Date: Sat, 6 Jun 1998 20:53:20 -0700 (PDT)
From: Dean Gaudet <dgaudet@arctic.org>
To: Apache <new-httpd@apache.org>
Subject: Re: time_t is 64 bits in linux/alpha
In-Reply-To: <Pine.LNX.3.96dg4.980606203616.26186G-100000@twinlark.arctic.org>
Message-ID: <Pine.LNX.3.96dg4.980606204247.26186H-100000@twinlark.arctic.org>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=ISO-8859-1
Content-Transfer-Encoding: QUOTED-PRINTABLE
Sender: new-httpd-owner@apache.org
Precedence: bulk
Reply-To: new-httpd@apache.org

Oh yeah, I wanted to rant about big-endian order (i.e. network order).=20

If you study the theory behind mod_unique_id's tokens you'll see that with
a "flag second" (second, or longer interval, in which no requests are
served and all software is simultaneously upgraded) you can change from
one token format to another... provided you keep the time_t at the front.=
=20
If the time_t were in little-endian format, this would work correctly
regardless of the time_t size on any of the hosts involved (assuming it's
at least 32-bits long).=20

But no... it's big-endian order...  To perform an upgrade from 32-bit time
stamps to 64-bit timestamps, and maintain the property that tokens are
unique for all time, we have to do something like this:=20

    __u32 time_low;
    __u32 time_high;

Each half is in BE order (network order).  But the 64-bit aggregate is in
a mixed little/big format.  What a waste.=20

If it was in little endian format to begin with we could just start using
"__u64 time;" and it would require no special coding.=20

Dean

On Sat, 6 Jun 1998, Dean Gaudet wrote:

> Yeah I was thinking about how to deal with this cleanly... it's too bad C
> doesn't define the result of "u >> 32" where u is 32 bits... because we
> could code it to expect a 64-bit time_t and break it into two parts, and
> for 32-bit time_t it would result in a bunch of dead code (eliminated at
> compile time).=20
>=20
> NSPR defines a 64-bit PRTime ... so it'll be easy to clean up in the nspr
> version.
>=20
> Dean
>=20
> On Sun, 7 Jun 1998, Alvaro Martinez Echevarria wrote:
>=20
> > On Fri, 5 Jun 1998, Dean Gaudet wrote:
> >=20
> > > Is time_t 64-bits on alpha-linux ?=20
> > > Dean
> >=20
> > Yes, as defined by /usr/include/asm/posix_types.h:
> >=20
> > typedef long            __kernel_time_t;
> >=20
> > and /usr/include/linux/types.h:
> >=20
> > typedef __kernel_time_t              time_t;
> >=20
> > So it's 64 bits. Right now I'm working on a patch for
> > mod_unique_id to fix this problem, and to make it more simple I
> > have changed the type of stamp from "time_t" to "unsigned int".
> > This way htonl and ntohl work properly (although we'll have to
> > change this back to 64bits before year 2106).
> > Probably I'll submit the patch in a few minutes.
> > Regards.
> >=20
> > .------------------------------------------------------------------.
> > |   Alvaro Mart=EDnez Echevarr=EDa   |      LANDER SISTEMAS            =
|
> > |        alvaro@lander.es        |      P=BA Castellana, 121         |
> > `--------------------------------|      28046 Madrid, SPAIN        |
> >                                  |      Tel: +34-91-5562883        |
> >                                  |      Fax: +34-91-5563001        |
> >                                  `---------------------------------'
> >=20
> >=20
>=20
>=20