On Mon, Jan 06, 2014 at 11:31:18AM +0100, Pino Toscano wrote:
On Tuesday 31 December 2013 14:27:11 Richard W.M. Jones wrote:
> This patch assumes that iconv doesn't actually modify the
> input buffer, even though it is declared as char *.
> ---
> [...]
> @@ -51,10 +51,11 @@ _hivex_recode (char *input_encoding, const char
> *input, size_t input_len, errno = err;
> return NULL;
> }
> - char *inp = input;
> + const char *inp = input;
> char *outp = out;
>
> - size_t r = iconv (ic, &inp, &inlen, &outp, &outlen);
> + /* Surely iconv doesn't really modify the input buffer? XXX */
> + size_t r = iconv (ic, (char **) &inp, &inlen, &outp, &outlen);
> if (r == (size_t) -1) {
This (the constness of the input argument) is the difference between the
iconv implementation provided in GNU libc and the other ones, even
though it reflects POSIX [1].
I found there's an "entertaining" bug about this:
https://sourceware.org/bugzilla/show_bug.cgi?id=2962
The iconv.m4 module in gnulib checks for this too, and
AC_DEFINE's
ICONV_CONST as "const" (or empty) depending on the actual prototype of
iconv(3).
[1]
http://pubs.opengroup.org/onlinepubs/9699919799/functions/iconv.html
I committed this patch [with the (char **) cast]. Using gnulib's
ICONV_CONST is possible, although I suppose it wouldn't make much
difference to us.
Rich.
--
Richard Jones, Virtualization Group, Red Hat
http://people.redhat.com/~rjones
Fedora Windows cross-compiler. Compile Windows programs, test, and
build Windows installers. Over 100 libraries supported.
http://fedoraproject.org/wiki/MinGW