Alice Averlong🏳️⚧️the best time to do win32 development is when microsoft's entire cloud has shit the bed
gotta pull out my visual studio 6 MSDN CD because the online stuff keeps timing out
and lets see if the internet archive has WinObj, because sysinternals is not responding
I'm just trying to read sectors off a floppy drive that's not mounted because it doesn't have a FAT12 filesystem!
surely this is an everyday occurrence in windows 11, why is this so hard?
ah-ha!
r"\\.\GLOBALROOT\Device\Floppy0"
I'm benchmarking how long it takes to read a floppy disk, for my video player. test one is calling read(512) 2880 times
305 seconds, that's five minutes.
My video is 252s/floppy, so that's sub-realtime. NOT GOOD
My video is 252s/floppy, so that's sub-realtime. NOT GOOD
½kb: 305s
1kb: 161s
2kb: 89s
4kb: 58s
8kb: 42s
16kb: 34s
32kb: 30s
64kb: 28s
128kb: 28s
256kb: 28s
512kb: 28s
1024kb: 28s
1440kb: 28s
1kb: 161s
2kb: 89s
4kb: 58s
8kb: 42s
16kb: 34s
32kb: 30s
64kb: 28s
128kb: 28s
256kb: 28s
512kb: 28s
1024kb: 28s
1440kb: 28s
That's speeds on a Teac FD-05PUW
a reasonable thing to do would be to adjust my video bitrate so that it stores something like 45-60s instead of 252s, since I can read one disk every 28s.
this'd get me MAXIMUM VIDEO QUALITY at the expense of increasing the video size from 12 floppies to 51-68
Greaseweazle + TEAC FD-235HF:
default settings: 97s
--revs 1: 66s
that makes sense. The USB Teac is a 2X drive, so it's running the disk at 600 RPM instead of the intended 300 RPM, and it gets roughly twice the speed.
there are 4x drives, but they're rare and I think they've mainly been used inside mavicas
I've got one but I'm not gonna build a video player that runs off a camera, that's just silly.
but it's good to know I have plenty of headroom on this 2x drive. even if the user takes a full minute to find and insert the next floppy disk, I can load it fast enough to keep the video going
next step would obviously be to measure how long it takes me to swap through 12 floppy disks
if I could instantly load the next floppy disk that'd be 336s to read all 12. obviously it's going to be more than that, but how much more?
maybe I'll run two tests: one with the disks sorted, and one with them shuffled.
I should mod a card shuffler to work on floppy disks
"The 4x USB Floppy Drive is 400% faster than a typical 1x floppy drive, making it the fastest drive available today. "
NOT HOW MATH WORKS
I just realized I'm writing a file/disk marker to each floppy that consists of "FLOP"
yes, of course they're a floppy, foone! this is not a useful distinguishing signature!
there we go, reading the metadata too:
>python readdisk.py
Reading 12 chunks of 131072 bytes each
FlopPlayer 0.1 video AMOKTIME disk 1 of 12
Exact file size is 16504149 bytes
Read 1474560 bytes in in 28.71s
I wonder if I can autodetect the next disk insertion. The drive definitely does detect new disks, it spins them up as soon as you put them in.
but does that info make it all the way to windows? I'm not sure.
and I don't want to do read(1) in a loop until it succeeds, that's nasty. and can be noisy on floppy drives
annoyingly I had mostly solved this problem in software I wrote back in 2018 but it's all python2 so adapting it to py3 would take a minute
fortunately my desktop has enough old python versions installed that I can get 2to3 to help me
okay I now have a working disk-swap-detector.
It basically just calls DeviceIoControl(IOCTL_DISK_GET_DRIVE_GEOMETRY) in a loop. That'll fail if the drive is empty.
It starts "full", waits for it to go empty, then waits for it to go back to full.
what if the user puts in the wrong disk?
PANIC
0%| | 0/1474560 [00:00<?, ?it/s]FlopPlayer 0.1 video 2j▐TDMí disk 2 of 12
Exact file size is 4039327740 bytes
Disk info mismatch! Expected title to be AMOKTIME, but was 2j▐TDMí
0%| | 0/1474560 [00:00<?, ?it/s]
Traceback (most recent call last
Exact file size is 4039327740 bytes
Disk info mismatch! Expected title to be AMOKTIME, but was 2j▐TDMí
0%| | 0/1474560 [00:00<?, ?it/s]
Traceback (most recent call last
oh I was trying to check that the title was the same on later disks, but the title is only saved on the first disk (can't waste 8 bytes)
whooops
this is arguably a mis-design of my disk format and I should fix it and make it use the same metadata format for all the disks
but consider:
that'd require rewriting 12 floppy disks
ugh. my test read was stopped by disk #5 being bad. it wasn't showing bad sectors when I formatted and wrote it an hour ago!
I can't believe these 30 year old floppy disks are unreliable.
DISK 8 TOO?!
how could this handwritten novel netware license disk fail me
now I gotta wait until my standing up recharges enough to go grab some more disks from the garage, as I only had one spare
originally I got 20 disks, and formatted them all. I tossed 7 failures, leaving me at 13 working disks, so I had 12 for the video + one spare.
I wasn't really expecting any of them to fail inbetween that format-test and now
okay I managed to go to the garage and acquire a sack of fresh new 30-year-old dumpster floppies. surely at least one of these works
Read all disks in 430.22s
336s is the maximum instant-disk-change speed, so I had 94 seconds of overhead, or about 10s to change each disk (with presorted disks)
my code is reading the whole of the last disk, which isn't exactly right. most of that last disk is zero padding
I'll need to figure out how to test that one easily, because other wise it only happens at the end of 7 minutes of disk-swapping
but I think that's enough (4 hours?!) for today. I've got the disk-reading code, the disk-swapping code, I've fixed the disks so I have 12 that work (FOR NOW), I confirmed I can read the correct data quickly, so next steps are:
* fixing the over-read at the end
* separating out the data from the metadata into a single stream for feeding to the video player
* integrating a swap-disk interface into the video player
* fixing the over-read at the end
* separating out the data from the metadata into a single stream for feeding to the video player
* integrating a swap-disk interface into the video player
Steve Syfuhs@foone just thinking about this and there is a super easy way to make this work where you don't risk the user taking too long to flip disks or put them in the wrong order: 12 disk drives doing one gigantic parallel read.
As long as you've sequenced the streams all you have to do then is sort by disk number. Easy peasy. Bonus is you can ramp the quality right up to the edge. I see no problems with this.
@SteveSyfuhs sadly I've had really terrible results with using multiple USB floppy drives at once
@foone filing a bug on the driver team would likely simultaneously make their day and also make them very angry. Still no downside.
@SteveSyfuhs I actually tested it on linux, figuring that would have a greater chance of working (no offense).
I'll have to try on Windows one of these days
llewelly
Colin
Stormgren@Stormgren I'm not! I formatted the disk and put my own shit on it