Steel Bank Common Lisp

On Fri, Nov 03, 2000 at 01:54:23PM +0100, martin@... wrote:

> I have experimented with code from cmucl-2.4.22, in order to find about
> the "special-variables"-problem I reported last time.
> Unfortunately I couldn't find out, why this bug happens, up to now.
> This bug prevents working with ILISP happily.
> In contrast to SBCL, this seems to be fixed in cmucl (I tested
> cmucl-2.4.19).
> 
> Anyway, I tried to patch compiler/ir1tran.lisp (looking at the cmucl
> source).
> The patch below fixes some "typos", I guess, but ut also prints less
> notes when compiling.
> This is especially due to the %coerce-to-callable-function construct.
> 
> This transformation is present in the sbcl-source; so I guess it was
> removed.
> But why??

SBCL's %COERCE-CALLABLE-TO-FUNCTION is more or less analogous to CMU
CL's %COERCE-TO-FUNCTION. The CMU CL code (at least the 2.4.8 code
that I started with -- I don't know about current code) was not
systematic about how it converted function designators to functions,
but had various hand-coded hacks in different places. E.g. in MAPHASH
it used

  (let ((fun (etypecase map-function
	               (function
        	        map-function)
	               (symbol
        	        (symbol-function map-function))))
	..)

(which is not quite correct).

I tried to convert everything to use the same function, and to
make the optimization work correctly for that function. I didn't
like the name %COERCE-TO-FUNCTION because %COERCE-CALLABLE-TO-FUNCTION,
like CMU CL's %COERCE-TO-FUNCTION, does something different from 
(COERCE X 'FUNCTION). I don't remember whether that was the only
reason that I changed the name, or whether I had other reasons as
well.

Is that what you wanted to know?

(I can try applying the CMU CL specials/closures patch myself, but
first I'd like to finish testing and integrating the stuff in the big
patch you sent me before. At the rate I'm going, that will take me at
least two more days, and probably more.)

-- 
William Harold Newman <william.newman@...>
software consultant
PGP key fingerprint 85 CE 1C BA 79 8D 51 8C  B9 25 FB EE E0 C3 E5 7C

[...]

> I tried to convert everything to use the same function, and to
> make the optimization work correctly for that function. I didn't
> like the name %COERCE-TO-FUNCTION because %COERCE-CALLABLE-TO-FUNCTION,
> like CMU CL's %COERCE-TO-FUNCTION, does something different from
> (COERCE X 'FUNCTION). I don't remember whether that was the only
> reason that I changed the name, or whether I had other reasons as
> well.
> 
> Is that what you wanted to know?
Yes, thanks a lot!

> (I can try applying the CMU CL specials/closures patch myself, but
> first I'd like to finish testing and integrating the stuff in the big
> patch you sent me before. At the rate I'm going, that will take me at
> least two more days, and probably more.)
That is probably the best way to go.
I think applying the patch directly is not possible. I suspect the patch
is against DTC's experimental version, and that seems to be quite
different from the debian-sources.
I'll have to get a better understanding of how the compiler works,
before I can try to apply the patch. And I'll do that, before I go on.
So I hope you will succeed.


Cheers,

Martin

-- 
Homepage: http://www.atzmueller.net/
Email: martin@...