Re: [PATCH v7 10/11] utimes: detect utimes() correctly on OS/2 kLIBC

[Paul Eggert]

On 01/13/2016 06:23 PM, KO Myung-Hun wrote:
> address@hidden
> +On OS/2, this funciton cannot set timestamp earlier than 1980 year in local
> +time.
> address@hidden
> +On OS/2, this function cannot set timestamp in odd seconds.

Misspelled "function", and some grammar issues. Please change to 
something like this:

On OS/2, this function cannot set the timestamp to earlier than the
year 1980 in local time.
@item
On OS/2, this function cannot set the timestamp to an odd number of
seconds.


> +  /* On OS/2, a file date should be since 1980 year in local time
> +     and even seconds.  */

Reword to "On OS/2, file timestamps must be on or after 1980 in local 
time, with an even number of seconds."

> +  static struct timeval timeval[2] = {{315620000 + 10, 10},
> +                                      {315620000 + 1000000, 999998}};
>   
>     /* Test whether utimes() essentially works.  */
>     {
> @@ -82,15 +86,18 @@ main ()
>             result |= 1;
>           else if (fstat (fd, &st0) != 0)
>             result |= 1;
> +        else if (write (fd, "\n", 1) != 1)
> +          result |= 1;
> +        else if (fstat (fd, &st1) != 0)
> +          result |= 1;
> +        /* utimes() of OS/2 kLIBC does not work on an opened file */
> +        else if (close (fd) != 0)
> +          result |= 1;
>           else if (utimes (file, timeval) != 0)
>             result |= 2;
>           else if (utimes (file, NULL) != 0)
>             result |= 8;
> -        else if (fstat (fd, &st1) != 0)
> -          result |= 1;
> -        else if (write (fd, "\n", 1) != 1)
> -          result |= 1;
> -        else if (fstat (fd, &st2) != 0)
> +        else if (lstat (file, &st2) != 0)
>             result |= 1;
>           else if (time (&t2) == (time_t) -1)
>             result |= 1;
> @@ -103,7 +110,8 @@ main ()
>               if (! (m_ok_POSIX || m_ok_NFS))
>                 result |= 32;
>             }
> -        if (close (fd) != 0)
> +        /* Ensure to close fd, but ignore an error if already closed. */
> +        if (close (fd) != 0 && errno != EBADF)
>             result |= 1;
>         }
>       if (unlink (file) != 0)

This doesn't look right, as the original order of updating the file's 
time stamps is needed to tickle the NFS bug. Instead, please change the 
code to attempt utimes() on the opened file, and if this fails the way 
that kLIBC fails, then close the file, invoke utimes on the file, and 
reopen the file. That way, the order will be preserved and there should 
be no need for any EBADF check.