23 May
2014
23 May
'14
12:23 p.m.
On Fri, May 23, 2014 at 12:28 PM, Paul Koning <paulkoning at comcast.net>wrote:
Was it you, or Paul Pierce, who proposed a format that
includes a way to
represent errors rather than only good data?
They may have. I proposed one, which went nowhere. I proposed a metadata
record of:
32-bit 0xffffffff (metadata indicator)
32-bit length (of data between the two length fields, not inclusive)
32-bit metadata type (needs to be a process for uniquely assigning
them)
metadata ((length - 4) bytes)
32-bit length (same as previous length field)
32-bit 0xffffffff
I proposed a metadata record for bad tape areas, with a flag word that
would contain flags for:
unknown record count - clear for single bad record, set for unknown
number of bad records
unknown record length
approximate record length
(For bad records of known length, neither the unknown record length nor the
approximate record length flags would be set.)
If I were writing such a proposal today, I'd replace the 32-bit metadata
type with a counted string containing a reversed-DNS name (as used for Java
package namespace), so that anyone with a registered DNS domain can assign
their own metadata record types.
Eric