Re: [PATCH v2 2/2] linux: always update select timeout (BZ #27706)

[Florian Weimer <fweimer@redhat.com>]

* Adhemerval Zanella via Libc-alpha:

> diff --git a/misc/tst-select.c b/misc/tst-select.c
> index dc7717a7a6..9c817f4035 100644
> --- a/misc/tst-select.c
> +++ b/misc/tst-select.c
> @@ -23,6 +23,7 @@
>  #include <support/timespec.h>
>  #include <support/xunistd.h>
>  #include <support/xtime.h>
> +#include <support/xsignal.h>
>  
>  struct child_args
>  {
> @@ -30,6 +31,12 @@ struct child_args
>    struct timeval tmo;
>  };
>  
> +static void
> +alarm_handler (int signum)
> +{
> +  /* Do nothing.  */
> +}
> +
>  static void
>  do_test_child (void *clousure)
>  {
> @@ -59,6 +66,22 @@ do_test_child (void *clousure)
>    xwrite (args->fds[1][1], "foo", 3);
>  }
>  
> +static void
> +do_test_child_alarm (void *clousure)
> +{
> +  struct sigaction act = { .sa_handler = alarm_handler };
> +  xsigaction (SIGALRM, &act, NULL);
> +  alarm (1);
> +
> +  struct timeval tv = { .tv_sec = 10, .tv_usec = 0 };
> +  int r = select (0, NULL, NULL, NULL, &tv);
> +  TEST_COMPARE (r, -1);
> +  TEST_COMPARE (errno, EINTR);
> +
> +  if (support_select_modify_timeout ())
> +    TEST_VERIFY (tv.tv_sec < 10);
> +}
> +
>  static int
>  do_test (void)
>  {
> @@ -98,6 +121,13 @@ do_test (void)
>    xclose (args.fds[0][0]);
>    xclose (args.fds[1][1]);
>  
> +  {
> +    struct support_capture_subprocess result;
> +    result = support_capture_subprocess (do_test_child_alarm, NULL);
> +    support_capture_subprocess_check (&result, "tst-select-child", 0,
> +				      sc_allow_none);
> +  }
> +
>    {
>      fd_set rfds;
>      FD_ZERO (&rfds);

I thought a bit about how this would interact with SIGALRM in the test
suite, but this should be okay: the SIGALRM handler is in the parent
wrapper process, not in the child process, and in --direct mode (without
a wrapper), there is no timeout at all.  So the test should be okay.

The actual code change looks okay to me too.

Thanks,
Florian