On 6/10/15 11:25 AM, Dennis Boone wrote:
The discussion of .tap format to which I referred earlier largely seemed to revolve around the question of representing errors. Block length markers occur at the beginning and end of data records, unless they're file marks (length 0). On a read error, should one write a single error mark, a pair of them, should one include the busted data, etc. The answers are complicated by the SCSI habit of hoarding the damaged data.
The current spec is 'undefined'. Bob extended it in Simh but behavior on read errors was never specified. John Bordynuik extended it in other ways in the tapes read by JBI, Inc., but I never found out what those extensions were. As was discussed on the Simh list, there is a bunch of information, like drive type and how it handles bad blocks, that should be in the tape image, but currently is not. I would think an error that returns no data would be a single error mark, but one with returned data should be two to handle the read reverse case.