I find that compass() is prone to causing system errors without explanation.  It seems to me that there are some inconsistencies if the call includes referenced points.  If I go back and give each point referenced an explicit name, the system error will go away.  In particular, a radius I specify is likely to be the length of some other object or the distance between two points [p1:p2].l, so that may be what does it.  But I don't get a syntax error, just a system error.  (I haven't tried to replicate the problem in a small program, but I've bumped up against it for a while now.)

Actually scratch that, it's when the points are a reference that there's a problem.  E.g.:

Ah, okay, that didn't occur to me, something to look out for.  I mostly use [p1:p2] in references, but I must use it carelessly instead of line(p1,p2) all the time. 

I went back and changed the definitions of l1 and l2 to use line(), but I'm still getting errors.  Maybe it's an abuse of notation, but most functions can take explicit points or referenced points.  Checking a little more, I find a problem with Compass() and Intercept().  The other ones I tried work the way I expect.  The lines that are commented out cause errors, or mistakes, but I think they should work, e.g. the three lines should create the same p4, or the same p7.

Thanks, I reported it to IT and they are looking into it!

The IT just released a fix, could you please check if it's ok now?

Thanks, it works!