Thomas Harte
83c7d34df2
Switched to populating the sector cache with everything in a track the first time anything on that track is requested. That avoids the problem whereby each request of a non-existent sector costs two spins.
2017-08-11 18:40:16 -04:00
Thomas Harte
a12671010a
Sector size is now reported, and CRC failures are merely indicated, not cause for a sector to be thrown away.
2017-08-11 16:23:33 -04:00
Thomas Harte
edb088526f
Simplified slightly, and updated TODO as to still-missing functionality.
2017-08-11 14:33:34 -04:00
Thomas Harte
80ebc63101
Updated the SSD file format container to specify sector sizes, now that it's no longer implicit.
2017-08-11 14:30:35 -04:00
Thomas Harte
cf1403bc79
Increased documentation.
2017-08-11 14:27:07 -04:00
Thomas Harte
fcf63a7547
Expands the [M]FM encoder to respect some new Sector flags: it will now wilfully make CRC errors, omit data, include data that is different than the ID's declared length, write deleted data, and can be commanded as to header/data gaps and what should be within them. All based around expanding towards the needs for reproduction of the CPC's .DSK file format.
2017-08-11 14:24:50 -04:00
Thomas Harte
0411b51582
Added an attempt to deal with 16-bit allocation units, and to ensure middle-of-file holes are respected.
2017-08-11 10:59:37 -04:00
Thomas Harte
026101a268
Killed logic_extents_per_physical, since I don't know how to handle it, and instituted tracks, to allow a decision about short versus long allocation units.
2017-08-11 10:46:50 -04:00
Thomas Harte
734099a956
Threw a sector cache into my MFM parser, in an attempt to cut down analysis costs. Also made it aware of multiple heads.
2017-08-11 10:29:13 -04:00
Thomas Harte
6be5851484
Cleaned up.
2017-08-10 22:34:29 -04:00
Thomas Harte
994179f188
Taking a whole bunch of guesses, this might be correct.
2017-08-10 22:33:08 -04:00
Thomas Harte
6a65c7a52a
Started working on a CPC-oriented analyser; for now I just want to be able to make a good guess at the appropriate file to load from a disk. As it turns out, the CPC simply adopts the CP/M format, so a generic parser is appropriate. This is its beginning.
2017-08-10 17:10:21 -04:00
Thomas Harte
7d1023ea98
Added a 'ready' getter to Drive, formally to let the drive take ownership of that test.
2017-08-08 21:15:56 -04:00
Thomas Harte
2b0dcf8573
Transcribed the status bits that I think actually need to be obeyed.
2017-08-07 12:37:45 -04:00
Thomas Harte
47732ffb98
Prevented the 8272 from overreading ID fields (and, by doing so, overrunning its internal buffer). Exposed the MFMController's CRC generator for inspection.
2017-08-07 12:37:22 -04:00
Thomas Harte
56d65ba6f3
Adapted slightly, to retain the ability to advertise an incorrect size, to adjust the confusion I've created by having two different types of thing both called new_sector, and to print a warning when ignoring error flags.
2017-08-07 12:12:04 -04:00
Thomas Harte
895a3cbf24
Corrected reading of the track size table for extended disks. My first extended disk has now loaded.
2017-08-07 11:38:19 -04:00
Thomas Harte
d951c8c1c2
Fixed search for track start position with extended disks: it's no longer an infinite loop. So that's a pretty good performance improvement.
2017-08-07 11:36:29 -04:00
Thomas Harte
a294963e98
Made an absolutely basic attempt to accommodate some extended disk images.
2017-08-07 11:26:15 -04:00
Thomas Harte
c839556a27
Fixed: rewind the file to check for 'EXTENDED' if 'MV - CPC' failed.
2017-08-06 19:47:10 -04:00
Thomas Harte
77da582e88
Switched the container in which events are passed to int, with the intention of subclasses extending the receivable range.
2017-08-06 12:35:20 -04:00
Thomas Harte
ffadb04761
Documented, and removed a couple of Event types that are WD-specific but had accidentally flown into here. Will need to figure out how best to expose the CRC result too, but I'm willing to let that one drop out naturally as I implement the 8272.
2017-08-06 11:36:36 -04:00
Thomas Harte
29288b690e
Switched disk controllers to be instantiated explicitly in terms of cycles, created an Amstrad-specific subclass of the 8272 to record the direct programmatic availability of all disk motors bundled together, and otherwise adjusted to ensure the thing is clocked and that the motor is enabled and disabled appropriately. The 8272 is also now formally a subclass of the incoming MDM controller.
2017-08-06 09:45:16 -04:00
Thomas Harte
4d60b8801c
Started trying to factor out just the PLL stream -> FM/MFM events part that is presently in the WD1770.
2017-08-05 22:26:15 -04:00
Thomas Harte
192f232d3f
Silenced warnings.
2017-08-05 11:53:29 -04:00
Thomas Harte
6e4d3b8a77
Added enough logic to produce some sort of version of a completely unprotected DSK. So enough to start bootstrapping an FDC emulation, at least.
2017-08-05 11:44:53 -04:00
Thomas Harte
8eda24261c
Removed unnecessary header — it's implied by being a child of FileHolder.
2017-08-05 11:44:06 -04:00
Thomas Harte
75c59fefab
Added an empty husk to begin support for Amstrad CPC disk image formats.
2017-08-05 10:02:10 -04:00
Thomas Harte
9d953421d8
After a quick check, added a couple of other _delegate initialisations. I should probably find a way to template this.
2017-08-01 07:07:43 -04:00
Thomas Harte
4abd62e62b
Standardises on const [Half]Cycles as the thing called and returned, rather than const [Half]Cycles & as it's explicitly defined to be only one int in size, so using a reference is overly weighty.
2017-07-27 22:05:29 -04:00
Thomas Harte
8361756dc4
Switched definitively to the works-for-now approach of requiring an explicit opt-in where somebody wants to clock a whole-cycle receiver from a half-cycle clock.
2017-07-27 07:40:02 -04:00
Thomas Harte
75d67ee770
Relocated ClockReceiver.hpp as it's a dependency for parts of the static analyser, and therefore needs to be distinct from the actual emulation parts.
2017-07-25 20:20:55 -04:00
Thomas Harte
a1e9a54765
Eliminated redundant uses of ClockReceiver and sought to ensure that proper run_fors are inherited all the way down.
2017-07-25 20:09:13 -04:00
Thomas Harte
8a2bdb8d22
Converted the TimedEventLoop and the things that sit atop it into ClockReceivers.
2017-07-24 21:19:05 -04:00
Thomas Harte
b3ae920746
Converted the DPLL and disk controller classes to be ClockReceivers.
2017-07-24 21:04:47 -04:00
Thomas Harte
45ec5f8eab
Eliminated implicit sign conversion.
2017-07-21 21:53:27 -04:00
Thomas Harte
1d01acce06
Fixed differing types of loop variables and targets.
2017-07-21 21:53:05 -04:00
Thomas Harte
163c0f1b44
Ensured offset means exactly one thing.
2017-07-21 20:58:17 -04:00
Thomas Harte
e5188a60dc
Settled on the new average-of-length approach to a PLL window sizing, eliminating the old errors-of-phase approach. Since it anchors automatically to the original target clocks per bit, killed the explicit mention of a tolerance.
2017-07-16 19:03:50 -04:00
Thomas Harte
e71d13c090
With the new PLL implementation, switching to a deeper window size returns the Acorn tape parser to: working.
2017-07-16 17:12:12 -04:00
Thomas Harte
51177e4e1f
Attempted a different implementation of the PLL, that responds to changes only once.
2017-07-16 16:49:04 -04:00
Thomas Harte
4489f120f9
Eliminated foolish double indirection on phase history.
2017-07-15 22:40:38 -04:00
Thomas Harte
e01f3f06c8
Completed curly bracket movement.
2017-03-26 14:34:47 -04:00
Thomas Harte
f94f34f053
Made an attempt at read track. Which means process_input_bit can't just swallow syncs any more; it now reports them as tokens of type ::Sync.
2017-01-01 20:39:19 -05:00
Thomas Harte
b3c33d993a
Made an attempt to explain the requirements placed upon Disk subclasses that wish to support writing.
2016-12-31 15:30:48 -05:00
Thomas Harte
8eb21c6702
The "MFM...Byte"s aren't MFM-specific, they're relevant to both FM and MFM encoding. So renamed them. Also slimmed syntax within MFM.cpp mostly where emigration from the Acorn disk analyser had left a residue of lengthy namespace specification.
2016-12-31 15:25:11 -05:00
Thomas Harte
52b850a3f5
Quick extra: make sure parsed tracks don't overflow the 6400 byte space available in an MFM disk. Which might be better expressed as 6250?
2016-12-31 12:51:52 -05:00
Thomas Harte
cfbab1448c
Switched to a track parsing that disallows synchronisation values within sector contents.
2016-12-31 12:23:08 -05:00
Thomas Harte
12549ff412
Might as well get the file offset before entering the critical section; also moved the lock guard down more explicitly to group with the second set of actions.
2016-12-31 11:48:46 -05:00
Thomas Harte
6f0b5427e4
Made an attempt to avoid repetition of sync bytes.
2016-12-31 00:20:00 -05:00
Thomas Harte
0123b37213
Made an attempt to include sync values in the stream and properly to align.
2016-12-31 00:11:31 -05:00
Thomas Harte
ea4d85e1cd
The virtual disk constructed is the same across all tracks. So why not just request zero?
2016-12-31 00:10:35 -05:00
Thomas Harte
f217d508b8
Completed first attempt at write support for Oric disk images.
2016-12-30 23:12:46 -05:00
Thomas Harte
1f625fad66
Decided that if this is an [M]FM parsing function then it should be something more intelligent than a mere PLL record. Which I guess conveniently implies Oric DSK-esque behaviour. But properly defined, rather than very vaguely.
2016-12-30 23:10:52 -05:00
Thomas Harte
632b3c63b1
Added the infrastructure necessary for Oric disks to appear writeable to the machine and to receive changed tracks.
2016-12-30 22:51:48 -05:00
Thomas Harte
d581294479
Added get_track to get the PLL output for a complete track.
2016-12-30 19:59:23 -05:00
Thomas Harte
3b29e6a473
Ensured SSD and ADFs are grown if required.
2016-12-30 18:08:12 -05:00
Thomas Harte
07dacff42d
Added writing for Acorn ADF disks, plus appropriate TODOs in both similar bits of boilerplate.
2016-12-30 18:03:30 -05:00
Thomas Harte
c85450648f
Fix: make sure copies have proper event lengths. Also made it much clearer what's going on with the initial copy to the heap.
2016-12-30 17:55:46 -05:00
Thomas Harte
c740d9655a
Fixed: index_count_ may have been left high by a previous call; reset it just in case.
2016-12-30 17:55:06 -05:00
Thomas Harte
d09e7ac1e8
Made an attempt at reacting appropriately if the very first thing that looks like a sector doesn't pan out.
2016-12-30 17:44:35 -05:00
Thomas Harte
5d63556870
Periods need a custom copy constructor too, if they're going to avoid sharing an event_source.
2016-12-30 17:39:52 -05:00
Thomas Harte
81a3cbac45
Ensured a copy is passed for writing back rather than the original.
2016-12-30 17:26:44 -05:00
Thomas Harte
63ff5165a4
After a quick bit of reading, discovered the virtual copy constructor pattern really is only a convention in C++, and conformed to it. Which hopefully gives copyable tracks.
2016-12-30 17:25:39 -05:00
Thomas Harte
71dbd78cf2
If asynchronous background processing is to occur on tracks then, given that they inherently have state, they'll need to be copyable, and ideally 'cheaply' (though it's not too great a priority). So started implementing appropriate copy constructors. Also introduced an extra level of indirection to PCMSegmentEventSource so that it can copy itself without copying the underlying PCMSegment, which is 95% of the heft of a track in all currently-implemented cases.
2016-12-30 14:23:26 -05:00
Thomas Harte
f88f3c65e9
Removed duplicated newline.
2016-12-30 14:21:36 -05:00
Thomas Harte
82bb78fb2d
Ensured that get_sector copes even if any invalid sectors are encountered.
2016-12-30 14:21:14 -05:00
Thomas Harte
6fc692cd34
Attempted to switch to an asynchronous means for continuous file updates. Testing with SSD, as usual.
2016-12-29 22:15:58 -05:00
Thomas Harte
b9fad184d7
Added just enough for a complete manual test of writing to a .ADF with the 1770 then getting the correct result parsing it back on the host side in order potentially to update a file.
...
... which means that now it's time to worry about when and how mounted files should actually update themselves. Which will make for some fun with threading, I dare say.
2016-12-28 23:00:47 -05:00
Thomas Harte
af1b396c9e
Found an ugly issue with Storage::Time as implemented (i) to be unsigned; and (ii) automatically to simplify. Will need to fix. Here's a quick workaround for this one segment of code.
2016-12-28 22:57:11 -05:00
Thomas Harte
9cb902cc4f
Experimentally marked ADF as writable too, immediately discovering a mistake in the analysing MFM decoder.
2016-12-28 22:34:22 -05:00
Thomas Harte
bfe6c0a0c1
Ensured that FileHolder gets a writeable file reference if one is possible, and records whether the file in hand is read-only. So now the SSD class can answer honestly.
2016-12-28 20:09:14 -05:00
Thomas Harte
46a93d2e12
Fixed errors to ensure that FM disks, at least, follow the same CRC generation rules when being built from sectors and when being parsed.
2016-12-28 19:48:46 -05:00
Thomas Harte
720b1e5802
Attempted to ensure proper CRC generation for FM-format input.
2016-12-28 18:56:53 -05:00
Thomas Harte
3a9ad3fb08
Fixed Oric .DSK handling, per my latest understanding. Which creates a desire to write shorts directly to the disk surface, so exposed that in the encoder.
2016-12-28 18:50:28 -05:00
Thomas Harte
a568172758
Made steps towards proper CRC generation. Am currently comparing against Oric disk images, as — amongst other things — they include precomputed CRCs.
2016-12-28 18:29:37 -05:00
Thomas Harte
0490a47058
Worked on the all-around framework for decoding sectors back from tracks when closing down a file. Hit the wall that the parser is more observant of CRCs than the WD. No, really. So I guess I have to stop avoiding that whole issue.
2016-12-26 14:24:33 -05:00
Thomas Harte
742c5df367
With lots of logging arising temporarily, fixed bug whereby conversion to a patched track would lead to holding a track with a distinct measure of time, leading to improperly-placed patches.
2016-12-25 22:00:39 -05:00
Thomas Harte
b538ee5bd8
Fixed discovery of correct active period and setting of track time, when seeking.
2016-12-25 21:32:50 -05:00
Thomas Harte
a6d038cad9
Eliminated special case that doesn't seek properly and isn't needed. Added TODO.
2016-12-25 21:32:14 -05:00
Thomas Harte
4fca30b81f
Made the Plus 3 less chatty, documented invalidate_track.
2016-12-25 21:06:58 -05:00
Thomas Harte
26710c988d
Modified SSD to ensure a fully-formatted surface is represented even if no track data is in the source file. This corrects the controller's sense of write success.
2016-12-25 20:40:06 -05:00
Thomas Harte
acc35885cd
Attempted to reduce track invalidations.
2016-12-25 20:38:25 -05:00
Thomas Harte
beaa868079
Factored the MFM parser out into encodings.
2016-12-25 20:00:57 -05:00
Thomas Harte
1349e85d83
[Mostly] fixed track write-back.
2016-12-25 19:19:22 -05:00
Thomas Harte
007c13ec16
Fixed: cycles_per_bit_ isn't a function of the rotational multiplier, it's absolute. Also made sure that exactly hitting the end of a bit counts.
2016-12-25 16:35:39 -05:00
Thomas Harte
98be6ede45
Shuffled a little to reduce risk of overflow, ensured writing is a loop, still seem to be writing too quickly for some reason.
2016-12-25 16:13:05 -05:00
Thomas Harte
d2ad2c756e
Added enough shovelling to write rubbish for an entire sector.
2016-12-25 15:46:49 -05:00
Thomas Harte
ec55a25620
It makes sense to simplify these ahead of time.
2016-12-25 12:32:25 -05:00
Thomas Harte
aceb7e3b6b
Started implementing write sector on the 1770, immediately deciding it would be useful to have a callback for end-of-queued-data-written from disk controller. So had a go at implementing that, naively. More investigation required.
2016-12-25 12:31:38 -05:00
Thomas Harte
901f19f89c
Added enough stuff that SSDs attached to a 1770 will now reach the entry point for writing.
2016-12-25 09:46:12 -05:00
Thomas Harte
9d555c4a02
Let's try just declining to pump the PLL while in write mode. Added documentation to explain.
2016-12-25 09:19:18 -05:00
Thomas Harte
b57038edc5
Actually, at least index holes will still be receivable while writing, so this wasn't entirely correct. Probably best to leave it in.
2016-12-25 09:16:09 -05:00
Thomas Harte
af69b21033
This is almost complete, except that it doesn't act appropriately if some bits are written but not enough to cover the entire writing period.
2016-12-24 22:51:26 -05:00
Thomas Harte
f601d796f5
Added documentation.
2016-12-24 22:37:20 -05:00
Thomas Harte
6e94d0c19f
Extended Storage::Disk::Disk to permit write-back of modified tracks, exposed some interface via Storage::Disk::Drive.
2016-12-24 22:11:31 -05:00
Thomas Harte
7f303cfceb
Continued the baby steps.
2016-12-24 21:54:43 -05:00
Thomas Harte
afc6f4129c
Withdrew unused tally.
2016-12-24 21:47:57 -05:00
Thomas Harte
1e416d4af0
Withdrew now-unused and never-implemented API from TimedEventLoop, and the redundant track time count from DiskController.
2016-12-24 21:02:10 -05:00
Thomas Harte
bedea48d03
This is a much better way of dealing with being partway into an incoming event. Subject to eliminating overruns, of course.
2016-12-24 20:54:27 -05:00
