This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [patch 1/9]#2 Rename `enum target_signal' to target_signal_t
- From: Pedro Alves <pedro at codesourcery dot com>
- To: Jan Kratochvil <jan dot kratochvil at redhat dot com>
- Cc: gdb-patches at sourceware dot org, Joel Brobecker <brobecker at adacore dot com>, Eli Zaretskii <eliz at gnu dot org>, Mark Kettenis <mark dot kettenis at xs4all dot nl>
- Date: Wed, 1 Sep 2010 20:59:36 +0100
- Subject: Re: [patch 1/9]#2 Rename `enum target_signal' to target_signal_t
- References: <E1Oq55N-0006ia-B0@fencepost.gnu.org> <201009012007.55983.pedro@codesourcery.com> <20100901191843.GA27558@host1.dyn.jankratochvil.net>
On Wednesday 01 September 2010 20:18:43, Jan Kratochvil wrote:
> On Wed, 01 Sep 2010 21:07:55 +0200, Pedro Alves wrote:
> > I wonder if switching on "-Wc++-compat" wouldn't catch these (at
> > least with recent enough gccs) and be more productive than
> > switching to a struct. /me ducks.
>
> I wanted to argue with this great point myself but during my tests (gcc-4.5)
> enum unfortunately IS compatible with int even in C++ (tried -Wall/-pedantic
> etc.). I haven't checked the C++ spec.
I think "int signal = TARGET_SIGNAL_TRAP" would be valid C++, while
"enum gdb_signal signal = SIGTRAP" would not. I thought the latter would
be the most common scenario to catch though. Maybe not.
Anyway, personally, I'd just do a enum target_signal/enum gdb_signal
rename, and stay with that.
But, here's another idea of how to get compiler warnings/errors,
that I think is more transparent to code throughout:
/* An empty struct. It's the instances we care about. */
struct gdb_signal_1
{
};
/* An array of all defined gdb signals. */
const struct gdb_signal_1 gdb_signals[NUM_GDB_SIGNALS];
/* gdb code uses gdb_signal, not gdb_signal_1. gdb_signal is a
pointer. */
typedef struct gdb_signal_1 * gdb_signal;
/* Define the constants. */
#define TARGET_SIGNAL_0 (&gdb_signals[0])
#define TARGET_SIGNAL_TRAP (&gdb_signals[5])
#define TARGET_SIGNAL_FOO (&gdb_signals[FOO])
...
Then, you can do this:
gdb_signal signal = TARGET_SIGNAL_TRAP;
and still do this:
for (gdb_signal foo = TARGET_SIGNAL_0; sig < TARGET_SIGNAL_LAST; sig++)
do_things(foo);
... all as before. But these:
int sig = TARGET_SIGNAL_TRAP;
gdb_signal sig = SIGTRAP;
gdb_signal sig = TARGET_SIGNAL_TRAP;
if (sig < SIGTRAP)
do_things ();
... give out a warnings, fatal with -Werror.
Getting at the signal integer is simply:
#define GDB_SIGNAL_NUMBER(gdb_sig) \
(((gdb_sig) - TARGET_SIGNAL_0) / sizeof (gdb_sig))
All the other macros TARGET_SIGNAL_EQ|NE|GT|... disappear.
Here's a compilable example:
$ cat sig.c
#include <signal.h>
struct gdb_signal_1 {};
typedef struct gdb_signal_1 * gdb_signal;
struct gdb_signal_1 signals[10];
#define TARGET_SIGNAL_TRAP (&signals[5])
int main ()
{
gdb_signal sig = TARGET_SIGNAL_TRAP;
int sig2 = SIGTRAP;
if (sig == sig2)
{
}
sig = sig2;
sig2 = sig;
return 0;
}
$ gcc sig.c -o sig.o -c -Wall
sig.c: In function âmainâ:
sig.c:16: warning: comparison between pointer and integer
sig.c:20: warning: assignment makes pointer from integer without a cast
sig.c:21: warning: assignment makes integer from pointer without a cast
--
Pedro Alves