06 января 2017

                              News
                         by Alone Coder

A few days passed, and a lot of hot material arrived, so I must
make a new issue just after :) Anyway I have New Year holidays
till 8th of January.

Hippiman/Conscience released a New Year game named Dizzy Rescues
Santa using DizzyAGE platform for Windows with a smooth scroll 
patch by Titus. I completed the game by 1th of January, and
while I played, I broke one Christmas tree ball and flew to
Great Ustyug (home of Ded Moroz) :) The next day Louisa won the
game, she also found a giant present from Ded Moroz. If anybody
has troubles with presents, there are just 25 of them (not
counting that giant one). English translation will appear soon.

Deathsoft released a couple of new versions of Unreal Speccy 
emulator and tries to reverse engineer the schematics of disk
controller chip 1818VG93 using micro-photographs. A half of the
chip appeared a programmable logic array.

Winter DiHalt Lite is not joined party again, however there is
preliminary evedence that there will be around 30 people - twice
the old number. Louisa and I don't go there this time.

I fixed several errors in my 3D engine and suddenly found that
its working zone with current math is limited to just 10x10 unit
cubes. This is enough only for inner rooms of Total Eclipse. 
That't how I struggled for 8-bit calculus. Let me think more :)

Polist culturologist Piotr Marecki who went to summer DiHalt
with Yerzmyey, wrote a big questionnaire for his future book. I
translated the questions. Piotr said that you can answer in
Russian as well. Send him your answers preferably before the end
of January (the deadline is 1Sth of February)!

Nikolay Amosov (AmoNik), formerly the editor of AvroNews
electronic paper, sent the news from Saint Petersburg (and a big
addition to SPbZXNet material from the previons issue of
ACNews - see the next article). 

                              News
                          by AmoNik

Sergey Smirnov (Gogin) revived his project to port Super Mario
to ZX Spectrum. There was a demoversion of its first level at
Your Game compo. Now Gogin rewrites the engine from scratch. 

Alone> As he said, "using the experience of last 15 years". 
And the engine is now in colour! (By the way, I tried to write 
my own port of Super Mario three times and always abandoned 
it.) 

Here you can see how much time it takes to output the graphics:
    https://monosnap.com/file/jбHxgvxd1YpJOCBgUaxzCdMdMOXЗaY
The video looks twitchy but as Gogin said, that because of
screen grabber, and the engine is smooth in reality. Even at
3,54 MHz, without turbo mode. 

Now his Scorpion is broken, and I try to give him
recommendations according to symptoms. Although he also has a
ZX Evolution. 

Last year I wrote a new HDD driver for iS-DOS operating system.
The old driver by Valery Kozhevnikov (Jason Frost) that I used
for a long time, had several flaws and mistakes that I found
when I made my own driver based on it.

The cause for me was a big stock of CompactFlash cards I
collected after I bought Amiga 600 in 2008. I purchased many
pieces every time, trying to find the one that would work with
Amiga via PCMCIA. It appeared that not all cards work this way,
and the ones that work might be very slow. In the end I found
the only card to work fast (Sony), and I tried to find use for
the rest.

I bought a CF-IDE adapter in a computer store. This adapter was
attached instead of IDE cable. I connected it to my Scorpion and
started experiments with the cards.

Alone> I also had an adapter like this, but it was made by 
CHRV/NedoPC. Sellers at the local computer store said that such
a device doesn't exist at all. I used it for a long time with 
Pentagon 1024 and ATM Turbo 2+ until I gave Pentagon 1024 with 
the adapter, a card, a CD-ROM, and a disk with Time Gal game to 
T(c)S. It's a pity but I have no place for it at home now...

It appeared that Scorpion is very choosy. Sometimes it can't
create a partition table on a card, sometimes it loses
information about mounted virtual disks. If it works without
bugs, it's very slow. Finally I found a few cards from a dozen
that worked fast and without bugs with Scorpion.

There I first encountered a bug: some programs didn't work.
Namely the HDD_0.1 plug-in for Real Commander and the iS-DOS
HDD driver. Both buzzed while disk access. I couldn't find the
author of the plug-in although we live in one city. The author
didn't release the sources but it was clear from the manual that
the bug in the plug-in and the driver had the same cause. The
plug-in was just a GUI wrapper over the driver by Jason Frost.
I studied the code in a debugger and made sure that the driver
code is the same. So, if I fix the iS-DOS driver, I fix the
plug-in.

I tried to decompile the plug-in several times, but every time I
got a non-working source file. The compiled program appeared
shorter than the original, or didn't work at all. Instead, I had
the sources of the iS-DOS driver and I modified it for testing.
However this didn't help me to understand why it buzzes.

I knew the HDD protocol mostly after Vega's (Vlad Sotnikov)
articles in ZX-NEWS electronic paper. I also had some excerpts
from an old version of ATA standard. According to these
documents, the driver worked OK. It didn't with CompactFlash
though. It was apparent that in the later versions of the
standard something got "broken", and the modern devices work
some other way. To be sure, I ordered a few old cards sized
32 to 512 MB, but they also failed with the old driver. So I
knew that changing a couple bytes is not enough, I must write a
new driver to comply with newer standards.

There was a moment I sold my GMX and downgraded my Scorpion
board to the initial state. But the board didn't work. After few
days, I switched to a back-up board, and also made an add-on
circuit to open DOS ports for fast drivers. Thinking twice, I
decided that I better patch the SMUC controller to open its
ports.

Now that I had access to the ports, and I occupied myself with
assembler, using hard copies of driver sources. I had few of
them: Jason Frost's driver, two versions of driver by Oleg
Grigoryev (Ruiner), and a couple of my hacks. Once I made my
versions of Grigoryev's drivers, with LBA support and something
else. It was easier for me to write my own that to understand
how other's code works. I used only sector read and write
procedures from the old drivers, and a structure for partitions
with a procedure to find sector address using the partition
information. The rest was mine from scratch.

I started with small fragments of block reading procedures that
read a given number of 256 byte blocks. I wrote them in ALASM 
and tested them. Then I made an iS-DOS driver from them. It
couldn't write but it reads very-very fast. When I debugged it,
I returned to ALASM and started the hardest part - saving
procedures.

iS-DOS works with 256 byte logical blocks, and a physical sector 
is 512 bytes, so I should write three distints saving
procedures. One writes blocks that lie in the first half-sector.
The second one writes blocks that lie in the second half-sector.
And the last one writes a full sector.

The logic was apparent, and I made the procedures in no time but
lost a few days debugging. I created one more partition on a
disk special for tests, and I hard-coded its coordinates into
the driver. Then I copied files there and looked what happens.
To be sure, I excluded possible reading failures of my new
driver, making all the reads via old slow driver. So all the
encountered bugs were because of my new code. And they were
there. There were not many but I spent much time on them.
I even wrote specific test programs and specific test files with
certain content to see what part of it fails and appears at a
wrong sector. Then I added debugging watches and looked at then
in a debugger. When I couldn't find a bug in my procedures, I
suspected that there are bugs in external code that I copied
from other's sources. I was right. I fixed them, and the driver
got worked.

I added the second IDE channel and an install/partition
configuration program. Then I made an official release. But in
2016 I changed a card for a bigger one (copied the contents
sector by sector) not testing the driver with it. That was my
mistake. Some time after that I found there files with checksum
error. Short inspection showed that some of the files were
already buggy at my old HDD (they were screwed up by Jason
Frost's driver). However, some of the files were destroyed with
my new driver. I opened the debugger again.

Test files helped me again. I found the error in a half hour.
This was not a driver bug but a standard change, again. Old
HDDs increased sector/head/track after reading a sector. So an
old driver didn't count these numbers every time. This gained
speed. My first CF card didn't change the numbers so I added
code to increase then manually using the disk geometry. However,
the other card returned random values.

So I should rewrite the driver again. For every sector that is
read or written not in a group, I count the coordinates and
write them in the registers of the device. So the new driver
will work with all the devices, includnig the old card and the
old HDD.

Bugs like that forced me to find a way to backup the iS-DOS
partitions on disk. I don't know a PC utility that can access
MFS partition and pick iS-DOS partitions from there. I'm not
very positive about using a disk editor with manual operations.
So I decided to backup disks via cable. First I thought about a
modem, but my ISA slot at SMUC didn't work. Something burned up
when I attached an internal modem - I connected power wrong way.
It might be unrecoverable if Altera died. I even didn't
investigate. I had Arduino Mega 2560 at my table, so why not use
it for this? At the same time I'll remember how to code in C.
I "remembered" the last time when I planned a bus-less
PS/2 keyboard controller for Scorpion. I stopped after few tests
- the device is too slow to "press" keys returning values on the
bus.

Today trd disks are already copied OK. I tested the speed with
simple text message, and it appeared that Scorpion can do
40.5 KB/s. With disk access (read sector, send it to the cable)
it will be slower, and a full trd is transmitted in 39 seconds.
This is for per-sector read and TR-DOS emulation via #3D13. In
iS-DOS, it will be a lot faster because of fast HDD driver.
Namely 39.5 KB/s for Scorpion in turbo mode. However this board
has non-usual turbo schematics that is faster than the original.