On Fri, Jun 07, 2013 at 09:37:45PM +0000, Thorsten Glaser wrote:
> Kurt Roeckx dixit:
>
> >tm_year should be an int, not a time_t or long. Note that it
>
> POSIX says it "must" be a long...
It doesn't say so here. It has it as an int.
Also note that time_t didn't have a requirement to be an integer
type, but as far as I know it does now. And you really want an
integer type for the values in struct tm.
> >counts in years, and so even a 16 bit int isn't going to cause
> >much problems for tm_year.
>
> ... but I disagree. All time_t seconds-since-the-epoch values
> need (in practice) to round-trip through "struct tm", which
> requires tm_year to be of a type of the magnitude time_t has:
If you add that requirement, it can be upto 24 bit smaller than
time_t. But as far as I know, there is no such requirement. In
fact POSIX defines an error code for that case.
> tg@blau:~ $ cat x.c
> #include <limits.h>
> #include <stdio.h>
> #include <time.h>
> int main(void) { struct tm *tm; time_t tmax = LLONG_MAX; tm = gmtime(&tmax);
> printf("%lld\n", (long long)tm->tm_year); return (0); }
> tg@blau:~ $ ./a.out
> 292277024696
tm should be NULL in this case, an errno EOVERFLOW. If that's not
the case I suggest you file a bug against libc. It's at least
what I get here.
Kurt
--
To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Archive: http://lists.debian.org/20130607220656.ga12...@roeckx.be
