bug#12929: case-lambda*

[ludo@gnu.org (Ludovic Courtès)]

Hi,

Indeed, there’s something wrong.  The “Odd length” error comes from the
‘bind-kwargs’ corresponding to the first case of the ‘case-lambda*’:

--8<---------------cut here---------------start------------->8---
scheme@(guile-user)> (case-lambda*
	     ((x #:key y)      2)
	     ((x y #:key z)    3))
$63 = #<procedure 22a0d40 at <current input>:4761:0 (x #:key y) | (x y #:key z)>
scheme@(guile-user)> ($63 1 2  #:z 3)
<unnamed port>:4761:0: In procedure #<procedure 22a0d40 at <current input>:4761:0 (x #:key y) | (x y #:key z)>:
<unnamed port>:4761:0: In procedure #<procedure 22a0d40 at <current input>:4761:0 (x #:key y) | (x y #:key z)>: Odd length of keyword argument list

Entering a new prompt.  Type `,bt' for a backtrace or `,q' to continue.
scheme@(guile-user) [1]> ,proc
$64 = #<procedure 22a0d40 at <current input>:4761:0 (x #:key y) | (x y #:key z)>
scheme@(guile-user) [1]> ,x $64
Disassembly of #<procedure 22a0d40 at <current input>:4761:0 (x #:key y) | (x y #:key z)>:

   0    (br-if-nargs-lt 0 1 :L49541)    ;; -> 34
   6    (bind-optionals/shuffle 0 1 0 1 0 2)
  13    (bind-kwargs 0 1 0 2 0)         
  19    (reserve-locals 0 2)            
  22    (local-bound? 1)                
  24    (br-if :L49542)                 ;; -> 31
  28    (make-false)                    
  29    (local-set 1)                   ;; `y'
  31    (make-int8 2)                   ;; 2
  33    (return)                        
  34    (assert-nargs-ge 0 2)           
  37    (bind-optionals/shuffle 0 2 0 2 0 3)
  44    (bind-kwargs 0 2 0 3 0)         
  50    (reserve-locals 0 3)            
  53    (local-bound? 2)                
  55    (br-if :L49543)                 ;; -> 62
  59    (make-false)                    
  60    (local-set 2)                   ;; `z'
  62    (make-int8 3)                   ;; 3
  64    (return)                        

scheme@(guile-user) [1]> ,regs
  Registers:
  $65 = ip = 19
  $66 = sp = #x2dacd58
  $67 = fp = #x2dacd38
--8<---------------cut here---------------end--------------->8---

Seems to me that for this to work, we’d need a variant of ‘bind-kwargs’,
say ‘br-if-kwargs-match’, that allows a jump to the next case upon
mismatch.

At least, I can’t think of any other way to generate a prelude that
would work (search for <glil-kw-prelude> in compile-assembly.scm.)

Thoughts?

Thanks,
Ludo’.