extract_data_from_object considered dangerous

extract_data_from_object considered dangerous

[Stefan Monnier]

I was just looking at the calling convention of extract_data_from_object
and it seems intrinsically dangerous:

    It returns a char* into a Lisp_Object without associating it with
    any specific Lisp_Object

In some cases, the Lisp_Object into which the result points is freshly
allocated during the call, so once we return from the function, there is
no Lisp_Object reference to it anywhere any more to keep it from
being GC'd.


        Stefan

Re: extract_data_from_object considered dangerous

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Date: Sun, 22 Sep 2019 09:39:54 -0400
> 
>     It returns a char* into a Lisp_Object without associating it with
>     any specific Lisp_Object
> 
> In some cases, the Lisp_Object into which the result points is freshly
> allocated during the call, so once we return from the function, there is
> no Lisp_Object reference to it anywhere any more to keep it from
> being GC'd.

The original idea was to use the pointer right away, so that GC
couldn't happen.  If we want to lift this restriction, I think the
function should return a Lisp object, not a char *.

Re: extract_data_from_object considered dangerous

[Stefan Monnier]

> The original idea was to use the pointer right away, so that GC
> couldn't happen.

Indeed, I think that it might be the case that currently this problem is
only hypothetical, but relying on "the GC probably can't happen now" is
too fragile for my taste.

> If we want to lift this restriction, I think the
> function should return a Lisp object, not a char *.

That's also my impression, but I was wondering is there was a good
reason why it wasn't done this way in the first place.


        Stefan

Re: extract_data_from_object considered dangerous

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Cc: emacs-devel@gnu.org
> Date: Sun, 22 Sep 2019 16:53:05 -0400
> 
> > If we want to lift this restriction, I think the
> > function should return a Lisp object, not a char *.
> 
> That's also my impression, but I was wondering is there was a good
> reason why it wasn't done this way in the first place.

AFAIR, the function was created by consolidating many repeated in-line
code fragments.