On 11/29/2011 03:12 AM, Steven D'Aprano wrote:
On Tue, 29 Nov 2011 13:57:32 +1100, Chris Angelico wrote:
I'm inclined toward an alternative: explicit recursion. Either a
different syntax, or a special-case on the use of the function's own
name, but whichever syntax you use, it compiles in a "recurse" opcode.
That way, if name bindings change, it's still going to recurse -
something few languages guarantee, and therefore few languages can
optimize.
As I recall, Forth uses (or used) a special RECURSE word which turned on
the recursion bit while compiling, so that the compiled word could see
itself.
By memory, the (incomplete) definition:
: fact dup 1- fact * ;
would fail, unless you happened to already have another word called fact
existing at compilation time. To make it recurse correctly, the compiler
needs to make sure that the namespace fact sees includes itself:
RECURSE : fact dup 1- fact * ;
which should work, apart from the embarrassing fact that I don't recall
the syntax for conditional jumps and so the recursion never terminates.
:)
The way I remember it, the current definition was "smudged" which made
it invisible (it basically changed the name to something unlikely)
during the compilation. After all, if you actually ran it at compile
time (which was frequently done), you could fall right into
uninitialized space. Anyway, some implementations had an immediate
SMUDGE word, which toggled the smudge bit and made it visible again.
Other implementations had an immediate word RECURSE, which compiled in
whatever word was being currently defined.
I'm pretty sure neither FIG nor Forth79 had either of these. But I
don't recall the ANSI standard (X3J14 ?), even though I was officially
an observer. I can't even remember what happened to my printed copy of
the standard.
The easiest word for conditional is IF/ELSE/THEN. IF will skip to the
ELSE or THEN if the condition is false. So something resembling:
: fact dup 1- dup 0<> if recurse * then ;
might do it. That's very rough, however. It's been a long time.
--
DaveA
--
http://mail.python.org/mailman/listinfo/python-list
