Re: segfault in CVS HEAD

[Dirk Herrmann]

Han-Wen Nienhuys wrote:

>  address@hidden writes:
>
> > address@hidden writes:
> >
> >> byrd:~/usr/src/guile/libguile$ cat t.scm (defiaine blub (cons
> >> (quote (1 . 2)) 2))
>>>
> >> for some reason, GUILE unmemoizes the procedure during print; of
> >> course, the expression is not memoized, since evaluation stops
> >> after "defiaine". I've seen similar problems with invocations of
> >> procedure-source on valid code.

Sorry for answering so late. The problem is due to some changes that I 
had introduced.

>  I've implemented a kludge in CVS: the for loop now has a SCM_CONSP as
>  the condition. It produces incorrect output in the example that I
>  sent, but at least it doesn't crash.

I will fix it. Sorry again for the delay.

>  BTW,
>
>  why are vectors quoted in the unmemoize output? from
>  unmemoize_expression():

Short answer: To make the unmemoized output R5RS compliant.

Long answer, for everyone interested: According to R5RS, vector 
constants must be quoted. The reason is the following:
 #((1 + 2))
evaluates to
 #((1 + 2))
rather than
 #(3)
To make this behaviour more obvious, R5RS demands to write
 '#((1 + 2))

Guile's executor does not need to see the quotes. That is, we strip away 
the quotes during memoization since this gives a slight performance 
improvement when dealing with vector constants. When unmemoizing the 
code again, however, in order to produce syntactically correct 
unmemoized code, the required quote is added again. Please compare the 
following:

guile> (define (foo) #(1))
guile> (foo)
#(1)
guile> (procedure-source foo)
(lambda () (quote #(1)))

guile> (define (bar) '#(1))
guile> (bar)
#(1)
guile> (procedure-source bar)
(lambda () (quote #(1)))

Best regards
Dirk Herrmann