Re: [Bug-wget] [PATCH] Fix some clang-analyzer warnings

[Darshit Shah]

On 11/18, Giuseppe Scrivano wrote:
> Tim Rühsen <address@hidden> writes:
>
> > Fixes these warnings:
> >
> >         gnutls.c:457:3: warning: Value stored to 'err' is never read
> >           err = 0;
> >
> >         http-ntlm.c:477:5: warning: Value stored to 'size' is never read
> >           size = (size_t) snprintf (ntlmbuf, sizeof(ntlmbuf),
> >
> >         http.c:1479:3: warning: Attempt to free released memory
> >           xfree_null (hs->error);
> >
> > The last one *might* result in a crash under special circumstances.
> >
> >
> > I think we should just have one xfree() macro instead of two (xfree and
> > xfree_null, some parts of the code even use free() directly).
> >
> > IMHO, a free'd pointer should always be set to NULL afterwards (as a good
> > programming convention). I suggest the following macro
> >
> > #define xfree(a) do { if (a) { free ((void *) (a)); a = NULL; } } while (0)
>
> I think you can skip the check, gnulib ensures that free(NULL) is a
> no-op.
I just hit this issue. (Lucky timing, I guess)

It seems to me that you and Tim agreed upon the same thing. We should have only 
a xfree() macro since gnulib will ensure that a NULL free(NULL) is a no-op.

However, the xfree() macro should explicitly set the pointer to NULL. That does 
seem like sound programming practice. And I just hit a snag with valgrind where 
it was complaining about a double free. But setting the pointer to NULL 
explicitly silenced it.

--
Thanking You,
Darshit Shah

pgpmMY3O8oFmF.pgp
Description: PGP signature