This commit allows OS X users to enjoy their current theme's custom splash graphic if it is available at ./Graphics/Common splash.png
I have been wanting to fix this for a fairly long time.
My fix is probably not the cleanest or proper way to do this. StepMania.cpp already looks up the current theme's "Common splash.png" and passes it off to the arch-appropriate LoadingWindow as a RageSurface. This fix repeats the lookup in LoadingWindow_MacOSX.mm, and loads the splash image into a Ragefile which is easily loaded into an NSImage.
Still, my fix does work, and if you actually know how to work with a Ragesurface and can help out here, I'll give you a cookie. :)
* Restore Windows installing to `C:\Games` by default.
* Restore Windows installing to the `Program` subdirectory.
* Use actual newlines in the `Build` documentation files.
* Remove the text that says that cmake is unstable.
RageTextureManager now keeps a map of textures that need updating, which is none of them because I couldn't find a class that inherits from RageTexture that doesn't have an empty Update function.
ArrowEffects now requires setting the current PlayerOptions before calling any functions. This might make moving to per-column mods easier, and reduces the direct usage of PlayerState.
Tipsy calculations for each column are done in ArrowEffects::Update instead of GetYPos and GetYOffset.
DrawHold changed to only call DrawHoldBody once. DrawHoldBody now does the normal and glow passes together instead of needing to be called twice.
DrawHoldPart changed to take most of its args in a struct.
Giant copy paste mess that NoteField was using to draw timing segment text replaced with a couple functions and a macro.
Selection glow for notefield is only calculated if a section is selected.
Added RageVector3 functions to CubicSplineN for NoteDisplay to use.
Actor effect period is cached now. Sprites and Models cache their animation length now.
ReceptorArrowRow no longer calls ArrowEffects::Update in gameplay.
On my Fedora 20 system, closedir apparently frees the dirent, and
reading it afterwards returns garbage data. This prevented my dance pad
from being detected.
A minor change to the code exposes the timeout parameter of `select()`, which was already in use for CanRead()/CanWrite(). The timeout parameter is used to block the current thread until data (for read) or space (for write) is available on the underlying socket.
A timeout of 0 suppresses any blocking (the original methods fix the timeout to 0).
The timeout value is supplied in milliseconds as an `unsigned int`; this is converted internally to the equivalent `timeval`.
An in-progress addition to my SextetStream input driver is implemented with non-blocking reads on a TCP socket. Without the block+timeout functionality, it must busy-wait for new input. Changing the reads to blocking works during a game but can cause the program to hang when exiting. The changes in this commit provide a solution to both issues without disrupting any existing code.
Joe