Re: [bug-gettext] xgettext segmentation fault javascript

[Jesper Fehrlund]

On 29/10/14 03:37, Daiki Ueno wrote:
> Jesper Fehrlund <address@hidden> writes:
>
> > I'm no c expert by any means, but does this not run xstrdup on a
> > dangling pointer?
> > (since best_cp->msgid and best_cp->msgid_plural points to the same
> > address which has been released in remember_a_message)
> Good point, thanks.  That part is really tricky.  In the code:
>
>    mp = remember_a_message (..., best_cp->msgid, ...);
>    if (mp != NULL && best_cp->msgid_plural != NULL)
>      remember_a_message_plural (..., best_cp->msgid_plural, ...);
>
> If MP is non-NULL, that means best_cp->msgid is saved in MP, and
> remember_a_message_plural can still refer to the address pointed by
> best_cp->msgid.
>
> Of course, it is an implementation detail and the confusing dangling
> pointer should be avoided.  I'm thinking to cleanup the code, also by
> moving the code conversion stuff to earlier phase.
>
> Regards,
> --
> Daiki Ueno
Hmm, are you sure?

The first time a message is encountered, I can see that this would be 
the case.
But, if the message already exists, then this code is executed:

  mp = message_list_search (mlp, msgctxt, msgid);
  if (mp != NULL)
    {
      if (msgctxt != NULL)
        free (msgctxt);
      free (msgid);
      for (i = 0; i < NFORMATS; i++)
        is_format[i] = mp->is_format[i];
      do_wrap = mp->do_wrap;
    }

And, as far as I can tell, at this point mp contains a string that is 
equal to msgid (before free) yes, but it's not the same pointer.
I suppose in the patch it should be fine to use mp->msgid instead of 
best_cp->msgid_plural for the copy

Regards,
--
Jesper Fehrlund