashing as a result of misconfiguration predates my changes, should be tackled eventually) (now I REALLY need to get to bed)

zoot, all "decal" actors must use terrain materials (e.g. "terrain_base.xml"). I changed all that I could find, so I guess he committed the rocks really recently and I didn't see them. I'll put up a page on the wiki tomorrow explaining the changes I've made. General, that might be a bug in the texture hotloading code. See if correcting the decal fixes it, though. Anyway, I'm off to bed, we'll continue the debugging tomorrow.

Did you follow the instructions to enable them? ^

The patches have now been committed and there are Windows binaries available on SVN (thanks to fcxSanya for the build). Anyone interested, please pull the files and test! To see the effects you need to make four little changes in your config first: preferglsl = true gentangents = true smoothlos = true materialmgr.quality = 10 Here's a few tests you can try: Modelmapping: Look at Roman CC and Mars Temple 2 models for AO/parallax/normal/specular/emissive mapping. Look at windy trees when in the game and in Atlas after pressing "Play". Smoothlos: Los should move smoothly around units. It should toggle correctly from the dev console. It should not be visible at all in Atlas. Terrainmapping: "Biome alpine/Alpine cliff" uses triplanar, normal and specular mapping. "Biome desert/Desert city tile" uses a custom alphamap. "Grass/Grass1 spring fancy" uses the special grass material. "Snow/Snow grass 2" uses normal and specular mapping. Heightmap: In atlas, File/Import heightmap. Choose an image file (not too huge!). In the first three, also check if ARB and Fixed modes work as expected (there should be no visible changes, with the exception of alphamaps). Should you encounter any bugs, please post them here.

I didn't change every xml file. I added the xml files. Terrains didn't use materials at all before now.

Nope. It's more expensive, so I'm leaving that up to the artists/mapmakers.

DONE!

Yes, I completely forgot about that! Well, too late now... I can only apologise in advance for the moderately large download. Flute's test error has also been fixed. So now, last thing: the heightmap importing. This should be considerably easier.

Wow, I keep forgetting why I dislike Subversion so much. Unlike git, it can't even detect when files were moved, so now it's re-uploading a bunch of textures. Ugh. Oh no. It's.. re-uploading all the terrain textures?! This could take a while.

About howlingflute's error: The LOSTexture code needs access to g_Renderer variable, which is global. However, in the test files the renderer isn't instantiated anywhere, hence the error when the value is accessed. Not sure if the renderer object can be instantiated on its own, just to please this test. I'll see what can be done, but I'll ask people to be a little patient... Edit: OK!

Committing more stuff now. Sorry for taking so long, I'm making sure all the xml files are up to date.

Thanks, howlingflute. I'll look into that error (it's probably nothing major). There's been some discussion about the preferglsl/gentangents before. While I believe the gentangents option should be kept separate, I'll do something about the crash. Loki1950, I'm sure people have already tested it on Windows before. Did you run update-workspaces before compiling? Edit: yeah, what historic_bruno said. Edit2: howlingflute, the roman CC should display the effects in GLSL mode. Just committed a simple fix for the crashing - the "quality" setting is now set to 0 by default, so it needs to be set to 10 to display the effects. Regarding the error, can you please post gdb output? Found the issue (doh!).

Yes, you need to add a second UV set that doesn't overlap itself anywhere. It's really easy to do in Blender, and we should eventually automate it with a Python script or something. Btw, smoothlos has been committed. I'll do the other two patches tomorrow.

You need a new material. Really, the actor editor should be modified to do that for you.

I do! They should probably go somewhere under the "lib" directory... I must admit I haven't spent any time thinking about those things actually... and now that you mention it, there might be a (small) file or two without the usual licensing blurb at the top. Edit: The "lib" directory not be the right place. Plus, that update-workspaces script is driving me up the wall.

It's kind of like landing a rover on Mars: the engineers test their creations and you get to look at rocks.

Modelmapping has been committed to SVN. Fingers crossed.

Probably not. It's definitely something to think about in the future.

Generally speaking they kinda do, but there is one important difference. Conditional defines are executed dynamically and (in the current implementation at least) re-evaluated by the renderer on each frame. The overhead is pretty minuscule, of course, but it's still there. OTOH, quality settings are evaluated only once, statically, as the materials are being loaded.

Then you'd use the "conditional_define" system I've added, which lets you set options for individual effects through the config.

Well, yes. That would be the plan!

Thanks for testing! The first error is because you used smoothlos=true without GLSL support, when there is no equivalent ARB shader available (yet). Just like the GPUskinning option, it's an experimental feature that is marked "requires GLSL". That is, the crashing is by design, as the ARB part of the equation is still missing. As for the second error, it's a real bug, but I think I know how to fix it.

Consider this example. Say the config setting is at 0 and the model definition points to Material 1. Material 1: <blah> <alternative (Material 2) quality=7/> ... </blah> If config < 7, jump to Material 2. Since config == 0, jump. Material 2: <blah> <alternative (Material 3) quality=3/> ... </blah> If config < 3, jump to Material 3. Since config == 0, jump. Material 3: <blah> ... </blah> There is no alternative, so this material is used iff config >= 0, so pick this one (this is the stopping condition for the recursion). So, Materials 1 and 2 are ignored, and we load Material 3. If config was 4, it would pick Material 2. If config was 9, it would pick Material 1. It's also possible to concentrate all the conditions in the same place to avoid the recursion, though obviously if a model starts off at Material 2 or 3, this system wouldn't work*. Material 1: <blah> <alternative (Material 3) quality=3/> <alternative (Material 2) quality=7/> ... </blah> This means if config < 3 jump to Material 3, else if config < 7 jump to Material 2, else pick Material 1 for anything >= 7. (I guess this is ascending order, not descending... ). There's no need for "alternative" definitions in Materials 2 & 3, though they would help with this* problem, but then we're duplicating stuff.