I can't verify directly, but my understanding is that the controller overwrites the entire sector data area on every write - it has no choice.
Consider that the sector headers are written once at format, but the sector data is led-in, resynchronised and written out in a single pass at each write. This comprises gap2, the DAM or DDAM, sector data, CRC and the first byte of gap3, and the previous contents of that part of the track are lost.
In the original IBM implementation, deleted data sectors kept to a special format starting with a control field, possibly additional to the 2^n bytes of content, and a vestige of this can be seen in Figure 20, but according to the microcode disassembly linked from the decap thread the 8271 writes deleted sectors on disc identically to normal ones, albeit with the variant address mark and without reading out the contents unless requested. (The "write data and deleted data" heading was most likely a typo, also corroborated by the disassembly.)
I do know that in past experiments with the 2791 I was able to format a track with only sector headers present, and write out the data fields in a second pass without any complaint from the controller.
Consider that the sector headers are written once at format, but the sector data is led-in, resynchronised and written out in a single pass at each write. This comprises gap2, the DAM or DDAM, sector data, CRC and the first byte of gap3, and the previous contents of that part of the track are lost.
In the original IBM implementation, deleted data sectors kept to a special format starting with a control field, possibly additional to the 2^n bytes of content, and a vestige of this can be seen in Figure 20, but according to the microcode disassembly linked from the decap thread the 8271 writes deleted sectors on disc identically to normal ones, albeit with the variant address mark and without reading out the contents unless requested. (The "write data and deleted data" heading was most likely a typo, also corroborated by the disassembly.)
I do know that in past experiments with the 2791 I was able to format a track with only sector headers present, and write out the data fields in a second pass without any complaint from the controller.
Statistics: Posted by regregex — Thu May 08, 2025 6:19 pm