>  o Now the bug manifests itself.  NOTE 32 in my draft of the scsi spec
>    says: "Since the data pointer value may be modified by the target
>    before the I/O-process ends, it should not be used to test for
>    actual transfer length because it is not reliable."  Contrary to

Cf. section 6.4.2:

"  On those occasions when an error or exception condition occurs and the 
target elects to repeat the information transfer, the target may repeat the 
transfer by either issuing a RESTORE POINTERS message or by disconnecting 
without issuing a SAVE DATA POINTER message.  When reconnection is completed, 
the most recent saved pointer values are restored."


Granted, this does not imply anything for the "non-error/exception" case..

-pk