diff options
author | Pedro Alves <palves@redhat.com> | 2013-03-11 12:22:20 +0000 |
---|---|---|
committer | Pedro Alves <palves@redhat.com> | 2013-03-11 12:22:20 +0000 |
commit | 39086a0e1346572fac07542ede2a1ef153308a2a (patch) | |
tree | 55287581f85b93530d40c06578f98c07ba535cac /gprof | |
parent | 905851752e89f4fa2cc13e6df0f2dbd22c82f623 (diff) | |
download | gdb-39086a0e1346572fac07542ede2a1ef153308a2a.zip gdb-39086a0e1346572fac07542ede2a1ef153308a2a.tar.gz gdb-39086a0e1346572fac07542ede2a1ef153308a2a.tar.bz2 |
Avoid invalid pointer to pointer conversions.
Casts between 'char **' <-> 'unsigned char **' and 'char **' <-> const
char **' are actually invalid:
http://gcc.gnu.org/ml/gcc-help/2013-03/msg00118.html
In a nutshell, char (and variants) can alias anything, but pointers to
chars get no special treatment (cf. C99/N1256, 6.5/7).
Turns out older gcc's actually warn/complain on these constructs,
though newer one's don't:
http://sourceware.org/ml/gdb-patches/2013-03/msg00429.html
http://sourceware.org/ml/gdb-patches/2013-03/msg00430.html
This patch fixes the cases I added last week. It also fixes one other
preexisting case in charset.c, though it seems even older gccs don't
complain of char * <-> const char * aliasing.
Tested on x86_64 Fedora 17.
gdb/
2013-03-11 Pedro Alves <palves@redhat.com>
* charset.c (convert_between_encodings): Don't cast between
different pointer to pointer types. Instead, make the 'inp' local
be of the type iconv expects.
(wchar_iterate): Don't cast between different pointer to pointer
types. Instead, use new pointer local of the type iconv expects.
* target.c (target_read_stralloc, target_fileio_read_stralloc):
Add new local of type char pointer, and use it to get a
char/string view of the byte buffer, instead of casting between
pointer to pointer types.
Diffstat (limited to 'gprof')
0 files changed, 0 insertions, 0 deletions