Re: the Courier font family and nroff history

[G. Branden Robinson]

At 2024-04-11T15:28:36-0400, James K. Lowden wrote:
> "G. Branden Robinson" <g.branden.robinson@gmail.com> wrote:
> > Yes, I reckon trapping into the kernel for every byte (or pair of
> > bytes) written to the screen would indeed have eaten all your
> > performance, and been pointless when there was no access protection
> > in any part of the address space anyway.  
> 
> Hmm, well, "trapping in to the kernel" wasn't so expensive because
> there was no context switch.

Right.  Had to laugh at myself here.  I have dwelt so long in the
civilized world that I forgot what it was like to have an "OS" that
didn't support CPU mode switches.

> Yes, interrupt processing was relatively expensive in terms of CPU
> cycles.  I remember implementing XON/XOFF because a 6 MHz 80286
> couldn't keep up with a 1200 bps modem.  

I vaguely recollect that this was solved by putting UARTs with (bigger)
buffers on the motherboard.  A 16-byte buffer on the 16550 comes to mind
for some reason.  I guess that was enough to make the difference if it
meant an interrupt handler grabbing up to 16 bytes at a time instead of
one.  I don't remember reading that the Intel 8250 UART had any
buffering at all.

> It wasn't 100% pointless to use the BIOS because it gave you a kind of
> device independence.  Different video cards resided at different
> addresses (above 640K) and had different modes.  The BIOS calls
> insulated the application from some of that.

Right.  I remember CGA and MDA occupying different memory regions, so
you could drive a dual-head setup, but I never saw this done in
practice.

> I'm the last one to defend it; present day Windows systems still
> suffer from brain-dead choices in MS-DOS that ignored contemporary OS
> thinking at the time.

And then made more of its own.  Hitching their wagon to UTF-16 for
character encoding when a machine register was already available seems
like a particular head-slapper to me.

> But it's also a "he is us" problem: MS-DOS vanquished numerous
> alternatives not because it was better but because it offered so
> little.  Except for the "disk" part, applications largely ignored it.

This helps me to reconceptualize MS-DOS a little bit.  Not really an
operating system so much as a bootloader with a fairly featureful
(bloated?) monitor program and a mass storage stack (driver, file
system).

It's so strange to me that Intel copied the VAX's 4 privilege levels
into the 8086, and Microsoft totally ignored them.  Whatever was going
on there, it wasn't engineering.  This must be why its first 100
employees or so are all categorically described as geniuses.

> They wrote to the hardware, and were faster for it.  The market for
> well architected slow systems is vanishingly small.

You can probably strike "slow" from this claim.  I don't want to say
that the most short-sighted people in any organization become purchasing
managers, but, well, I just did.

> If in doubt, ask IBM about TopView.

I had never heard of it, and I feel myself unusual in being acquainted
with relatively esoteric stuff like the iAPX 432 and the Sprite
operating system.  :P

I wish I'd gotten more familiar with VMS, but DCL was so jagged and ugly
that I was repeatedly driven away.  Too bad about all the fiefdoms in
DEC.  An intelligent fusion of VMS and Unix could have ruled the world.
On the other hand, we'd have been at the mercy of yet another monopoly.
I'm attaching a piece about DEC's struggles.

> Debugging by the light of an ADM-3A.  Those were the days!

Closest I got was an ADM-5.  It seemed hopelessly retro at the time,
like someone hauled it off the set of the original Star Trek series.
When I started out on Unix without a phone line in my way, I was in
love with the Sun SPARC console for a couple of months until I
discovered the X Window System.  I've remained loyal to xterm ever
since.

Regards,
Branden

DEC - The Mistakes that led to its Downfall.pdf
Description: Adobe PDF document

signature.asc
Description: PGP signature