Subject: Re: On-disk time_t
To: None <>
From: Bill Studenmund <>
List: tech-kern
Date: 12/30/1999 15:23:15
On Thu, 30 Dec 1999, Eduardo E. Horvath wrote:

> > On Thu, Dec 30, 1999 at 08:14:15AM -0800, Eduardo E. Horvath wrote:
> > > 
> > > I think we should create a new type `dtime_t' for use in on-disk
> > > filesystem structures so we can migrate time_t to a 64-bit type
> > > eventually without impacting filesystem formats.
> > > 
> > > Thoughts?

The one which immediatly comes to mind is that this is a ufs-centric thing
to do. :-)

> > How would this help? Yes, we could manipulate life-system times, but couldn't
> > unambigously touch files in the file system...
> Once you have separated the on-disk structure from the kernel's
> representation you could either version the structures and create a
> new one with 64-bit `time_t's, or grab another unused (and currently
> all zero) field to store the high 32 bits.

Or go to a different time format in the on-disk structures. Like 44/20,
will take us to past 559000 AD while still giving us millisecond

Or see what FreeBSD did (I hope the versioned the inodes when they did it)
so we might stay fs-compatible with them. :-)

Take care,