Re: [PATCH v2 2/3] textconv: support for blame

From: Matthieu Moy <matthieu.moy@grenoble-inp.fr>
Date: Jun 15, 2010, 10:51 UTC
Message-ID: <vpqbpbc4lh3.fsf@bauges.imag.fr>
In-Reply-To: <192517e06785fed4fa799bee9a11ae28@ensimag.fr>

bonneta <bonneta@ensimag.fr> writes:

Show 6 quoted lines

> But we have to do:
> textconv_object(read_from, null_sha1, &buf.buf, (unsigned long *)
> &buf.len))
> where buf.len is size_t.
>
> Is that ok?

I don't think it fixes the problem. You're assuming sizeof(unsigned long) == sizeof(size_t), otherwise, textconv_object will write the incorrect number of bytes at the given adress.

If you have to use this pass-by-adress, you want

size_t buf_len; /* textconv_object needs a last parameter of type
                   (size_t *) */
textconv_object(..., &buf_len); /* <-- no cast here */
buf.len = buf_len; /* This is a cast, but not a pointer cast. The
                      compiler will do the actual conversion if
                      needed (while pointer casts are just a matter of
                      typing, the generate no code). */

-- 
Matthieu Moy
http://www-verimag.imag.fr/~moy/