This is necessary because when Python is initialized, it takes the codec to use from GetConsoleCP() and GetConsoleOutputCP() without bothering to check whether the given codec is available. However, in most cases, the console codepage will be the same as the ANSI codepage (ie. GetACP()) which is always supported by Python via the 'mbcs' codec.
So what we do is we check whether the console codepage is frozen in, and if not, we set the console codepage to the ANSI codepage and set the stdin/stdout/stderr encoding to 'mbcs'.
This is still not a perfect solution because the ACP may not be able to encode all characters that the application is printing, which would still result in unexpected errors. Ideally, we'd pull in Python 3.6's _io._WindowsConsoleIO class, which bypasses this whole mess by directly using the wide-character Windows APIs to write to the console.
A notable change is that the color scale is now applied *before* texture blending, matching the FFP. If this breaks anything, we might want to add a configuration option for this.
This also implements the remaining combine modes, CM_dot3_rgb and CM_dot3_rgba, and fixes the broken CM_subtract mode (which currently worked the same way as CM_add).
Fixes: #189
Set "mirror" camera to always have the same roll as the mirror
-> otherwise the model in the mirror will move unexpectedly
Set y coordinate of the mirror corners to the cameras distance
-> This fixes a crash and keep the mirrored model at correct sizer
Closes: #194
This remove support for coercing non-ReferenceCounted types that are neither default-constructible nor move-assignable, but it turns out none of the classes we really need it for matches that.
It further cuts down on the amount of code that is being generated to support coercion in cases where it makes absolutely no sense.
This makes it possible to run pytest in the root directory. It also lets us store metadata such as the current version number, preventing us from having this in several different places, and allowing us to phase out parsing dtool/PandaVersion.pp.
This test just makes sure that created windows match the default
WindowProperties (with a few exceptions).
This also adds the following fixtures:
* graphics_engine - scope='session', GraphicsEngine
* graphics_pipe - scope='session', default GraphicsPipe
* window - scope='test', GraphicsWindow with default framebuffer and window properties
These tests were converted from the tests branch with some changes:
* Convert from unittest to PyTest
* Update code to use new API (LVector3 instead of Vec3, properties, etc.)
Now the debug drawing will no longer happen if the debug node is not being visited by the cull pass, ie. in another scene graph. Furthermore, the generation code has been optimized a bit more.
This change means it no longer inherits from GeomNode. Future improvements could include better culling (which is currently disabled entirely).
Closes: #130
