Re: [PATCH 9/9] server: Avoid using pointer value after realloc.

[Rémi Bernon]

On 6/3/22 11:39, Dmitry Timoshkov wrote:
> Rémi Bernon  wrote:
> > - char *buf, *cur, *tmp;
> > -
> >
> > char *buf, *cur;
> >  int count = 0, buf_size = 16 * sizeof(struct hardware_msg_data);
> >
> > if (!req->buffer_size) buf = NULL;
> >
> > @@ -3373,13 +3373,13 @@ DECL_HANDLER(get_rawinput_buffer)
> >           if (cur + data->size > buf + buf_size)
> >           {
> >               buf_size += buf_size / 2 + extra_size;
> > -
> >        if (!(tmp = realloc( buf, buf_size )))
> > -
> >        cur = (char *)(cur - buf);
> > -
> >        if (!(buf = realloc( buf, buf_size )))
> >         {
> >             set_error( STATUS_NO_MEMORY );
> >             return;
> >         }
> > -
> >        cur = tmp + (cur - buf);
> > -
> >        buf = tmp;
> > -
> >        cur = buf + (size_t)cur;
> >     }
>
> Reusing 'cur' as an offset doesn't look very elegant to me. Perhaps
> a new variable to hold the offset could be more appropriate here?

I actually agree, I considered renaming the variable to "pos", which 
could maybe do better as both a position or an offset in the buffer, but 
maybe changing the code to use buf + offset everywhere would be better.

It was a bigger diff though, so it made the change less obvious.