Are you simply trying to pass command line options to the kernel which is used fromt he FE? Or are you looking to evaluate code on startup, not through init.m?
@Szabolcs This is going to be a problem on OSX and probably Win too. In Linux, everytime a new FE is launched
The situation is really crappy because what I probably can support from within IDEA is to start Mathematica, open a notebook and set the working directory (which lets you load the dev-package you are working on right away).
@Szabolcs All the time I was so damn sure that I can supply at least code that is started so that I can set $Path and load the dev package etc..
(btw, you MATHKERNELINIT does work here kind of)
@Szabolcs Does the MATHKERNELINIT method work on OSX? Because here, the code is really evaluated every time the kernel is started in this session, which is a good thing.
@halirutan If I run /Applications/Mathematica\ 10.app/Contents/MacOS/Mathematica from the command line, it starts a new instance of the FE every time on OS X. That is what you need, right? So it might actually work.
This should be a good quick and dirty method. What the Workbench does is more "proper", but I don't think it's worth spending the time to implement it ...
@MichaelHale I think I managed to extrapolate that there are 301359406576520469424323428352000000 valid variations of face triangles for icosahedron in the "Hamiltonian cycles routing" problem, give or take rotation and mirror symmetries. There might be some computational challenges ahead...
@kirma Hm, yeah. I think someone would have to find some clever proof based on the number of possible rotations of the tile compared to the number of faces on the polyhedron, which somehow adds some constraints on the path lengths or something.
@MichaelHale Actually, now that I think of it, it's probably easier to approach the problem by trying to generate a counterexample... but that's not what I want! :)