As long as we're not actually using a different shader for meshes
without bones, we need to upload an identity matrix so there's defined
data in the bone slot.
Doing skinning on the GPU shows a noticeable performance improvement in
pretty much any situation, but especially so in scenes with lots of
animated objects with high polygon counts.
Calling CStdGL::CheckGLError calls glGetError, which is really, really
slow because it has to flush the pipeline to check whether there's an
error or not. Plus it's not like we can do anything about it anyway. If
you want to be notified when an error happens, pass --debug-opengl to
the executable.
Instead of transforming all vertices on the CPU every time an animation
progresses, we now only recalculate the skeleton, leaving the heavy
lifting for the GPU. This also means we no longer have to push all
vertices onto the bus every frame, because the mesh isn't changing and
can therefore be stored in a GL_STATIC_DRAW VBO when it's first loaded.
The downside of this approach is that there's only a limited number of
uniforms and vertex attributes we can pass to the shader. At the moment
these limits are a maximum of 128 bones per skeleton, and no vertex can
be influenced by more than 8 bones at once. So far this is no problem,
as the most complex skeleton in the base game uses less than 64 bones
and no more than 6 bone weights per vertex.
Instead of having the default vertex shader hard-coded into the engine,
allow to load it from Graphics.ocg. There's still a fall-back version
wired into the engine because we can't return an error from
GetVertexShaderCodeForPass.
Unlike std::make_shared, std::make_unique was unfortunately missing from
C++11. It's a useful utility though.
Technically, declaring a new name in the std namespace is undefined, but
the other way to make make_unique available to all callers regardless of
C++ version, putting it into a distinct utility namespace and importing
the declaration from std if available, makes for more ugly code.
While we're still not doing skinning on the GPU, copying the vertex data
to a VBO immediately after updating the animation allows us to re-use
that data for unanimated meshes. It also allows us to store unanimated
data on the GPU, instead of transferring it over the bus for each frame.
This means we can drop the opaque parameter. Where additional data is
required, a lambda is the way to go here. It also means we don't have to
put half of the parameters into a global variable.
Makes the whole thing more robust in corner cases. Also means that the
terrain a Clonk walks over doesn't matter that much anymore, which is
probably good.
On the other hand, now walls can be "partially invisible", which can
cause strange light effects. We'll have to see whether this is looks bad
in practice.
Descend collision was broken because of two reasons - one complicated and
one easy. Firstly, we would not re-visit beams after eliminations if they
were closer to the light source than the remaining beam. Especially nasty
because the comments claimed the opposite. Secondly, the coordinates
passed to find_cross were actually flipped.
I took the opportunity to clean up the control structure a bit, update
the comments (brr), and fix the bug in one swoop.
In the very common case where the C4Surface only uses a single texture
to store its data, a lot of work GetTexAt is actually unnecessary. Split
it up so we can inline the fast path and only fallback to the slow path
when the surface is split up into multiple textures.
glGetString(GL_EXTENSIONS) is deprecated starting with OpenGL 3.0.
Instead, you're now supposed to retrieve the list of extensions one by
one with glGetStringi.
We're not inheriting from C4PVP, and all of these internal functions
really shouldn't be called from outside the class. So private is what
they should be.
Sven2 reports he can't get reliable stack traces from his debugger when
the assertion handler is installed. Since there's no need for the hook
when we're already running under a debugger, don't install it.
We've been using OpenGL 2.1 features for some time now, and hardware has
started supporting OpenGL 2.1 in 2005. I doubt this will make anyone
unable to run the game, and it's certainly better than crashing because
of a nullpointer dereference when some GL function we use can't be
found.
The MSDN reference for wglMakeCurrent states that the first (hdc)
parameter is ignored when the second one is NULL. This is incorrect: it
checks validity of the hdc parameter before doing any work. Since we
have a DC anyway, it's no problem to pass that to wglMakeCurrent.
In applications targeting the GUI subsystem, stdout and stderr aren't
valid file handles, and trying to write to them sets the system error
code to ERROR_INVALID_HANDLE.
GL startup failures call Application.Clear(), which will at some point
before creating the error dialog post a WM_QUIT. When the dialog box's
message loop retrieves that message, it will shut down the dialog box,
thus ensuring that the user will never see it.
So before showing the dialog box, we have to dispatch any pending
messages, then retrieve the WM_QUIT ourselves, run the dialog box, then
re-post the WM_QUIT.
Depending on how current your headers are, the userParam parameter to
GLDEBUGPROCARB may be const, or it may not. The ARB has added the const
qualifier at some point after publishing the specs. Hooray for breaking
API changes.
Assertion failures or crashes during process shutdown must not write to
the log file if it is already closed, or the debugging CRT will raise
another assertion.
This introduces a new command line parameter "--debug-opengl", which
will create special debug OpenGL contexts and attach a callback that the
driver will invoke when it detects a problem. The callback will then
write the error message to the logfile, and break into the debugger if
one is attached.
Currently only works on Windows.
gluErrorString returns latin-1 encoded strings. Our code expects to
receive UTF-8 encoded strings everywhere, so make sure that the strings
are converted before passing them on.
A line object must have at least two points to plot a line through. At
any vertex count less than that, we were accessing invalid memory by
dereferencing the first element of an empty vector (VtxNum==1), or
issuing an OOB read from the vertex data (VtxNum==0).
C4Group::View is only used in the c4group command line tool, so the
right place for that code is inside the tool. Added two public constant
accessors to return the header data and entries.
If vsnprintf fails for other reasons than too small a buffer, it returns
a negative value. Comparing that with a size_t value promotes the
negative value to unsigned, which makes for a very large result, almost
guaranteed to be larger than the buffer size.
Nicolas Hake