bug#58066: pretty-print will make data modified by set-cdr! to random value

[Jean Abou Samra]

Le 24/09/2022 à 12:01, slbtty a écrit :
> GNU Guile 3.0.8
>
> Using `guile3 < file.scm` to run the code below will makes the result
> of tp unpredictable (pointing to very random things)
>
> (use-modules (ice-9 pretty-print))
> (define tl '(1 2))
> (define tp (cdr tl))
> (set-cdr! tp (cons 3 '()))
>    (pretty-print tl)
>    (pretty-print tp)
> ---
>
> However, if i use plain old (display tp), the code will out put
> expected valve (1 2 3) (3)
> (define tl '(1 2))
> (define tp (cdr tl))
> (set-cdr! tp (cons 3 '()))
>    (display tl) (newline)
>    (display tp) (newline)
>
> ---
>
> Context:
>
> I am using guile3's repl and whenever i access a variable's by typing
> its name after (set-cdr!), the result will goes wild.



Interesting. It looks like scm_is_mutable_pair is broken? It's
defined (in pairs.h) as

static inline int
scm_is_mutable_pair (SCM x)
{
  /* Guile embeds literal pairs into compiled object files.  It's not
     valid Scheme to mutate literal values.  Two practical reasons to
     enforce this restriction are to allow literals to share share
     structure (pairs) with other literals in the compilation unit, and
     to allow literals containing immediates to be allocated in the
     read-only, shareable section of the file.  Attempting to mutate a
     pair in the read-only section would cause a segmentation fault, so
     to avoid that, we really do need to enforce the restriction. */
  return scm_is_pair (x) && GC_is_heap_ptr (SCM2PTR (x));
}


I don't know why that breaks or how to fix it, though.

As you can read from the comment, this code is actually
invalid Scheme, since it mutates literal data. However,
it is true that Guile should raise an error for this,
not segfault (which is what it does for me) or start
printing random things.

If you define tl as (list 1 2 3) instead of '(1 2 3),
everything will be fine.