Re: [PATCH] posixtm: don't reject a time with "60" as the number of seconds

[Jim Meyering]

Eric Blake wrote:

> Jim Meyering <jim <at> meyering.net> writes:
>> FYI, here's the new test, in case anyone feels like reviewing:
>>
>> +static struct posixtm_test T[] =
>> +  {
>> +    { "12131415.16",     13, "  1039788916 Fri Dec 13 14:15:16 2002" },
>> +    { "12131415.16",     13, "  1039788916 Fri Dec 13 14:15:16 2002" },
>
> I like how you made the test independent of the year, later on.

I've just noticed that those two lines are identical.
Humph.
I suppose one should have lacked the .SS suffix,
for coverage of that case.

>> +    { "000001010000.00", 13, "-62167132800 Sun Jan  1 00:00:00 0000" },
>> +    { "190112132045.52", 13, " -2147483648 Fri Dec 13 20:45:52 1901" },
>> +    { "190112132045.53", 13, " -2147483647 Fri Dec 13 20:45:53 1901" },
>> +    { "190112132046.52", 13, " -2147483588 Fri Dec 13 20:46:52 1901" },
>> +    { "190112132145.52", 13, " -2147480048 Fri Dec 13 21:45:52 1901" },
>> +    { "190112142045.52", 13, " -2147397248 Sat Dec 14 20:45:52 1901" },
>> +    { "190201132045.52", 13, " -2144805248 Mon Jan 13 20:45:52 1902" },
>
> I would feel more comfortable if we split this into two arrays, one supported
> by 32-bit signed time_t run on all hosts, and another for 39+-bit time_t run
> only on supporting hosts.  That way, even 32-bit time_t gets some test
> exposure, for the values which are in range.

Ok.  I'll do that.

>> +  /* This test data also assumes that time_t is signed and is at least
>> +     39 bits wide, so that it can represent all years from 0000 through
>> +     9999.  A host with 32-bit signed time_t can represent only time
>> +     stamps in the range 1901-12-13 20:45:52 through 2038-01-18
>> +     03:14:07 UTC, assuming POSIX time_t with no leap seconds, so test
>> +     cases outside this range will not work on such a host.  */
>> +  if ( ! TYPE_SIGNED (time_t))
>> +    {
>> +      fprintf (stderr, "%s: this test requires signed time_t\n");
>> +      return 77;
>> +    }
>
> Is this true?  It seems like unsigned 64-bit time_t will still pass this test.

I don't know off hand.  I trusted the comment ;-)

>> +  if (sizeof (time_t) * CHAR_BIT < 39)
>> +    {
>> +      fprintf (stderr, "%s: this test requires time_t at least 39 bits
> wide\n");
>> +      return 77;
>> +    }
>
> Given my above idea for splitting into two arrays, it would be nice if this
> message occurred after testing the first array, and included wording such
> as "skipping remainder of test".

Sure.