Remove and/or fix casts
cover Remove and/or fix casts
Hi Michael,
[PATCH 1/7] sock_diag.7: Remove unneeded casts
[PATCH 2/7] pthread_sigmask.3: Remove unneeded casts [PATCH 3/7] msgop.2: Remove unneeded casts
[PATCH 4/7] user_namespaces.7: Remove unneeded cast
[PATCH 5/7] dlopen.3: Remove unneeded cast
[PATCH 6/7] bsearch.3: Fix intermediate type and remove unneeded cast [PATCH 7/7] qsort.3: Fix casts
Here's a set of patches removing unneeded casts when they are unneeded or incorrect, and fixing those that are incorrect but can't be removed.
Regards,
Alex.
PATCH 1/7 sock_diag.7: Remove unneeded casts
The type struct sockaddr_nl * is implicitly casted to void *. Explicitly casting can silence warnings when mistakes are made, so it's better to remove those casts when possible.
--- man7/sock_diag.7 | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
man7/sock_diag.7 +2 -2
@@ -680,7 +680,7 @@ send_query(int fd) .iov_len = sizeof(req) }; struct msghdr msg = { - .msg_name = (void *) &nladdr, + .msg_name = &nladdr, .msg_namelen = sizeof(nladdr), .msg_iov = &iov, .msg_iovlen = 1
@@ -765,7 +765,7 @@ receive_responses(int fd) for (;;) { struct msghdr msg = { - .msg_name = (void *) &nladdr, + .msg_name = &nladdr, .msg_namelen = sizeof(nladdr), .msg_iov = &iov, .msg_iovlen = 1
Signature · 1 line
2.28.0
PATCH 2/7 pthread_sigmask.3: Remove unneeded casts
In reply to Alejandro ColomarThe type sigset_t * is implicitly casted to void *. Explicitly casting can silence warnings when mistakes are made, so it's better to remove those casts when possible.
--- man3/pthread_sigmask.3 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
man3/pthread_sigmask.3 +1 -1
@@ -154,7 +154,7 @@ main(int argc, char *argv[]) if (s != 0) handle_error_en(s, "pthread_sigmask"); - s = pthread_create(&thread, NULL, &sig_thread, (void *) &set); + s = pthread_create(&thread, NULL, &sig_thread, &set); if (s != 0) handle_error_en(s, "pthread_create");
Signature · 1 line
2.28.0
PATCH 3/7 msgop.2: Remove unneeded casts
In reply to Alejandro ColomarThe type struct msgbuf * is implicitly casted to const void *. Not only that, but the explicit cast to void * was slightly misleading.
Explicitly casting can silence warnings when mistakes are made, so it's better to remove those casts when possible.
--- man2/msgop.2 | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
man2/msgop.2 +2 -2
@@ -637,7 +637,7 @@ send_msg(int qid, int msgtype) snprintf(msg.mtext, sizeof(msg.mtext), "a message at %s", ctime(&t)); - if (msgsnd(qid, (void *) &msg, sizeof(msg.mtext), + if (msgsnd(qid, &msg, sizeof(msg.mtext), IPC_NOWAIT) == \-1) { perror("msgsnd error"); exit(EXIT_FAILURE);
@@ -650,7 +650,7 @@ get_msg(int qid, int msgtype) { struct msgbuf msg; - if (msgrcv(qid, (void *) &msg, sizeof(msg.mtext), msgtype, + if (msgrcv(qid, &msg, sizeof(msg.mtext), msgtype, MSG_NOERROR | IPC_NOWAIT) == \-1) { if (errno != ENOMSG) { perror("msgrcv");
Signature · 1 line
2.28.0
PATCH 4/7 user_namespaces.7: Remove unneeded cast
In reply to Alejandro ColomarCasting void * to struct child_args * is already done implicitly. Explicitly casting can silence warnings when mistakes are made, so it's better to remove those casts when possible.
--- man7/user_namespaces.7 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
man7/user_namespaces.7 +1 -1
@@ -1222,7 +1222,7 @@ proc_setgroups_write(pid_t child_pid, char *str) static int /* Start function for cloned child */ childFunc(void *arg) { - struct child_args *args = (struct child_args *) arg; + struct child_args *args = arg; char ch; /* Wait until the parent has updated the UID and GID mappings.
Signature · 1 line
2.28.0
PATCH 5/7 dlopen.3: Remove unneeded cast
In reply to Alejandro ColomarCasting void * to double (*cosine)(double) is already done implicitly.
I had doubts about this one, but gcc -Wall -Wextra didn't complain about it.
Explicitly casting can silence warnings when mistakes are made, so it's better to remove those casts when possible.
--- man3/dlopen.3 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
man3/dlopen.3 +1 -1
@@ -581,7 +581,7 @@ main(void) dlerror(); /* Clear any existing error */ - cosine = (double (*)(double)) dlsym(handle, "cos"); + cosine = dlsym(handle, "cos"); /* According to the ISO C standard, casting between function pointers and 'void *', as done above, produces undefined results.
Signature · 1 line
2.28.0
Re: [PATCH 5/7] dlopen.3: Remove unneeded cast
Hello Alex,
Quoted 17 lines · Alejandro Colomar
Casting
void *todouble (*cosine)(double)is already done implicitly.
I had doubts about this one, butgcc -Wall -Wextradidn't complain about it.
Explicitly casting can silence warnings when mistakes are made, so it's better to remove those casts when possible.Signed-off-by Alejandro Colomar <[email protected]>--- man3/dlopen.3 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)man3/dlopen.3 +0
-0
man3/dlopen.3 main(void) · line 587
Quoted 8 lines
man3/dlopen.3 +1
-1@@ -581,7 +581,7 @@ main(void) dlerror(); /* Clear any existing error */ - cosine = (double (*)(double)) dlsym(handle, "cos"); + cosine = dlsym(handle, "cos"); /* According to the ISO C standard, casting between function pointers and 'void *', as done above, produces undefined results.
This cast really is needed. See the comment just below, and also try compiling the code with your patch applied:
cc -pedantic -Wall prog.cd.c: In function ‘main’:
d.c:21:19: warning: ISO C forbids assignment between function pointer and ‘void *’ [-Wpedantic]
21 | cosine = dlsym(handle, "cos");
| ^
Thanks,
Michael
Signature · 3 lines
Michael Kerrisk Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/ Linux/UNIX System Programming Training: http://man7.org/training/
Re: [PATCH 5/7] dlopen.3: Remove unneeded cast
Hola Michael,
Quoted 32 lines · Michael Kerrisk (man-pages)
Hello Alex,
Quoted 17 lines · Alejandro Colomar
Casting
void *todouble (*cosine)(double)is already done implicitly.
I had doubts about this one, butgcc -Wall -Wextradidn't complain about it.
Explicitly casting can silence warnings when mistakes are made, so it's better to remove those casts when possible.Signed-off-by Alejandro Colomar <[email protected]>--- man3/dlopen.3 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)man3/dlopen.3 +0
-0man3/dlopen.3 main(void) · line 587
Quoted 8 lines
man3/dlopen.3 +1
-1@@ -581,7 +581,7 @@ main(void) dlerror(); /* Clear any existing error */ - cosine = (double (*)(double)) dlsym(handle, "cos"); + cosine = dlsym(handle, "cos"); /* According to the ISO C standard, casting between function pointers and 'void *', as done above, produces undefined results.This cast really is needed. See the comment just below, and also try compiling the code with your patch applied:
Quoted 3 lines
cc -pedantic -Wall
prog.cd.c: In function ‘main’:
d.c:21:19: warning: ISO C forbids assignment between function pointer
and ‘void *’ [-Wpedantic]
Quoted 2 lines
21 | cosine = dlsym(handle, "cos");
| ^Hmmm, not sure about it.
The thing is, standard C doesn't allow this, no matter how. POSIX does allow it, however.
The only thing with the casts is to avoid the warning, but they don't avoid the possible undefined behaviour (only in non-POSIX systems). But that warning, -pedantic, is specifically targeted to warn about whatever code that is not strict standard C, which this code isn't, so the warning is legit IMHO, and anyone using -pedantic would probably be warned about this line, and anyone not wanting to be warned about this line should probably disable -pedantic.
So, in POSIX, without -pedantic, that line without casts will result in correct code and no warnings, as expected.
And in non-POSIX, with -pedantic, that line without casts will correctly result in a warning.
And more importatnly, in non-POSIX, with -pedantic, that line with casts will result in no warnings but undefined results.
I'd say that no casting is less problematic than casting, although both have their problems.
Saludos,
Alex
Re: [PATCH 5/7] dlopen.3: Remove unneeded cast
Hi Michael,
Quoted 2 lines · Alejandro Colomar
so the warning is legit IMHO, and anyone using
-pedanticwould probably
want to
Quoted 2 lines
be warned about this line, and anyone not wanting to be warned about this line should probably disable
-pedantic.
Quoted 1 line · Michael Kerrisk (man-pages)
Quoted 1 line
I've applied all of these ,except 5/7 (I'll reply to that mail).
Actually all of the patches were applied, so probably you applied 5/7 by accident. My branch is up to date with upstream/master.
Quoted 1 line
Oops -- and I meant to say: thanks for the patches.
:-)
Cheers,
Alex.
Re: [PATCH 5/7] dlopen.3: Remove unneeded cast
Quoted 4 lines
Quoted 1 line
Quoted 1 line
I've applied all of these ,except 5/7 (I'll reply to that mail).
Actually all of the patches were applied, so probably you applied 5/7 by accident. My branch is up to date with upstream/master.
D'oh! I just now reverted that patch :-).
Cheers,
Michael
Signature · 3 lines
Michael Kerrisk Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/ Linux/UNIX System Programming Training: http://man7.org/training/
Re: [PATCH 5/7] dlopen.3: Remove unneeded cast
In reply to Alejandro ColomarHi Alex,
Quoted 38 lines · Alejandro Colomar
Hola Michael,
Quoted 32 lines · Michael Kerrisk (man-pages)
Hello Alex,
Quoted 17 lines · Alejandro Colomar
Casting
void *todouble (*cosine)(double)is already done implicitly.
I had doubts about this one, butgcc -Wall -Wextradidn't complain about it.
Explicitly casting can silence warnings when mistakes are made, so it's better to remove those casts when possible.Signed-off-by Alejandro Colomar <[email protected]>--- man3/dlopen.3 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)man3/dlopen.3 +0
-0man3/dlopen.3 main(void) · line 587
Quoted 8 lines
man3/dlopen.3 +1
-1@@ -581,7 +581,7 @@ main(void) dlerror(); /* Clear any existing error */ - cosine = (double (*)(double)) dlsym(handle, "cos"); + cosine = dlsym(handle, "cos"); /* According to the ISO C standard, casting between function pointers and 'void *', as done above, produces undefined results.This cast really is needed. See the comment just below, and also try compiling the code with your patch applied:
Quoted 3 lines
cc -pedantic -Wall
prog.cd.c: In function ‘main’:
d.c:21:19: warning: ISO C forbids assignment between function pointer
Quoted 7 lines
and ‘void *’ [-Wpedantic]
Quoted 2 lines
21 | cosine = dlsym(handle, "cos"); | ^Hmmm, not sure about it.
The thing is, standard C doesn't allow this, no matter how.
Agreed.
Quoted 1 line
POSIX does allow it, however.
Yes, POSIX is explicit on this point, and the specification gives an example of the use casts in the manner shown in the manual page.
Quoted 2 lines
The only thing with the casts is to avoid the warning, but they don't avoid the possible undefined behaviour (only in non-POSIX systems).
Yes. (But, dlopen() is a "UNIX-only" API, and thus non-POSIX doesn't matter here.)
Quoted 9 lines
But that warning,
-pedantic, is specifically targeted to warn about whatever code that is not strict standard C, which this code isn't, so the warning is legit IMHO, and anyone using-pedanticwould probably be warned about this line, and anyone not wanting to be warned about this line should probably disable-pedantic.So, in POSIX, without
-pedantic, that line without casts will result in correct code and no warnings, as expected.
Quoted 8 lines
And in non-POSIX, with
-pedantic, that line without casts will correctly result in a warning.And more importatnly, in non-POSIX, with
-pedantic, that line with casts will result in no warnings but undefined results.I'd say that no casting is less problematic than casting, although both have their problems.
Two things:
* The standard uses the casts, and allows the extension on top of what the C standard permits here. So, I think the manuial page better use the casts also.
* Sometimes people have to compile a large body of code using certain compiler options, perhaps including "-pedantic", so they need to at least be aware of the warning that the cast may incur.
To address the second point, I make use of the appropriate pragma, to eliminate the waring from -pedantic:
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wpedantic"
funcp = (void (*)(void)) dlsym(libHandle, name);
#pragma GCC diagnostic pop
Thanks,
Michael
Signature · 3 lines
Michael Kerrisk Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/ Linux/UNIX System Programming Training: http://man7.org/training/
Re: [PATCH 5/7] dlopen.3: Remove unneeded cast
Hi Michael,
Quoted 5 lines · Michael Kerrisk (man-pages)
Quoted 2 lines
The only thing with the casts is to avoid the warning, but they don't avoid the possible undefined behaviour (only in non-POSIX systems).
Yes. (But,
dlopen()is a "UNIX-only" API, and thus non-POSIX doesn't matter here.)
Agreed. The cast is useful in this case.
The warning that is completely useless this time.
Cheers,
Alex.
PATCH 6/7 bsearch.3: Fix intermediate type and remove unneeded casts
In reply to Alejandro ColomarCasting const void * to struct mi * should result in a warning if done implicitly. The explicit cast was probably silencing that warning. const can and should be kept.
Now, casting const void * to const struct mi * is done implicitly.
--- man3/bsearch.3 | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
man3/bsearch.3 +2 -2
@@ -114,8 +114,8 @@ struct mi { static int compmi(const void *m1, const void *m2) { - struct mi *mi1 = (struct mi *) m1; - struct mi *mi2 = (struct mi *) m2; + const struct mi *mi1 = m1; + const struct mi *mi2 = m2; return strcmp(mi1\->name, mi2\->name); }
Signature · 1 line
2.28.0
PATCH 7/7 qsort.3: Fix casts
In reply to Alejandro Colomarp1 (and p2 too) is const void * and it comes from a const char ** (for legacy reasons, argv is not const but should be treated as if it were). That means, the ultimate char is const:
"a pointer to a pointer to a const char".
Let's see what is going on before the fix first, and then the fix.
Before the fix:
(char *const *) (I removed the space on purpose) casts p1 to be "a pointer to a const pointer to a non-const char". That's clearly not what it originally was.
Then we dereference, ending with a char *const, which is "a const pointer to a non-const char". But given that the pointer value is passed to a function, const doesn't make sense there, because the function will already take a copy of it, so it is impossible to modify the pointer itself.
The fix:
(const char **) The only thing that is const is the ultimate char, which is the only thing that matters, because it is the only thing strcmp(3) has access to (everything else, i.e. the pointers, are copies).
Then, after the dereference we end up with const char *, the type of argv (more or less, as previously noted), which is also the type of the arguments to strcmp(3).
--- man3/qsort.3 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
man3/qsort.3 +1 -1
@@ -137,7 +137,7 @@ cmpstringp(const void *p1, const void *p2) pointers to char", but strcmp(3) arguments are "pointers to char", hence the following cast plus dereference */ - return strcmp(* (char * const *) p1, * (char * const *) p2); + return strcmp(*(const char **) p1, *(const char **) p2); } int
Signature · 1 line
2.28.0
Michael Kerrisk (man-pages)
AppliedHello Alex,
Quoted 10 lines · Alejandro Colomar
Hi Michael,
[PATCH 1/7] sock_diag.7: Remove unneeded casts
[PATCH 2/7] pthread_sigmask.3: Remove unneeded casts [PATCH 3/7] msgop.2: Remove unneeded casts
[PATCH 4/7] user_namespaces.7: Remove unneeded cast
[PATCH 5/7] dlopen.3: Remove unneeded cast
[PATCH 6/7] bsearch.3: Fix intermediate type and remove unneeded cast [PATCH 7/7] qsort.3: Fix casts
Quoted 3 lines
Here's a set of patches removing unneeded casts when they are unneeded or incorrect, and fixing those that are incorrect but can't be removed.
I've applied all of these ,except 5/7 (I'll reply to that mail). Mea culpa for 2 and 4...
Thanks,
Michael
Signature · 3 lines
Michael Kerrisk Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/ Linux/UNIX System Programming Training: http://man7.org/training/
Quoted 13 lines · Michael Kerrisk (man-pages)
Hello Alex,
Quoted 10 lines · Alejandro Colomar
Hi Michael,
[PATCH 1/7] sock_diag.7: Remove unneeded casts
[PATCH 2/7] pthread_sigmask.3: Remove unneeded casts [PATCH 3/7] msgop.2: Remove unneeded casts
[PATCH 4/7] user_namespaces.7: Remove unneeded cast
[PATCH 5/7] dlopen.3: Remove unneeded cast
[PATCH 6/7] bsearch.3: Fix intermediate type and remove unneeded cast [PATCH 7/7] qsort.3: Fix casts
Quoted 6 lines
Quoted 3 lines
Here's a set of patches removing unneeded casts when they are unneeded or incorrect, and fixing those that are incorrect but can't be removed.
I've applied all of these ,except 5/7 (I'll reply to that mail). Mea culpa for 2 and 4...
Oops -- and I meant to say: thanks for the patches.
Cheers,
Michael
Signature · 3 lines
Michael Kerrisk Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/ Linux/UNIX System Programming Training: http://man7.org/training/