Re: Better support for tailcalls

[Markus Pizka <pizka at biretta dot cam dot uk dot eu dot microsoft dot com>]

David Starner wrote:

> On Thu, Feb 03, 2000 at 10:04:16PM +0000, Markus Pizka wrote:
> > Hello,
> >
> > to support the implementation of new front ends for gcc (especially
> > functional languages) I am highly
> > interested in integrating a fully general tailcall feature into gcc.
>
> Cool. Considering my grant proposal for adding a Scheme frontend to
> GCC, this could come in very useful.

there are a quite a lot of people feeling the same way ... but hardly any
progress

> > I wonder, has anyone succeeded, yet or is there a partial implementation
> > that could be used as a starting point?
> fgrep -ir recursive gcc-cvs | fgrep -i tail | wc
> 10  97  766

> Look in flow.c and stmt.c.

the current support for tailcall elimination is very limited as it is
restricted to
self-recursion of non-void functions only!
(correct me if I am wrong)

> > As a first step, I think it would be sufficient if the gcc back end
> > would not automatically detect
> > tailcalls but only provide a mechanism to transform a tail call
> > specified by the front end
>
> Why? Personally, I was planning on doing it (if I had to) in tree,
> as that should have all the information necessary. I don't see
> what you'd gain by having each frontend try to interpret it.

agreed. As I said, this would be a first step. Automatic detection could
be added later. The reason is, that there are languages (such as C--)
which provide and explicit tail call concept. That is a tail call - let's
say a non-local goto with arguments - is specified in the program.
To implement this concept it would be helpful to be able to explicitly
specify
a tailcall instead of a call expression which is (hopefully) optimized into
a tail call. To me, the automatic detection and application of the tail call
mechanism is an additional feature build on top of the pure mechanism.

Markus