Tips and pointers (*berylhackingtips):
* A very small portion of beryl is the actual OpenGL magic, most of working with plugins is manipulating the main data structures (CompScreen, CompWindow) and working with Callbacks, core is a different beast. Appropriate data structures are more important than impressive algorithms!
* IPCS can be used for communication between plugins see beryl-core/src/ipcs.c and Cube/Rotate for good examples. Avoid use of this however as it is both slow and ties plugins together in ways you may not wish.
* Premature optimization is the root of all evil, profile before needlessly wasting time on optimization. To profile beryl you can compile with -pg (CFLAGS="-pg" ./configure; make; blabla) and when beryl exits it drops a 'gmon.out' in it's working directory. From that directory run gprof /path/to/beryl/executable > berylprof, and you can view the profile in a text editor, or a profile tool such as kprof (highly recommended).
* WRAP/UNWRAP works a bit like a stack of sorts, examine a backtrace from gdb to make things clear. This is relatively inflexible and ideas for improvement post 0.2.0 are welcome.
* Beryl can be debugged without crashing your system, compile with debugging symbols (-g) and run in a tty with gdb beryl, set breakpoints with 'break function' run with 'run' when X freezes because of the break use alt+sysrq+r then ctrl+alt+f1 to get to the tty, examine things, and continue execution.
* Tieing things whether it be plugins, settings/core, or settings/an individual plugin is bad and should be avoided. 'Write (programs) to do one thing and do it well'.
* As a rule avoid pixel shaders as they aren't exposed by AIGLX as of current. Ala water.
* Plugins should work within core, at the same time changes to core can be made to accomadate a specific feature such as transparent cube.
* Optimizations are always welcome, benchmark isn't horribly accurate but if you consistently notice a ~10-20 FPS improvement, or a decrease in lag spikes , or something that's not '1 FPS more when rotating cube left' go ahead and commit it, worst case is it's reverted. Things to look at: Switcher, particle system in animation, input enabled zoom.
* A keybinding to replace windows (run through placement algorithm again) would be an easy to implement and useful feature. Bonus points if the windows fly to their new positions.
* Comment code! not only does this help everyone in their tasks it helps you learn the codebase, pretty easy to do.
* Allow annotate to draw directly on to the window texture. This would involve procing drawWindowTexture as in beryl-plugins/src/neg.c and binding an opengl rendering context to the windows GLXPixmap, beyond that it's pretty similar to the rest of annotate.
* Improve plugin short descriptions, maybe make long descriptions a bit more wordy as well as they are being moved to tooltips.
* Anyway to speed up IPCS?
* Refactor some of the code in place, it's spaghetti code as it stands.
* Exempt specific windows from window decorations as part of the emerald menu. Remember this setting for windows? look at all kwin has for stuff like this, duplicating a small amount would be nice.
* Contribute to I18N
* Work on kberylsettings the PyQT settings manager in branches.
* Add CompOptionFont to settings
* Look for functions that can be inlined. use __attribute__((always_inline)) to force the inline even without -finline-functions
* Look at using LIKELY/UNLIKELY macros for if/else statements in tight loops with a high discrimination towards one or the other. Profile to find such areas, premature optimization is bad.
* Participate in discussion and planning on IRC, it's invaluable. Make yourself a part of the beryl development community rather than a lone wolf.
* Take lengthy discussions on the mailinglist.
* Add more code documentation in Documentation/ about interfaces used by Beryl.
* Look in Documentation/TODO-Multiscreen for multiscreen bugs to fix during 0.3.x
* Add more stuff here!