ACNews #66

News - AmoNik has measured the power consumption of IDE different devices, I am mostly occupied with NedoLang compiler

                              News
                        by Alone Coder

Today I am mostly occupied with NedoLang compiler.

The main part of the program (10 modules, formerly more because
of header files), that in fact compiles, is already converted in
NedoLang language, module after module. I added few structures 
in he language to do this, namely import of constants,
variables, and functions (header files and#include are not
used). This is not the final version of the syntax yet because I
still can't compile NedoAsm -"switch" and defines are needed
(or maybe enums that are also missing). Anyway, I can sum up the
situation after 6 months in the project:

- 346 KB of sources, not counting the GUI. 191 KB of them is the
compilator itself (that generates assembly text). The rest is
asm tokenizer, asm exporter (de-tokenizer), and the assembler.

- with comments off, the output assembly text of the compiler
(that was 191 KB) comprises 500 KB. While compilation of the
largest module, 11 KB of labels is generated (there was 10 KB
reserved before, now it doesn't fit).

- binary file generated from this assembly text is 57 KB
(including label buffer and other memory buffers). Assembler
uses around 60 KB for labels (I reserved 128 KB because the
assembler acts as a linker and must know all labels in all
modules).

This is too early to release. If I add filesystem and screen
output, the compiler will exceed the address space, the
assembler already did. And how many floppies would be needed to
self-compile... three at once?

I can work in different ways:

- optimize the compiler source (for example, remove superfluous
text messages);
- make better code generation (register allocator at least);
- support far calls to write as many code as you want;
- support far pointers to use as much memory as you want;
- somehow remove local labels during assembly;
- assemply part by part and add a linker;
- stream the assembly text to save disk space. For this, I need
to buffer constants and variables in memory. Now they are just
emitted in another assembly text. However, if I stream assembly
text, I can't compile programs part by part (is it needed? we
will know when we measure compilation speed).

I don't know how much time any of these jobs will take. There
were other ideas - overlays for example (proposed bydeathsoft
who has hand-counted around 4000 transistors in VG93 disk
controller, and has found commonly withLord Vader a few bugs in
NedoLang using three different compilers). But I don't want to 
bind the system to overlay loading. I want the whole system
(editor, compiler, assembler, and programm running) in memory.
For this, OS support is needed, of course.

I'd better run self-compilation as early as possible, at least
to know what places to optimize for speed. I don't know the
speed yet. Judging from the size of assembly file, today's
version will take more than a minute at 3.5 MHz. It is desirable
to fit in 1 minute.

Maybe integrating the assembler in the compiler is profitable.
Of course, built-in assembler will not be as fast as a dedicated
one (that reads tokenized text). Although the future work
scenario for NedoLang assumes prototyping a program in
high-level language then converting the slowest parts in
assembly. There's no need to convert all parts. However, someone
might want writing in assembly from scratch.

As a compromise, I can include the tokenizer in NedoLang, so it
will output tokenized assembly text. This adds 98 KB to 191 KB
of sources mentioned earlier, and will not fit in the current
binary.

The third option, most comfortable for the end user and the
hardest to do - re-write the assembler to compile both
separately and inside the compiler (so the binary will be twice
as big). However, we need assembly text output anyway.

Of course, I can make the language simpler, like Pseudo-C, with
one loop command, two types, without function arguments, modules
etc. But this will be unusable for real-life tasks.

The reserve method is re-writing the compiler in assembly, but
that will stall the development of the development system. We
shall see what we shall see. Maybe the speed will be sufficient.
Especially with background compilation :)

                            * * *

This time I couldn't escape neither toJohn Silver, nor to the
gathering at local film studio (namely we visited the studio
withLouisa but found a quiz there, not lessons). John said he
writes a song. In general, he repairs PCs.

By the way, as I wrote this issue, my very old PC has broken -
first the keyboard went crazy, then something clicked, the
computer stopped and refused to reboot. The mouse was played to
the end byDasha even earlier. A couple of laptops remained. All
of this at the very moment whenWanderer (author of MURKЗЗ20,
MURKЗЗ27 games) asked me to copy Spectrum floppies with a source 
for an unreleased game. Somebody else should do it. Suddenly I
realized that all my working computers are at least 6 years old.
This was unthinkable in 2000, even at Speccy...

There are marks of music by other authors, and the articles for
the e-zine are also being written. Sadly there's still no
graphics.
AmoNik sent a lot of data - old archives, echo conferences,
memories.

In between times,AmoNik has measured the power consumption of
different devices (I've seen a dispute in the Internet about
this today) - see in separate article. He also fixed the IDE
driver for iS-DOS, written in 1999 byOleg Grigoriev. It had the
same problem (see#65 ) - it didn't work with modern devices,
namely his CompactFlash card that doesn't return sector
coordinates in registers.

Let him say:

┌───────────────────────────────────────────────────────────────┐
I copied the original driver inSMUC.ZIParchive in mid-2000s 
from the author,Oleg Grigoriev.He wrote a fast and compact 
driver but hadn't finished an installer for it, arguing that he 
had few iS-DOS partitions at his HDD, so he had hard-coded their 
coordinates directly in the driver. As far I understood,Oleg's
driver is based onJason Frost'sdriver with some update. After 
the update the driver became shorter and more logical. 

In 2006, I made an LBA version of it, but for some reason the 
new driver wasn't tested successfully, and I forgot about it. In 
2017, I was cleaning test versions of programs fromTEMP
directory at my system disk, and I found this driver in a 
directory namedNOTWORK.ING.There weren't any text notes. So I 
couldn't know what was broken. I copied the driver just in case, 
and tested it at a test partition. The driver completed reading 
tests successfully but failed in writing. 

I converted the driver in ALASM and wrote a test suite. After 
debugging for some time, I found and fixed the fragment not 
working at modern devices. Then I remade partition switcher, 
modelling it after HDDLSERV, then I tested the driver thoroughly 
by copying my 8 MB system disk to test partition of MASTER and 
SLAVE devices. Checksums showed that the driver in working. 

As always, the driver requires open ports of SMUC IDE and works 
in both iS-DOS Classic and Chic. I think this driver is better 
because it's 230 bytes shorter than my HDDLSMUC. I also wrote a 
universal disk mounter named HDUNISRV. It works with either 
HDDLSMUC or RHDDLBA depending on the driver name in the command 
line. For example: 
hdunisrv Rhddlba - mounts disks to Rhddlba driver.

The originalGrigoriev'sarchive is also here. There are several 
versions of his driver, some sketches of install utility, and 
testing procedures. Maybe somebody will be interested.