[mercury-users] Circular lists

[Thomas Charles CONWAY]

Richard A. O'Keefe, you write:
> 
> 	I guess we could emit a warning every time you return a string from C
> 	code, or pass or return a complex type from C code,
> 
> Why are these things dangerous?  In Quintus Prolog, passing strings in either
> direction was guaranteed safe, thanks to copying.  In ESH, a very nice Scheme
> system from Sun, passing C data structures to Scheme and back again works
> perfectly safely.

Yes, we could fix the returning string problem by copying, though
currently we don't. Actually, we probably should (at least for the
"user" C interface, rather than the "implementor's" C interface) -
when I wrote the tcl/tk binding, there was indeed a bug where strings
were changing "mysteriously" - I wasn't copying the result string
from the interpreter, just assigning the pointer....

> 
> 	but to do anything
> 	more would require impossible analysis of the C code itself (and
> 	separate compilation only makes the problems worse).
> 
> You apparently envisage checking arbitrary code;
> I envisage checking _checkable_ code.  If the code is for some reason
> impossible to analyse, that is in itself a danger sign.

One of the possible pieces of work on the C interface that Tyson
alluded to as summer work for a student was to automatically
generate interface predicates directly from the C prototypes,
although this has problems to do with memory mangagement and
C's conflation of passing in-by-reference and out-by-reference.

Thomas
-- 
ZZ:wq!
^X^C
Thomas Conway               				      conway at cs.mu.oz.au
AD DEUM ET VINUM	  			      Every sword has two edges.