Re: [PATCH] Use __nonnull for the epoll_wait(2) family of syscalls

[Alejandro Colomar <alx.manpages@gmail.com>]

[-- Attachment #1.1: Type: text/plain, Size: 3574 bytes --]

Hi Adhemerval!

On 5/29/23 19:39, Adhemerval Zanella Netto wrote:
> In fact, checking I am seeing a regression:
> 
> ../sysdeps/unix/sysv/linux/tst-epoll.c: In function ‘do_test’:
> ../sysdeps/unix/sysv/linux/tst-epoll.c:194:11: error: argument 2 null where non-null expected [-Werror=nonnull]
>   194 |   int r = epoll_pwait2 (-1, NULL, 0, NULL, NULL);
>       |           ^~~~~~~~~~~~
> In file included from ../include/sys/epoll.h:2,
>                  from ../sysdeps/unix/sysv/linux/tst-epoll.c:27:
> ../sysdeps/unix/sysv/linux/sys/epoll.h:144:12: note: in a call to function ‘epoll_pwait2’ declared ‘nonnull’
>   144 | extern int epoll_pwait2 (int __epfd, struct epoll_event *__events,
>       |            ^~~~~~~~~~~~
> cc1: all warnings being treated as errors
> 
> And I am not sure why it was not caught by buildbots.

I didn't catch it either.  I didn't run tests, since I thought if anything failed, it
would probably be at compilation stage, so I only compiled.  Not sure about the
buildbots, though; those should have caught it.

> 
> The check is only for test for epoll_pwait2 support, so I think it would be simpler to
> just suppress the warning:
> 
> diff --git a/sysdeps/unix/sysv/linux/tst-epoll.c b/sysdeps/unix/sysv/linux/tst-epoll.c
> index 66f091c202..e2fd34e0e6 100644
> --- a/sysdeps/unix/sysv/linux/tst-epoll.c
> +++ b/sysdeps/unix/sysv/linux/tst-epoll.c
> @@ -18,12 +18,13 @@
> 
>  #include <errno.h>
>  #include <intprops.h>
> +#include <libc-diag.h>
> +#include <stdlib.h>
>  #include <support/check.h>
>  #include <support/support.h>
>  #include <support/xsignal.h>
> -#include <support/xunistd.h>
>  #include <support/xtime.h>
> -#include <stdlib.h>
> +#include <support/xunistd.h>
>  #include <sys/epoll.h>
> 
>  /* The test focus on checking if the timeout argument is correctly handled
> @@ -191,7 +192,12 @@ do_test (void)
>      xsigaction (SIGCHLD, &sa, NULL);
>    }
> 
> +  /* The NULL tests here is only to check if epoll_pwait2 is supported by the
> +     kernel and to simplify the rest of test.  */
> +  DIAG_PUSH_NEEDS_COMMENT;
> +  DIAG_IGNORE_NEEDS_COMMENT (7, "-Wnonnull");
>    int r = epoll_pwait2 (-1, NULL, 0, NULL, NULL);
> +  DIAG_POP_NEEDS_COMMENT;
>    TEST_COMPARE (r, -1);
>    bool pwait2_supported = errno != ENOSYS;
> 
> Could you send a v3 with the change?  Another possibility is to remove the
> pwait2_supported and handle it on the test itself (it would require more
> extensive changes).

I suggest a slightly different approach: passing the address of a dummy variable.
It is even simpler.  See the suggested diff below.  If you like it, I'll send a
revision with it (which IIRC, should be v2).

Cheers,
Alex

$ git diff
diff --git a/sysdeps/unix/sysv/linux/tst-epoll.c b/sysdeps/unix/sysv/linux/tst-epoll.c
index 66f091c202..abda45c427 100644
--- a/sysdeps/unix/sysv/linux/tst-epoll.c
+++ b/sysdeps/unix/sysv/linux/tst-epoll.c
@@ -180,6 +180,8 @@ epoll_pwait2_check (int epfd, struct epoll_event *ev, int maxev, int tmo,
 static int
 do_test (void)
 {
+  struct epoll_event ev;
+
   {
     struct sigaction sa;
     sa.sa_handler = handler;
@@ -191,7 +193,7 @@ do_test (void)
     xsigaction (SIGCHLD, &sa, NULL);
   }
 
-  int r = epoll_pwait2 (-1, NULL, 0, NULL, NULL);
+  int r = epoll_pwait2 (-1, &ev, 0, NULL, NULL);
   TEST_COMPARE (r, -1);
   bool pwait2_supported = errno != ENOSYS;
 

-- 
<http://www.alejandro-colomar.es/>
GPG key fingerprint: A9348594CE31283A826FBDD8D57633D441E25BB5

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]