> Something has changed in the environment manager which means that clouds
> generateion is now inconsistent. I'm still tracking it down, as my recent
> changes shouldn't have affected this.
Well, the cause was a bug in my code, but it didn't expose itself until we moved
to multiple cameras. The attached patch fixes the problem.
I've also put in a new heuristic to improve the frame-rate. Clouds that are already sorted
are likely to still be sorted in subsequent frames. Therefore I've put in a back-off
mechanism for the bubble-sort pass. This should mean that if you stay completely
stationary, once the clouds become sorted they will eventually only perform a
bubble sort pass every 128 frames.
> Warning: detected OpenGL error 'valeur non valide' after RenderBin::draw(,)
Fixed in the patch below. For some reason the shader didn't like index 16 being used...
The patch also fixes the chequer-board effect that was causing very sparse cloud cover.
Attached is a small patch for 3D clouds.
It provide the following:
1) Proper spherical distribution of sprites (previously they were distributed cylindrically - whoops)
2) Better shading, so the bottom of the cloud is darker than the top.
3) Fixed a couple of texture sizing bugs.
Attached is a small fix for the sorting in CloudShaderGeometry.cxx.
I think the sorting problem stems from the osg idiosyncracy
to store transposed matrices...so the intuitive
osg::Vec4f p = vm * osg::Vec4f(_cloudsprites[i]->position.osg(), 1.0f);
needs to be replaced with...
osg::Vec4f p = vm.preMult(osg::Vec4f(_cloudsprites[i]->position.osg(), 1.0f);
The patch also optimizes the distance calculation - it evaluates the distances
in model space instead of eye space, which reduces computation to a dot-
product instead of a matrix multiplication.
It fixes the following issues (to a greater or lesser extent):
1) Performance. Quad trees used to improve culling, and the sprites are placed on the surface of a sphere rather than
randomly throughout the cloud, requiring fewer textures. This saves about 5-10fps on my machine.
2) Disabled 3D clouds have no performance impact. Previously they were still in the scenegraph. Now they are removed.
3) Clouds are now loaded on start-up, and don't require the scenario to be changed, they also work with METAR.
4) The cloud field is shifted as you travel. There's a small bug in that the clouds "jump" as you reach the edge of the field.
5) Iterative sorting of sprites. This doesn't appear to solve the alpha blending problem completely, but may help a bit.
so hopefully not too much breaks). New syntax features:
1. Call-by-name function arguments. You can specify a hash literal in
place of ordered function arguments, and it will become the local
variable namespace for the called function, making functions with many
arguments more readable. Ex:
view_manager.lookat(heading:180, pitch:20, roll:0, x:X0, y:Y0, z:Z0,
time:now, fov:55);
Declared arguments are checked and defaulted as would be expected:
it's an error if you fail to pass a value for an undefaulted argument,
missing default arguments get assigned, and any rest parameter
(e.g. "func(a,b=2,rest...){}") will be assigned with an empty vector.
2. Vector slicing. Vectors (lists) can now be created from others
using an ordered list of indexes and ranges. For example:
var v1 = ["a","b","c","d","e"]
var v2 = v1[3,2]; # == ["d","c"];
var v3 = v1[1:3]; # i.e. range from 1 to 3: ["b","c","d"];
var v4 = v1[1:]; # no value means "to the end": ["b","c","d","e"]
var i = 2;
var v5 = v1[i]; # runtime expressions are fine: ["c"]
var v6 = v1[-2,-1]; # negative indexes are relative to end: ["d","e"]
The range values can be computed at runtime (e.g. i=1; v5=v1[i:]).
Negative indices work the same way the do with the vector functions
(-1 is the last element, -2 is 2nd to last, etc...).
3. Multi-assignment expressions. You can assign more than one
variable (or lvalue) at a time by putting them in a parenthesized
list:
(var a, var b) = (1, 2);
var (a, b) = (1, 2); # Shorthand for (var a, var b)
(var a, v[0], obj.field) = (1,2,3) # Any assignable lvalue works
var color = [1, 1, 0.5];
var (r, g, b) = color; # works with runtime vectors too
This eliminates jitter and other rendering problems.
For the moment this is dependent on an osg fix.
Also, don't read wind properties from FlightGear; provide a mechanism
for fg to set the wind.
It is not safe to call this function from the database pager thread;
in any event, state sets and textures created in the database pager
will get passed through the SharedStateManager anyway.
- this exposed a bizarre issue on Mac where dragging in <AGL/agl.h> in
extensions.hxx was pulling in all of Carbon to the global namespace
- very scary. As a result, I now need to explicitly include CoreFoundation
in fg_init.cxx.
- change SG_USING_STD(x) to using std::x
Issues:
- the logic for X11 and Win32 in RenderTexture and extensions is tortured,
please see if you agree I got all the ifdefs correct.
the MSVC and MipsPro warning stuff).
As a result of this patch, simgear/sg_traits.h can be deleted. So can SGCMath.h,
but I'll do that separately.
There is one more 'mechanical' change to come - getting rid of SG_USING_STD(X),
but I want to keep that separate from everything else. (There's another mechnica
l change, replacing <math.h> with <cmath> and so on *everywhere*, but one step a
t a time)
PLETE_FUNCTIONAL from SimGear and FlightGear.
As a result, SG_HAVE_STD_INCLUDES is now *always* set, so I will get the boring
fixes for that done, but separately. I'm still auditing the other things in comp
ilers.h - there's a lot that can die now BORLAND is gone.
