[Chicken-users] Re: Problems with run-time evals involving srfi-17..

[felix]

Daniel B. Faken wrote:
> Hello,
>
>   (this is with chicken 1.51)
>
>   A lot of my code uses run-time evals (I'm passing code around a 
> network) and srfi-17 (generalized set!).
>   The right combination of features to use in chicken has always been a 
> little obscure to me, but I've bumbled along by requiring almost 
> everything almost everywhere.. but now, I can't get this simple code to 
> run:

Well, using macros at runtime is rather unintuitive, I admit.

> -------------------------------------------------
> (declare (uses extras srfi-1 srfi-4
>                eval syntax-case) (run-time-macros))
> (require-extension chicken-more-macros)
>
> ;(install-highlevel-macro-system 'r5rs 'srfi-0 'extensions)
> (require-extension srfi-17)
>
> (define node-var-table (make-hash-table))
> (define node-var
>   (getter-with-setter
>    [lambda (name) (hash-table-ref node-var-table name)]
>    [lambda (name val) (hash-table-set! node-var-table name val)]) )
>
> (set! (node-var 'tracker-node?) #t)
> (set! (node-var 'gui-tracker-node?) #t)
> (set! (node-var 'vrpn-node?) #t)
>
> (eval `(set! (node-var 'vmd-node?) #t))
> ----------------------------------------------------
>
>   ... at least, when compiled: I get (also with various modifications)
> Error: invalid syntax: (set! (node-var (quote vmd-node?)) #t)
>
> (even if I use -H2 and/or -run-time-macros for csc (which shouldn't
> be necessary anyway))

The proper way to do it is like this:

(require-extension srfi-17)

(register-feature! 'srfi-17)

(define x (list 1 2 3))
(eval '(set! (cadr x) 99))
(print x)

If you want to use the hygienic macro system at runtime, I recommend
using the `-hygienic-at-run-time' option, unless the non-standard macros
are needed. Should that be the case, the best way is:

(require-extension syntax-case)
(install-highlevel-macro-system 'extensions)

The `register-feature!' above is all that's needed to tell the macro-expander
at runtime that non-identifiers should be allowed, the runtime-support code is 
already loaded.
(the `register-feature!' is in fact unnecessary and should be done by 
srfi-17-support.scm,
I have changed the srfi-17 egg accordingly)

The run-time-macros declaration only works for lowlevel macros.

Another point: your invocation of `install-highlevel-macro-system' is slightly
incorrect: you should pass _either_ 'r5rs or 'extensions, the two options are 
mutually
exclusive (I have added a clarification to the manual, since this wasn't that 
clear
from the documentation).

>  It works in csi unless I uncomment that
> (install-highlevel-macro-system..), which causes this mysterious error:
> ; loading /home/dfaken/.csirc ...
> ; loading /home/dfaken/lib/chicken/readline.so ...
> ; loading htcam/tst.scm ...
> ; loading /home/dfaken/lib/chicken/share/chicken/chicken-more-macros.scm 
> ...
> Error: unbound variable: srfi-17

Passing 'r5rs to `install-highlevel-macro-system' assumes you only want
R5RS (and it ignores the 'extensions), so `require-extension' is not recognized.
>   Is (require-extension ..) valid during eval-time?  I've used (require..)
> in the past with no problems.  

I recommend not to use require-extension at run-time.

> Are there any other gotcha's for using 
> macros at run-time vs. compile-time?  (I think I've seen all the sections 
> in the manual) 

Yes, there are several, but I think you already got caught by most of them. ;-)
The basic problem is that Chicken uses two different, mostly incompatible
macro systems.

>   Does anyone else do a lot of run-time macro-needing evals? :)

Normally I use the low-level macro system, exclisively and do the equivalent of

(declare (run-time-macros))
(include "chicken-more-macros")
(include "chicken-match-macros")  ; if needed

This includes all macro definitions in the compiled code and also will define
them at run-time for being used with eval.


cheers,
felix