Initial test-suite pass about llvm-mos

[mysterymath]

In the initial reckoning, it looks like 420 test cases pass, 55 fail, and 1321 fail to compile or are otherwise missing.

[johnwbyrd]

I'm surprised that so many tests pass, this early on! I vaguely recall that at least some of the tests may depend on floating point.

[mysterymath]

I'm surprised at how much of printf seems to work, once I fixed a few key issues. LLVM does a remarkable amount of verification and asserting on the inside, so I guess it's actually kinda difficult to get really really broken stuff through that. Just regular amounts of broken so far.

Anecdote:
A bunch of garbage was getting printed after the 0 in printf("%d\n", 0);, but only on the c64. I added a bunch of diagnostic __putchar('X')-es throughout the code, and to my surprise, it looked like the garbage was after printf! I was returning back to basic after I'd walked all over its variable structure, and basic was printing the garbage.

After changing the return to BASIC to an infinite loop, seeing printf("HELLO %d %s\n", 6502, "COMPILER") print "HELLO 6502 COMPILER" was a surprisingly big thrill, given all that's involved in that actually working.

[mysterymath]

On second though it's going to be difficult to keep a "working baseline" established while the work of repairing the test suite is underway. Fixing a compilation error can (and has) activated a hundred or so additional test cases, which can create more failures than it's reasonable to fix in the same commit. I'll be attempting to alternate between fixing compilation issues and fixing test executions so that the suit stays "as working as possible", but the actually number of failing tests will likely oscillate.