shithub: femtolisp

ref: 5d480a2944a16a22fcaad448cd0822b45858b222
dir: /todo-scrap/

View raw version
- readable gensyms. have uninterned symbols, but have all same-named
  gensyms read to the same (eq) symbol within an expression.
- fat pointers, i.e. 64 bits on 32-bit platforms. we could have full 32-bit
  integers too. the mind boggles at the possibilities.
  (it would be great if everybody decided that pointer types should forever
   be wider than address spaces, with some bits reserved for application use)
- any way at all to provide O(1) computed lookups (i.e. indexing).
  CL uses vectors for this. once you have it, it's sufficient to get
  efficient hash tables and everything else.
  - could be done just by generalizing cons cells to have more than
    car, cdr: c2r, c3r, etc. maybe (1 . 2 . 3 . 4 . ...)
    all you need is a tag+size on the front of the object so the collector
    knows how to deal with it.
    (car x)      == (ref x 0), etc.
    (rplaca x v) == (rplac x 0 v), etc.
    (size (cons 1 2)) == 2, etc.
  - one possibility: if we see a cons whose CAR is tagptr(0x10,TAG_SYM),
    then the CDR is the size and the following words are the elements.
    . this approach is especially good if vectors are separate types from
      conses
  - another: add u_int32_t size to cons_t, making them all 50% bigger.
    access is simpler and more uniform, without fully doubling the size like
    we'd get with fat pointers.

Notice that the size is one byte more than the number of characters in
the string. This is because femtoLisp adds a NUL terminator to make its
strings compatible with C. No effort is made to hide this fact.
But since femtoLisp tracks the sizes of cvalues, it doesn't need the
terminator itself. Therefore it treats zero bytes specially as rarely
as possible. In particular, zeros are only special in values whose type
is exactly <tt>(array char)</tt>, and are only interpreted in the
following cases:
<ul>
<li>When printing strings, a final NUL is never printed. NULs in the
middle of a string are printed though.
<li>String constructors NUL-terminate their output.
<li>Explicit string functions (like <tt>strlen</tt>) treat NULs the same
way equivalent C functions would.
</ul>
Arrays of uchar, int8, etc. are treated as raw data and zero bytes are
never special.