Re: compiling with -DSCM_DEBUG=1

[Ken Raeburn]

On Sep 3, 2009, at 17:04, Ken Raeburn wrote:
> [...] Scheme compilation bug is still there, though.

I'm still not sure where the bug is, but here's what I've traced  
through so far;

The error is happening in eval.i.c, line number in the high 800s (I've  
got a bunch of tracing lines added in my copy):

  while (!scm_is_null (args))
    {
      /* More arguments than specifiers => CLASS != ENV */
      SCM class_of_arg = scm_class_of (SCM_CAR (args));
      if (!scm_is_eq (class_of_arg, SCM_CAR (z)))   // <== HERE!
        goto next_method;
      args = SCM_CDR (args);
      z = SCM_CDR (z);
    }

At this line, at one point during the compilation of debugger.scm, z  
is not a pair.  (Earlier the problem was triggering somewhere else for  
me, now it's in debugger.scm; not sure why.)

Some tracing of the containing do loop indicates that these are the  
sorts of values used before the error happens:

do-loop with hash_value 0 arg1 (#<struct 10a6370:1092060> #<struct  
10a6d40:10b94e0>) z (no-method)
do-loop with hash_value 0 arg1 (#<struct 10a6370:115cb0> #<struct  
10a6d40:115a60>) z (no-method)
do-loop with hash_value 0 arg1 (#<struct 10a6370:1010de0> #<struct  
10a6d40:1010db0>) z (#<struct 10a73a0:10a6370> #<struct  
10a0bc0:10a6d40> . #<procedure #f (#{gf\ 2116}# #{m\ 2117}#)>)
do-loop with hash_value 0 arg1 (#<primitive-generic eqv?> #<struct  
10a6d40:1099310>) z (#<struct 10a73a0:10a6370> #<struct  
10a0bc0:10a6d40> . #<procedure #f (#{gf\ 2116}# #{m\ 2117}#)>)
[...]
do-loop with hash_value 3 arg1 (#<<generic> apply-generic (0)>  
#<<method> (<generic> <top>) 10bbd80>) z (#<<entity-class> <generic>  
10a59a0> #<<class> <method> 10a6420> . #<procedure #f (#{gf\ 2116}# #{m 
\ 2117}#)>)

But then:
do-loop with hash_value 0 arg1 (#<<entity-class> <generic> 10a59a0>  
#:name compute-applicable-methods) z (#<<class> <entity-class>  
10a6910> #<<class> <keyword> 10a3b90> . #<procedure #f (#{class\  
2660}# . #{initargs\ 2659}#)>)

Here, the arg list has three elements instead of the two in the calls  
shown above, but the initial 'z' value isn't any longer; they're  
walked in parallel in that while loop, 'z' runs out first, and SCM_CAR  
is applied to a non-pair:

Non-pair accessed with SCM_C[AD]R: `#<procedure #f (#{class\ 2660}# .  
#{initargs\ 2659}#)>'

Assuming the bug doesn't jump out as obvious to someone from this,  
I'll dig into it a little more later.

Ken