TODO

TODO
====
  - LDS and STS instructions don't address I/O
  - ATmega8 CALL instruction is bogus
  - recursion - sigsegv
  - Conditional assembly with forward referenced label on .DW directive
    generates incorrect code
  - AVRA complains about "((str_start_%)<<1)" beeing invalid
  - if last line in include is a remark line, it is written twice in the listing
  - Printing of diagnostic messages with hexademical numbers is inconsistent
    (some are %x; others %X)
  - Args handling is overcomplicated. Use something along the lines of
    suckless.org args.h instead.
  - mneumonic.c is a giant if-elseif-elseif-... mess. It would be cleaner to
    have one function for each instruction/group of similar instructions and
    store function pointers in the instruction table. (Same thing with
    directives.)
  - Rethink how device flags are handled? I.e., how we determine which devices
    have which capabilities. Current solution is error prone (I've seen
    multiple cases where devices have wrong flags) and messy. Maybe look to
    gcc-avr for this, which seems to get most of the way by just classifying
    devices into 16 sub-architechtures. Another idea that would yield the most
    flexibility is to associate ~20 bytes with each device, where each bit
    corresponds to whether a device has a particular instruction. We could then
    predefine sets of instructions corresponding to each subarchitecture, while
    still allowing individual instructions to be removed on a device-by-device
    basis, without using much memory.
  - Add a true C preprocessor, like AVRASM. Or, simpler, recommend the use of
    an external preprocessor and give a warning whenever assembler directives
    are used with the '#directive' syntax, since it confusingly makes it look
    like a C preprocessor directive.
  - My general impression of the lexical analysis code is that it is too
    tightly integrated with everything else. It would be cleaner to have more
    of a separation, but this is a little vague and requires investigation.
  - Ifdef/ifndef storing file/line number is kinda hacky. Do something better.
  - Some of the OS-specific makefiles look old and are probably broken.
    Test and fix them.
  - Establish code style and enforce it throughout (i.e., trivial things like
    brace placement).
  - Since we now use C99 instead of ANSI C, we can use stdbool.h instead of
    custom bool type.
  - Don't always use unistd.h. It's only used for unlink, so it should be
    fairly easy to just add some ifdefs to make it work on POSIX systems and
    Windows.