The buffer offset was encoded in the VAO which would be shared across all
submeshes. Instead, don't encode the buffer offset into the VAO but apply it
with glDrawElementsBaseVertex().
glDrawElements needs an IBO when using a core profile. The particle
system's IBO is actually quite static since it's always a triangle
strip with 2 triangles followed by a PRI. Therefore, this reduces the
amount of data we have to send to the GPU compared to the previous
solution.
Also, remove the workaround when glPrimitiveRestartIndex is not
available since it is always available with OpenGL 3.1 and when using
a core profile we are guaranteed to have OpenGL 3.1 anyway.
This is required for glDrawElements() with a core profile. Furthermore, for
meshes that have a fixed face ordering (non-transparent meshes at 100%
completion), this puts more static data on the GPU that we don't have to
upload every frame.
For meshes with non-fixed face ordering, we still have to upload faces every
frame, since the ordering might change if the mesh rotates. We could still
improve this by figuring out if the order actually changed after the sort
step, and only update to the GPU if it has. In many cases it hasn't, so there
is some potential for more optimization there. But that's for later.
This code was basically never used anyway, since the
ARB_non_power_of_two extension is always available since OpenGL 2.0.
The GLEW check for the extension fails however as soon as we select
a OpenGL 3 core profile, and the texture sizes will be unnecessarily
adjusted again.
Basically add another layer of indirection around accessing VAOs. The problem
is that VAOs are not shared between OpenGL contexts. This mechanism allows to
treat them mostly as if they were shared if they are only accessed through
the API defined in CStdGL.
This is accomplished by caching all existing VAOs per context. If a VAO is
being accessed, it is checked whether that VAO exists in the currently
selected context. If yes, return it, otherwise create a new VAO and return it,
together with a flag that indicates that the VAO needs to be initialized.
The former is not available in a OpenGL 3 core profile, but glGenerateMipmap
is available since OpenGL 3.0, and so is always available when we have a
core profile.
This prevents a crash when an incompatible engine function is used as an
engine callback function.
Unfortunately, this breaks any scripts that have wrong type information in
engine callbacks and only worked because those were ignored.
The engine has a few more usages of the operators, but they don't look
prone to overflowing. The other operators in Script already used SetInt,
which always truncates to 32 bit.
Since mesh instances can now be attached to attached definitions, attached
definitions need to denumerate pointers recursively to make sure that the
reference to the mesh instance is restored correctly.
I am not sure why the 'isMainWindow' was there. It wasn't there in 7319f7b3cc and got introduced in the major rework in 049088be78 - I guess it was just an oversight that was not noticed because usually the UI windows have the text or other things that need to scroll on a deeper level.
Anyway, checking whether the window is a script-root window does not make any sense as far as I can see now.
Gamepad support is currently not working properly as many menus cannot be
navigated with a gamepad. Simply don't load any control assignment sets that
have gamepad enabled.
Otherwise the font chooser in the startup options shows those fonts, but
loading them will fail. Ideally we replace this by enumerating the system
fonts at some point.
Some sort of smooth scroll information was passed down to the engine which
didn't handle it properly, or the units with which they were reported to us
were different from what the engine expected. For now, just use fixed scroll
offsets until we implement proper delta-y-aware scrolling in C4MouseControl.
The fix for #1574 might have also helped with this in recognizing when Control
is pressed and when it isn't.
For example, keydown events when pressing/releasing the shift, alt, or control
keys. With this version this is needed ingame e.g. to pick up objects with
Shift.
Reuse the windows code for this. Both versions had three parts: Enabling
the right menu items, adding the selection-specific ones, and showing the
menu. The second part is now common code, with the platform code in a new
function, since it grew big enough to be worth sharing.
glObjectLabel would result in invalid VBOs if called on a not-instantiated object. glGenBuffers does apparently only reserve the name and not create the buffer itself.
For me the landscape and some objects were drawn completely black.
More information: http://stackoverflow.com/questions/26614130/opengl-debug-extension-with-globjectlabel
Found via --debug-opengl
which lead to
gl: high severity API error: GL_INVALID_VALUE error generated. ObjectLabel: unknown buffer object <name>
In earlier versions, DrawPatternedCircle was a fairly complicated function.
Nowadays, it just copies pixels in memory, bounded by a circle. Since we
need to convert our surfaces to a format GTK+ can use anyway, combine that
step with the circle boundaries. Also use a cairo surface instead of a
GdkPixbuf. Cairo uses almost the same pixel format, except for
premultiplied alpha. Ignore this for now.
Specifically, GtkGrid. For most cases, this is an increase in lines of
code, but the landscape tools dialog can now be done with just three layout
widgets.
While at it, use a toolbar and move the mode buttons into it, too.
9e771ac moved the Win32 ClearInput implementation into SetInputFunctions,
but broke at least GTK while doing that. 9a7d57c0d2 fixed the assertion
failure in GTK that resulted, but failed to restore the removal of the old
completions.
Fix this by folding ClearInput into SetInputFunctions in all implementations.
This reverts commit 790219ac7e.
This commit broke support for Mac OS X since Apple only supports OpenGL 3.0
with a core profile, not a compatibility profile. Revert this commit
temporarily for the 7.0 release, until we port all OpenGL usage to the core
profile with 8.0.
This fixes#1495.
This would crash for me when starting a game in developer's mode. Not sure if this is the best place for the fix. Maybe there is some other underlying issue. ck?
In doing so, refactor the way C4FoWDrawStrategy works slightly. Get rid of
immediate mode OpenGL, and instead populate a vertex array and an index array
as vertices are added to the draw strategy. Then render everything as
triangles.
The actual rendering is then performed in C4DrawStrategy::End(). It passes
the accumulated vertices to the GL. Also get rid of the notion of multiple
passes in C4DrawStrategy. Instead, C4DrawStrategy::End() can simply just draw
the bunch of vertices multiple times. This allows us to do the vertex
generation and GPU upload only once, instead of repeating it for every pass.
This allows the strategy to keep some state between invocations. I want to
use this in order to store a VBO in the draw strategy, and to avoid having
to re-create the VBO for every rendering.
We are already using VBOs for meshes, so we can rely on them being available.
VBOs are shared between contexts, so #1195 is not an issue, either. However,
keep the workaround in place for the VAO, which is not shared between
contexts.
I don't expect much of a performance benefit from this, but it works toward
removal of legacy OpenGL usage. Note that we are already using VBOs for
meshes, so this does not require any functionality that we don't require
already.
Changing the mesh material to something that does not use alpha blending
restores the originial face order. This caused a crash for meshes with
Completion below 1, since all original faces were copied, but the buffer
in the mesh instance was only large enough to hold the number of faces of
the incompleted mesh. To fix this, simply run the standard
select-faces-for-a-given-completion-value procedure.
This probably enlarges the maximum landscape height for borderbound objects.
The "infinity" constant +1000000 is larger than the biggest C4Fixed, so
would have been truncated to 16960.
The DynamiteBox sets the PictureTransformation to a degenerate matrix, to
prevent it from being rendered. Exit early in this case, since we are not
going to render anything anyway, and avoid the assertion when inverting the
matrix.
I think C4Game::InitKeyboard relies on the callback targets outliving the
keybindings and every other class is prepared to be deleted before
C4KeyboardInput::Clear(), so needs to hold a ref to the keybinding to
delete that in its destructor.
C4KeyBinding is the class that is used for that, and it sets the reference
count to 1 in the constructor, to represent the reference held in the
owning class.
On the other hand, C4CustomKey gets immediately forgotten by C4Game and the
only reference is held in the keymap.
To prevent this from happening again, make the C4CustomKey constructors
protected and C4Game a friend of the class.
Also remove an unused C4CustomKey constructor.
Replace the hardcoded VAI_ constants with that. The VAI constants are now
moved to C4DrawGL as C4SSA_ constants, similar to the C4SSU ones. This allows
to introduce other attributes to replace vertex positions, normals, colors
and texture coordinates with attributes in later commits. This, in turn, is
needed because the built-in attributes are no longer available in the OpenGL
core profile.
Also, while at it, cleanup C4Shader a bit. Use std::vector<> instead of
maintaining an own array of uniform names. Delete the vertex and fragment
objects after the full shader program has been linked. Make sure that
C4Shader::Init keeps the old program in place if the new one cannot be
compiled or linked.
Surfaces are initially created in locked state, and only when unlocked the
memory buffer with the surface data is uploaded to the GPU. The FoW code
however never unlocks the surface but instead operates on the OpenGL texture
directly. To fix this, introduce a new flag for C4Surface to create it in an
unlocked state.
This is not a memory leak in the traditional sense as a pointer to the buffer
is still available and is being freed when the surface is being destructed
at the end of the game. However, during the game there might be several
megabytes allocated that are never used.
I'm not really sold by this Lock/Unlock mechanism. Instead, there should
simply be a function to obtain the current texture data, and a function to
upload new texture data. That's for another time though...
For some reason, the C4TexRef constructor first uploaded a set of nulled
texture data, just to be replaced by the actual image data when Unlock()
is called. Skip the unnecessary step and instead just reserve the video
memory in the constructor.
This also fixes objects which previously jumped out of the water continuously, by adding an extra check to also capture faster moving objects. It could be made speed dependent, but that seems to be overkill for the situation.
The only use of C4RTF in its final moments was parsing out plain text
from RTF files anyway, so why even go to all the trouble instead of just
storing plain text in the beginning?
This was used to name snapshot releases of the Network2 branch, and has
seen almost no use since.
C4ENGINEINFO(LONG) was a duplicate of C4ENGINENAME and C4ENGINECAPTION.
Code outside of C4FoWRegion should not care about the internals of the
class. Therefore, we remove direct access to the backing surface (and
secondary buffer surface) and replace it instead with accessors that
return those few values that are required by outside code.
In order to avoid duplication, PlatformAbstraction.h has to be included at
the start of every source file, so that various headers can rely on it
being there. To avoid confusion, always rely on that, instead of sometimes
randomly including it or parts of it again.
Video recording and playback only worked on Windows, and recording only
handled video, not game audio. As such, it was of fairly limited use,
and there's lots of software available these days that handle both.
GUI_Multiple windows are supposed to be updated from the root window in a full refresh. However, they also had the "dirty" flag set (that is only used for non GUI_Multiple windows after the initial startup), but they did not unset it when updating.
So they were first updated with the correct parent width (full screen) and then, because of the dirty flag, were updated again like a non GUI_Multiple window with the parent's size. That was incorrect. Only another update would then fix it again (because the "dirty" flag would then have been unset).
If the user switched away from OC while it was loading, there was
probably a reason for that. We'll no longer assume we're the most
important app in the world and steal the focus away from whatever the
user was doing.
It's still pretty useless because nobody cares in the slightest about
which extensions are available and which ones aren't, but at least now
it doesn't throw an unintelligible list of numbers at you.
In the game initialization code, we're setting the thread locale to the
user's default locale. This is bad when generating shader code because
it may lead to floats being written with commas instead of dots for the
decimal separator, which in turn confuses the GLSL compiler (because it
parses the comma as the comma operator). So now we just temporarily set
the C locale while generating shader code.
In a surprising touch, using shaders to affect pretty much every pixel
that gets drawn on screen means there is little reason for code to ask
the renderer if it knows about shaders.
There are only a couple of error values worth considering, so we can just
write out own function for it. Disable error checking in C4FoWRegion.cpp as
well, since we have the --debug-opengl flag now.
This should allow us to get rid of the GLU dependency soon.
Otherwise, when starting the next scenario, random spots would be initially
re-lighted. That's not necessarily a problem, but unnecessary work, and it
triggered the assertion in C4FowAmbient::UpdateFromLandscape if the relight
was scheduled for a region that's outside of the landscape boundaries of the
second scenario.
Instead, compute the projection, modelview and normal matrices explictly
and upload them as shader uniforms. This is one step towards using the
OpenGL 3 core profile.
Translate() used to select the script which it checked for translations
before actually looking up the string key. Change it so if the object
the context of which it was called in doesn't have a matching
translation, the string table of the script containing the call is used
instead.
Hardly anyone ever tests the headless build, because it requires you to
run CMake a second time with -DUSE_CONSOLE=On. So now it's an included
project which you'll get whether you want it or not.
I'm well aware that this could be solved more nicely, and that we could
be splitting unchanged files out into a separate support library, but
that's left for a later date.
Another piece of code that hasn't been tested in ages, gone. The default
puncher address was still pointing to clonk.de, which I'm very certain
isn't providing UDP hole punching services anymore.
Instead of a bespoke hole punching implementation we should use a STUN
lib.
It was almost never appropriate (played underground etc.), could not be turned off and was very repetitive. Mountain scenarios with lots of wind may activate the sound.
A playlist change would reset SCounter and eventually cause all songs to be non-eligable.
Since playlists can now select differently sized subsets, the old way of selecting random songs without too much repetition no longer works. Rewrite to use a more robust algorithm.
Getting the error summary written to the log/console automatically may
be useful for general game usage, but when testing Aul we just end up
with thousands of useless messages in the output. Let the caller log the
message if it's really desired.
Also move all object sounds to global Sound.ocg so they are still played.
This is to prevent naming collisions. All sounds are moved because otherwise all scripts would have to be adjusted. Also, the distinction between sounds in definitions and sounds in Sound.ocg was rather arbitrary.
It will also ensure most sounds are reloaded when sound is turned off temporarily, although that should better be fixed properly.
Deserializing players neglected to turn an enumerated proplist back into
a live one. This would explode later during player initialization when
someone didn't check that the deserialized value was a real, live
proplist.
None of these functions require access to any state of C4Landscape, and
they've only been used from within C4Landscape.cpp, so don't need to be
part of the header.
When clearing parts of the landscape, solid masks must be disabled
beforehand because we don't want to remove the actual solid mask itself.
Theoretically we should also temp remove solid masks before DigFree,
ShakeFree, and all of the other landscape modifiers, just in case
someone overrides the Vehicle material with stupid values.
Pressing the Alt key enters the menu modal loop, which takes control of
the thread that everything happens on. As a quick fix, we'll suppress
this loop for simple Alt key presses, but this will still happen when
the system menu gets opened by Alt+Space or clicking the icon in the
window's title bar.
A more permanent solution would be to detach the message loop from
everything else, by having rendering etc. run in a separate background
thread.
* Show max one message per second per client
* Do not show ready message locally
* Fix observer flag reset by ready state change
* Reflect ready state change in client list immediately
The C++ standard library comes with perfectly fine implementations of
these functions, so there's no point in reimplementing them just for the
hell of it.
Using memset to initialize non-POD types doesn't work. Or rather, it may
work right now, but will fail when somebody adds a member that relies on
its constructor doing something (like for example any STL container).
Either way it's undefined behavior and needs to go. Furthermore, using
it to reinitialize an object also prevents any dtors from doing their
work when needed.
A new helper function InplaceReconstruct will take an object of nothrow-
default-constructible type, and call the dtor to properly clean up
before placement-new reconstructing the object in the same location.
This is still bad design, but unfortunately removing the Default/Clear
functions from every object currently using them is a herculean task.
The problem comes from the fact that the landscape texture can be larger
than the actual landscape, and that for the last row or column of pixels
drawn at the edge of the landscape, the shader might access a texel that's
beyond the landscape boundaries, at least as part of its interpolation
when accessing a texture.
This patch fixes this by always making the landscape surface the same size
as the landscape, and then GL_CLAMP_TO_EDGE behaviour will prevent access to
pixels beyond the edge of the landscape. Modern hardware should handle
NPOT textures just fine, and they are already used in other places in Clonk.
If this becomes a problem and needs to be reverted, we could instead upload
the "real" landscape size to the shader, and clamp the texture coordinates
there to make sure to not read beyond the actually filled landscape surface.
Before, material normal bias was calculated only from a column/row of pixels right
next to the pixel in question. This sometimes led to "lines" appearing in edge cases.
The new code modified this approach to look at a whole rectangle instead, which
gives a *lot* smoother results and is only marginally more expensive.
Removing solidmasks temporarily as part of the relighting procedure caused
a landscape update which modified the list of to-be-relighted regions that
is currently being iterated over. This could cause relighting of a region in
which solidmasks have not been removed, leading to vehicle pixels in the
landscape surface used by the landscape shader, which show up pink on the
screen.
Fixed this by introducing C4Landscape::_SetPix2Tmp, which changes a pixel
without causing relighting (or material count updates), and use that when
temporarily removing or putting solidmasks. This should also avoid unnecessary
computations (relighting, material count updates) when updating the landscape
or moving objects with solidmasks.
Instead of Y - Hgt > LandscapeHeight (or, equivalently, Y > LandscapeHeight + Hgt), it was checked for Y + Hgt > LandscapeHeight, which would remove objects too early.
The minimum GTK version is now GTK+ 3.4, which is available since 2012.
It's part of Ubuntu LTS 12.04, and so should be available on any halfway
modern linux distribution.
This should allow getting rid of using deprecated GTK+ API much easier.