Re: locale-dependent assertion failure in gawk 3.1.5

[Aharon Robbins]

I have not yet read all this thread. The intent is that n->wstptr is NOT valid
if the WSTRCUR bit is clear. Or to flip it around, only if the bit is set should
you mess with n->wstptr.

I have cleaned up the WSTRCUR code in my code base, but have not posted diffs.

IF I can steal the time over the next few weeks to clear my backlog of
unapplied patches I will start an alpha cycle.  Unfortunately, that's
a rather big "if".

Arnold

> Date: Mon, 10 Jul 2006 08:58:21 -0400
> From: "Andrew J. Schorr" <address@hidden>
> Subject: Re: locale-dependent assertion failure in gawk 3.1.5
> To: address@hidden
> Cc: =?iso-8859-1?Q?J=FCrgen?= Kahrs <address@hidden>
>
> On Fri, Jul 07, 2006 at 06:05:00PM -0400, Andrew J. Schorr wrote:
> > Given that NODE is a union, I'm not clear on whether it's OK to
> > test the wstptr value if the WSTRCUR flag is not set.  I guess
> > one has to assume that wstptr may contain random garbage if WSTRCUR
> > is not set; is that the correct approach?  Or would it be safe to
> > assume that if STRCUR is set, then
> >    (wstptr != NULL) <==> ((flags & WSTRCUR) != 0)?
>
> In particular, I'm not certain that I understand the model implied
> by this code at the top of node.c:str2wstr():
>
> NODE *
> str2wstr(NODE *n, size_t **ptr)
> {
>       ...
>         if ((n->flags & WSTRCUR) != 0) {
>                 if (ptr == NULL)
>                         return n;
>         }
>
>         if (n->wstptr != NULL) {
>                 free(n->wstptr);
>                 n->wstptr = NULL;
>                 n->wstlen = 0;
>         }
>
> Suppose that ((n->flags & WSTRCUR) == 0) (the flag bit is not set).  In that
> case, is it valid to access n->wstptr?  I had imagined from reviewing other
> parts of the code that we should consider n->wstptr undefined if the WSTRCUR
> flag is not set...
>
> In other words, would it be more correct to write:
>
>         if ((n->flags & WSTRCUR) != 0) {
>                 if (ptr == NULL)
>                         return n;
>
>               assert(n->wstptr != NULL);
>               free(n->wstptr);
>               /* no need to reset wstptr and wstlen since they will
>                  be set below */
>       }
>
> Or are there some well-defined circumstances under which it is OK to presume
> that the wstptr field is valid even if the WSTRCUR flag is not set?  And
> if so, is that concisely expressible in code (or a comment)?
>
> Regards,
> Andy