On Fri, 4 Jul 2003, Manuel Kasper wrote:
> On Thu, 3 Jul 2003, Fred Wright wrote:
> 
> > Hmm.  I thought tar was supposed to handle hardlinks correctly.
> 
> Apparently not when you use the -O option to extract individual files...

Ah, OK.  Probably it handles them in some asymmetric way, where one name
gets the content and the rest just point to it, leading to a two-pass
extraction algorithm that doesn't get along well with pipes.  The
difficulty could be avoided by associating the content with the *last*
name, but that would require a two-pass (or at least 1.5-pass) creation
algorithm.

> > You don't say what the symptom is.  Complete zap of the date/time?
> 
> Yes, sorry - it was thoroughly discussed on the Soekris list some time
> ago, so I presumed everybody knew that. ;)

The last time I looked at the Soekris list archives (no longer possible
without subscribing), the traffic looked like it was higher than I wanted
to deal with.  Perhaps if it were split into a discussion of issues
directly involving the Soekris and its interactions with software and
general issues with sofware that happens to be running on the Soekris...

> Yep, I think that's exactly what happens. AFAIK the date/time writeback
> stuff was modified quite a bit for 5.x; those changes haven't been MFC'd,
> though. I guess we'll just live with this little fix - it's not the only
> kernel patch we use in m0n0wall anyway. ;)

Personally, I don't think it's necessary to *use* the century, as long as
the ambiguity is resolved with a "floating-window" approach (GPS already
does this over an interval of less than 20 years).  But the century should
still be *set* properly, just to maintain compatibility with software that
*does* use it.  I don't know if there's any need to make this conditional
for compatibility with old BIOSes.

In any case, it should be pointed out to everyone that resetting the date
in the Soekris BIOS dialog is the workaround for the "1903 problem".  I
tested this by manually screwing up the century byte, which had to be done
with direct CMOS commands since it won't accept 1903 as a valid year.

					Fred Wright