On Nov 26, 10:28 am, dreish <[EMAIL PROTECTED]> wrote:
> Now that you've gone this far, why not do this?
>
> - class clojure.lang.TailCall contains an AFn
> - Compiler checks for a tail call position and instead of calling it,
> returns new TailCall(AFn)
> - In invoke, while returnvalue instanceof TailCall, returnvalue =
> returnvalue.fn.invoke(returnvalue)
>
> I confess I don't yet have much familiarity with the compiler; I'm
> just reading through it now. Is the above idea workable?
>
> Upside: you could have real TCO tonight, with J2SE 5.
>
> Downside: as soon as they finally get around to putting TCO in the
> JVM, it becomes useless cruft.
>
> I would think you could take it back out, though. And who knows when
> they'll finally do it? And wouldn't it be better to put TCO workaround
> cruft in the compiler, rather than in the language definition, around
> which people are already (foolishly, perhaps) starting to build
> serious programs?
>
There's a very simple reason - such built-in trampolines are
incompatible with interoperability. Anyone can build a trampoline
system into their lang to get tail calls and make sure it is used
consistently within lang - but what do those TailCall return values
mean to a Java caller?
Also, in what you've described every invoke turns into a loop, every
tail call allocates etc.
I made a conscious decision here, and it was a pragmatic one. People
using Clojure are pragmatic, not foolish.
The vast majority of tail calls requiring TCO are self-calls handled
by recur.
I'm all for TCO, but trampolines are simplistic TCO and no substitute
for real jumps.
Rich
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups
"Clojure" group.
To post to this group, send email to clojure@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at
http://groups.google.com/group/clojure?hl=en
-~----------~----~----~----~------~----~------~--~---
