Improve UDI disk image support

[morozov]

Open .udi from Finder

master doesn't register .udi as a document type, so it can't be opened from Finder. This PR adds the registration and an icon, mirroring fdi (#43); no controller change is needed, since UDI already routes through the shared DISK_GENERIC open path.

Description block rejected on load

A UDI may set header byte 11 bit 0 to flag a NUL-terminated description stored after its last track¹. master's open_udi scans tracks to EOF, so it reads the first description byte as a track-type byte, fails the check, and returns DISK_UNSUP:

fusex/peripherals/disk/disk.c

Line 1055 in a0fb141

for( i = 0; buffer->index < eof; i++ ) {

Both scan loops are bounded to sides × cylinders tracks, leaving the description unparsed. Spectrofon #09 (SP9_O.udi, header byte 11 = 0x01) is rejected on master and loads with this change.

More than 85 cylinders rejected

The shared GEOM_CHECK macro rejects any image over 85 cylinders as DISK_GEOM:

fusex/peripherals/disk/disk.c

Lines 831 to 833 in a0fb141

	#define GEOM_CHECK \
	if( d->sides < 1 || d->sides > 2 || \
	d->cylinders < 1 || d->cylinders > 85 ) return d->status = DISK_GEOM

The UDI max-cylinder field is one byte (0x00–0xFF, up to 256 cylinders)¹, and the drive reaches FDD_MAX_TRACK (99), so 85 was below both limits. The cap is raised to 86. GEOM_CHECK is shared by every disk loader and track storage is allocated from d->cylinders, so the wider cap is safe across formats.

Spectrum Expert (SP_EXP1O.UDI) is an 86-cylinder example: its CRC is valid and it has no description block, so master rejects it only at GEOM_CHECK, and raising the cap admits it.

Sectors with imperfect MFM sync don't read

read_id and read_datamark require textbook MFM sync: three 0xffa1 marks before the address mark, and in read_datamark a 0x00 data sync and a gap of only 0x4e/0x00. Any stray gap byte errors the read:

fusex/peripherals/disk/wd_fdc.c

Lines 297 to 305 in a0fb141

	for( i = 40; i > 0; i-- ) {
	fdd_read_data( d );
	if( d->data == 0x4e ) /* read next */
	continue;
	if( d->data == 0x00 ) /* go to PLL sync */
	break;
	return 1; /* something wrong... */

Disks imaged from protected or marginal originals violate this — stray gap bytes, or one or two 0xffa1 marks with no 0x00 sync. A real WD1793 reads them; master does not.

The fix scans the gap for the A1 run within the datasheet's data-mark window — 43 byte-times in MFM, 30 in FM² — then locks onto the A1 and the following mark byte. It presets the address-mark CRC to 0xcdb4, the value after three A1 bytes, and resumes from the mark byte. The surviving A1 count therefore never affects validation: a standard three-A1 disk validates identically, and the datasheet's detector keys on a single missing-clock A1 regardless².

This lives in shared WD1793 core, so it also covers +D; the FM path is untouched.

Unsupported CRC algorithms

A UDI whose stored CRC32 was computed with a standard algorithm instead of the format's signed-accumulator variant¹ is rejected on load; this PR doesn't relax that. Возрождение (VOZRAZH1.udi) is such an image — its stored CRC matches the textbook algorithm, not the format's.

Footnotes

1. UDI format specification — Alex Makeev, 2002 (Sinclair Wiki transcription). ↩ ↩² ↩³

2. WD179X / FD179X datasheet — 1986 Western Digital Storage Management Products Handbook. ↩ ↩²