OS/2 1.x versions were released in 1987..90, so the contemporary Windowses would be primarily 2.x (1987..89) - no protection there, right?

Grzyb wrote on 2023-04-08, 08:26:

OS/2 1.x versions were released in 1987..90, so the contemporary Windowses would be primarily 2.x (1987..89) - no protection there, right?

Yes, maybe. Not sure, I guess it depends on the point of view. 🤷‍♂️
Personally, I simply go by the look of Presentation Manager.

OS/2 v1.0.. Like DOS (text mode)

OS/2 v1.1.. Like Windows 2.x

OS/2 v1.2x and v1.3.. Like Windows 3.0

When it comes to WLO/Willow, I'm thinking of OS/2 v1.3 only.
The v1.2 series could already run it, but related like a beta version to v1.3, I assume.

Version v1.3 was the mature version that everyone had accepted as being usable.
It also supported SCSI based HDDs, CD-ROMs and had most device drivers at hand.

Protection.. Good one, good point I mean. 🙂👍
I assume that Windows 3.x running either of the two Protected Mode kernals qualifies. 🤷‍♂️
Mainly, because of the involvement of a memory management unit, that can prevent/allow access to segments.

The 80286 also has a separate address bus decoding unit that can create the operand of segment address, index, base address and displacement.
Prior the 8018x series, the 8086/8086 had to do this in ALU. This is not directly relevant here, maybe, but gives an idea about how the 80286 was a more robust architecture.
Memory management was way more sophisticated in 80286 systems.

Windows 1.x had no protection, I think, since Windows programs were still free to do address arithmetics (ie, could play with segments+offeset, like DOS programs do).

Windows 2.x is.. more complicated. I'll to explain why I think that.
Firstly, because Windows/386 2.01 was the very first release of Windows 2 (as an OEM version). Followed by plain Windows 2.03 as a retail box.
That means that Windows/386 was the real basis, the real ancestor of Windows 3.0 and 386 Enhanced-Mode.
Standard Mode/286 support was the last addition, interestingly. In Windows, I mean.
OS/2 started with 286 16-Bit Protected Mode (used by Standard Mode kernal, except on Windows 3.1x on 386+).

Edit: Quick info here: The version numbers and the /286, /386, [plain] editions were not directly related.
Windows, Windows/286 and Windows/386 were available with various internal 2.x version numbers.
At its core, there was always WIN.COM or WIN86.COM in the box that could run on a PC with 8086/8088 CPUs.
The /286 and /386 editions merely had extra features for AT and AT386 PCs.
The former had himem.sys aboard and could use the 64KB in the HMA, but that feature conflicted with DOS 5/6 (DOS=HIGH).
Whereas the latter could provide EMS via V86,-similar to EMM386 years later-, and added a multi-tasker. Windows 2.x itself was rectricted to a single DOS box.
- Not unlike DOS support was restricted to a little DOS box itself on 16-Bit OS/2 (version 1.x). Those parallels are astonishing, IMHO. 😳

So this means that certain elements of memory protection were already in Windows/386.
Not for Windows applications yet, but DOS applications and Windows drivers
(V86 mode, as used, requires VXDs or similar tech, even if it's just for mouse/keyboard virtualization or video driver enable/disable).

That being said, what's even more uncertain, is how applications relate to memory protection.
Since Windows 3.o, newly made Windows 2.x applications are being encouraged to be "clean".:

That means they don't fiddle with memory anymore and that they play nice with scalable fonts.
That's why the warning message about compatibility pops up on Windows 3.x.
- But merely in Protected Mode! Real-Mode Windows doesn't complain, as far as I remember.

So I assume that Windows 3.x in Protected-Mode is similar to OS/2 1.3.
Both are similarly advanced and WLO can translate between both worlds.

Interestingly, though, WLO requires a strange type of Windows application it will host:

a) a Real-Mode application which
b) doesn't use memory arithmetics (which was typical for Windows 1.x/2.x programs)
c) and uses pure Windows 3.0 API and file formats

In short, it requires an unicorn of an Windows application. A model type, the perfect "mister clean", the nice guy.
You may imagine it as the Pierce Prosnan of the Windows world (ie, sexy, smart, charismatic). 🙂

That's something interesting, I think. It maybe means that WLO does prefer a Windows application
that fits snugly into 64KB segments or restricts itself to "small" type memory model.

Please excuse me for being a bit uncertain here, I'm just a layman/hobbyist, after all. 😅

Maybe it just means that those Windows applications compatible with Real-Mode
are not interfering with OS/2's own memory management.

They essentially act like COM files in DOS. Not technologically, but in terms of behaving.
Both stay within their boundaries, make no attempts to break out, ie don't make far calls, in practice often stick to plain 8086 code.
In essence, they're both very modest/humble. Simply that.

Again, I'm just a layman here. An expert may say that these comparisons are nonsense. He/she/they may be right.

That being said, it's fascinating to tinker with that stuff and talk with others about it.
Even if we're sometimes walking in the dark, so to say.
Because, step by step, things will reveal themselves eventually. IMHO. 😉

Edit: Formatting fixed (at home on PC).
Edit: Re-wrote a sentence (various instead of different).

Quick update. I've attached the modified GNU Chess binary for testing purposes. It has a red icon, so it can be better separated from the original. Hope that's okay.
Maybe some of you can find out what's wrong with the binary, or simply wants to see the how it looks like.
- I was wrong about the chess engine not working, also. It must be working due to the internal benchmark being functional. 😀

Edit: Screenshots attached.

Quick update. Got PM Chess running on OS/2 v1.3.. 😁
PM Chess is the native OS/2 port of GNU Chess for Windows by author K. Cedola.
Ie, it's based on GNU Chess 3.1 (Windows) by D. Baker.

In this version, the chess pieces can be selected normally, like with Windows 3.x.
Unfortunately, there's no source code included in the archive I've found.
It maybe was made available to download separately, originally.

Highly interesting, when it comes to the chess game. Perhaps MrFlibble knows where , or perhaps he can find the source for that game. After all, he has made some great threads about either Open source and free games 😀 So perhaps he can help with finding the source 😀 Other than that, you could perhaps find the source of an old game. And port it to OS/2 1.x Willow 😀

I'm no MrFlibble that's for sure, but here it is 😁
Found it at (you guessed it) archive.org, what else.
https://archive.org/details/ProfitPress_Mega_ … 31_OS2_1991_Eng

As Jo said it was a separate package. There's some other versions on that same CD-ROM too. Anyway, nice thread 😁

WolverineDK wrote on 2023-04-11, 19:32:

Highly interesting, when it comes to the chess game. Perhaps MrFlibble knows where , or perhaps he can find the source for that game.
After all, he has made some great threads about either Open source and free games 😀 So perhaps he can help with finding the source 😀
Other than that, you could perhaps find the source of an old game. And port it to OS/2 1.x Willow 😀

Thanks, the idea sounds fun. I *think* some Win 3.x games had source included.
I vaguely remember "Hyperoid", for example. Because there was a 32-Bit version, too.

Which means, maybe someone else than the original author had ported the game to Win32s.
But I'm not 100% sure. I can also try porting little utilities and "fun stuff" programs, if they ship with source code.

I think this would be fun to the retro/vintage community.
Imagine, if we could finally run some interesting applications on OS/2 v1.2x/v1.3.

This would be another reason to build some hot-rod 286 machines running under OS/2. 😁
I mean, Windows 3.0 already works in the DOS box of OS/2 1.x, but it's a bit short on memory.

- Sure, running Windows 3.1 natively is always a nice alternative, of course.
But using OS/2 would be a refreshing new experience, maybe.

Not much unlike giving another DOS compatible OS a chance.
It would be interesting to tinker with.

mr.cat wrote on 2023-04-13, 12:12:

I'm no MrFlibble that's for sure, but here it is :D Found it at (you guessed it) archive.org, what else. https://archive.org/det […]
Show full quote

I'm no MrFlibble that's for sure, but here it is 😁
Found it at (you guessed it) archive.org, what else.
https://archive.org/details/ProfitPress_Mega_ … 31_OS2_1991_Eng

As Jo said it was a separate package. There's some other versions on that same CD-ROM too. Anyway, nice thread 😁

Thank you very much! ^ ^

The source you found can surely help troubleshooting Gnu Chess for Windows!
- Maybe I'll find out why WLO has problems with the chess board.

Quick update. I had a quick look at the source code of PM Chess.
- I'm just a layman when it comes to OS/2 and Win16 programming, but I think that board.c has some useful information.
The OS/2 port has an extra part about figuring out mouse pointer position. Maybe that's in parts related to the problem with WLO ?
It could be that GNU Chess for Windows can't get the actual mouse pointer position if run on OS/2, for some reason.
Edit: Or more precisely, the translation between mouse pointer coordinates (x, y) and chess board coordinates (A...H, 1-8) isn't working ?
And that's why the fields can't be selected. Maybe. I mean, the coordinate system is made upside-down between Presentation Manager / Windows.
So it's not that far fetched. Maybe an API call/function call was made without proper declaration of a function. Windows 3.x maybe just ignores it or uses defaults, but WLO has trouble.
On the other hand, the other Windows programs like Paintbrush work as expected when run through WLO on OS/2.

mr.cat wrote on 2023-04-13, 12:12:

I'm no MrFlibble that's for sure, but here it is :D Found it at (you guessed it) archive.org, what else. https://archive.org/det […]
Show full quote

I'm no MrFlibble that's for sure, but here it is 😁
Found it at (you guessed it) archive.org, what else.
https://archive.org/details/ProfitPress_Mega_ … 31_OS2_1991_Eng

As Jo said it was a separate package. There's some other versions on that same CD-ROM too. Anyway, nice thread 😁

Yo brother 😀 (said with respect and politeness) 😀 I am no programmer, but that is awesome 😀 respect ! 😀

Jo22: As you said, you think it could maybe be fun for the retro/vintage community.. heck let us be a bit blatant, if it gets enough traction, then perhaps a certain indie-retro-news site ( hint hint 😉 I will not link to the site, but it is a great site for indie retro news 😀 ) will perhaps read this thread, and others too. And it will embrace this idea. Cause it is about embracing, not forking and or dividing people 😀 just like the interview I saw/heard with Bob from retrorgb and Hans Baier, whom is programming the Mister FPGA framework to other FPGA and ARM platforms. With an embrace, and he calls it Mistex. Cause it is a play on words, such as Mister and LiteX (an FPGA framework) or Mister FPGA expanded 😀 And the cool thing about the framework, he is using to port the Mister FPGA framework, cane be used in such a way, that you only need the LiteX . to set it to either make the core for any console or machine, as either the DE10 Nano FPGA board, or for that matter a Zylinx FPGA and so forth in the framework he is using to port the Mister FPGA framework. And he is not looking for splitting/dividing people but embracing them and so forth 😀

Jo22: you are doing a fantastic job. That is what I am saying, also I hope my last post is seen the same way. With me dreaming, and bringing positivity to the thread 😀

WolverineDK wrote on 2023-04-16, 00:55:

Jo22: you are doing a fantastic job. That is what I am saying, also I hope my last post is seen the same way. With me dreaming, and bringing positivity to the thread :-)

Thank you very much for your support! ^^
I'll do some more experiments soon, I'm working on another short OS/2 video, too.
Maybe I have some free time at the weekend to finish things.

Thanks also to mr.cat, the shareware link contains a copy of Hyperoid w/ source code. :)

Edit: Interesting, the archive contains old OS/2 1.x graphics drivers not available on files.mpoli.fi..
Like Trident 8900/9000 graphics drivers for OS/2 1.2 and 1.3 (256c in 640x480 and up)..
A few rare applications and OS/2 1.1 drivers are included, too. I also saw drivers for digitizers (graphic tablets, scanner).

Hi. Just saw an interesting IBM PC/AT video in the YT channel Adrian's Digital Basement..

https://www.youtube.com/watch?v=bxnJhUyhszk

- Now that's what I call a properly configured AT.
It has network card, SB 1.x, 1200 or 2400 Baud modem, VGA and most importantly, 4 MB of RAM. 😉

Jo22 wrote on 2023-04-23, 07:03:

Hi. Just saw an interesting IBM PC/AT video in the YT channel Adrian's Digital Basement.. https://www.youtube.com/watch?v=bxnJhU […]
Show full quote

Hi. Just saw an interesting IBM PC/AT video in the YT channel Adrian's Digital Basement..
https://www.youtube.com/watch?v=bxnJhUyhszk
- Now that's what I call a properly configured AT.
It has network card, SB 1.x, 1200 or 2400 Baud modem, VGA and most importantly, 4 MB of RAM. 😉

Nice, but it's no longer a genuine IBM PC/AT 😜
With the genuine IBM motherboard, installing 4 MB of RAM wouldn't be so easy.

NOW, A CHALLENGE!
Find me a price list - from back in the era - offering a 286 PC with more than 1 MB of RAM.

Because I seriously doubt if such machines were ever sold stock...
just picked a random magazine, and found the first listing with a 286:

(BYTE, March 1991, page 2)

Grzyb wrote on 2023-04-23, 22:39:

Jo22 wrote on 2023-04-23, 07:03:

Hi. Just saw an interesting IBM PC/AT video in the YT channel Adrian's Digital Basement.. https://www.youtube.com/watch?v=bxnJhU […]
Show full quote

Hi. Just saw an interesting IBM PC/AT video in the YT channel Adrian's Digital Basement..
https://www.youtube.com/watch?v=bxnJhUyhszk
- Now that's what I call a properly configured AT.
It has network card, SB 1.x, 1200 or 2400 Baud modem, VGA and most importantly, 4 MB of RAM. 😉

Nice, but it's no longer a genuine IBM PC/AT 😜
With the genuine IBM motherboard, installing 4 MB of RAM wouldn't be so easy.

Hi there! 😁

Gratefully, it's not an IBM original. 😁 It needs a clone PC to gain true PC or PC/AT compatibility, after all.
- Just kidding. Well, more or less. 😉 IBM made good hardware, but its software...

Thing is, even back in the 1980s, it was common practice to replace an IBM BIOS at some point (or the whole board and keep RAM/CPU).
The hardware was solid, the chassis beautiful. But the original IBM AT Model 5170 had a horrible BIOS, without CMOS Setup Utility in ROM.
Which we can't blame it for, of course. It invented the whole thing, after all. So it had a prototype status, essentially.

Anyway, for changing settings, an external utility was required, thus. Like with Compaq PCs of the 80s/early 90s.
That external utility on the IBM diskette didn't even feature 1,44MB floppy option, however.

Something like GSetup (a third-party freeware) would have been needed.
Or DRIVER.SYS/DRIVPARM on DOS. Or an equivalent to "X2M BIOS" (didn't exist then).

A Type 47 HDD entry was missing, too, I assume: https://vintage-pc.tripod.com/types.html
By the mid-late 80s, this was going to be unacceptable. Capacities of fixed-disk were increasing so quickly.
The setup utility for the Commodore PC BIOS of 1987/88 had listed a 112MB model entry, even.
I mean, let's imagine we had no XT-IDE Universal BIOS these days! Unthinkable. The replacement BIOSes had a similar significance at the time.

Replacing the AT BIOS on a 5170 motherboard by a "backup copy" of a friend's Quadtel or AMI BIOS wasn't being unheard of thus.
I do imagine it was as about as half as common as an upgrade as installing a NEC V20 in an XT PC.
Just less being talked about openly at the time, I suppose, for obvious reasons.

I mean, there always was a computer freak or ham around the city who could burn you an EPROM for a rip of an VHS, a beer or another little gift.
- In the C64 days, EPROM programmers were a common item among computer enthusiasts (homebrew was more of a requirement than a hobby),
because only a few parts were required to build them. The 90s were a different time, though. Prommers (=PROM-mers) were highly expensive then, albeit some users kept their old equipment.
It took roughly 15 years up until EPROM programmers started to become common among computer users again (G540, TL866 and their clones).

The reverse was also true, I think. In the early days of XT clones, the ROM BASIC still mattered (and when BASICA was used, before GW-BASIC replaced it).
That's why some poor souls had decided to install an original IBM PC/XT Model 5160 BIOS in a Taiwanese clone board.
Which was a problem if the BIOS wasn't V20/V30 ready. Another reason for using replacement BIOSes at the time. 😉

To its defense, the board in the video is an earlier model without SIPP/SIMM slots, requiring 3/4 the size of the original.
It uses classic DIP/DIL RAM chips instead, so it might be from the 1987/1988 time frame (roughly).

Edited.

Edit: It seems that the original IBM AT BIOS has a built-in speed check, which halts the system if it's running above 8 MHz.
In that case, a replacement BIOS can help, too.

Grzyb wrote on 2023-04-23, 22:39:

NOW, A CHALLENGE! Find me a price list - from back in the era - offering a 286 PC with more than 1 MB of RAM. […]
Show full quote

NOW, A CHALLENGE!
Find me a price list - from back in the era - offering a 286 PC with more than 1 MB of RAM.

Because I seriously doubt if such machines were ever sold stock...
just picked a random magazine, and found the first listing with a 286:
dell286.jpg
(BYTE, March 1991, page 2)

This is tricky, because at this point in time, the 80286 PCs were considered to be lower end.
The last heyday of the 80286 platform was in 1992, before its popularity declined a lot the following year.
By 1993, many 32-Bit games came out and subsequent Windows 3.x relases dropped Standard-Mode, and with it, 80286 support.
In essence, EMS aware applications (DOS utilities such as SmarDrive, MSCDEX) and GUIs (Windows, PC GEOS) needed memory beyond 1MB the most.

DOS Days says this about the situation in 1991:

"Desktop PCs

A broad range of desktop PCs were available in 1991. A typical desktop PC would be a 486SX-25 with 2 MB of RAM and a VGA monochrome or colour monitor.
A choice of 1.2 MB or 1.44 MB floppy disk drive and up to 200 MB internal hard disk.
They were usually bundled with both DOS and Windows 3.0 along with a serial mouse.

At the low end were branded 286-12 systems with 1 MB of RAM, a 40 MB hard disk and 14" colour VGA display at the $1,500 price point.
Budget no-name 286-12 PCs with Hercules-compatible monochrome graphics would be as low as $600 retail.

The biggest brands were Dell, Everex, Mitac, ALR, Hewlett-Packard, Acer and AST.

Moving up to the large mid-market meant you were in 386 territory.
A cheap 386SX-based PC would have 4 MB of RAM and come bundled with Windows 3.0 in addition to DOS 3.3 or 4.01.
With the same colour VGA as the 286 mentioned, this pushed the price tag up to $1,900.
To get a 386DX-based computer would add a further $400-$600 for the same spec PC, though you would also get an 80 MB hard disk."

Source: http://www.dosdays.co.uk/topics/1991.php

Now, a 386SX system isn't that different to a 80286, except for the CPU. The mainboard designs or chipsets used are (were) quite similar.
They may also used similar VGA chips on-board, SIPP/SIMMs (pairs of two, because of 16-Bit I/O), drew little power (80L286 CPU was very efficient).
So it wasn't too unrealistic to opt for a cheaper, 80286 based systems at the time with otherwise similar specs.

Edit: What I mean by saying this: It was possible to ask a PC seller (computer shop) for changing an exisitng PC model,
ie. if the motherboard could be replaced by a different model. Asking for a custom PC build, so to say.
Real computer shops did offer this at the time, I remember. They had real PC technicians and a little PC lab/workbench for assembling PCs.

Edit: Speaking of that "Budget no-name 286-12 PCs with Hercules-compatible monochrome graphics would be as low as $600 retail.":
My father indeed had such a configuration, too! In circa 1989/1990, I think.. But with an 80MB HDD and 4MB of RAM (because, full memory expansion?).

Jo22 wrote on 2023-04-25, 10:55:

This is tricky, because at this point in time, the 80286 PCs were considered to be lower end.

You're free to choose any point in time, in the era of OS/2 1.x - 1987..92.

My point is: OS/2 1.x was designed for 286 with 4+ MB of RAM, which... did never exist on the market.
Prove me wrong!

Grzyb wrote on 2023-04-25, 12:08:

You're free to choose any point in time, in the era of OS/2 1.x - 1987..92. […]
Show full quote

Jo22 wrote on 2023-04-25, 10:55:

This is tricky, because at this point in time, the 80286 PCs were considered to be lower end.

You're free to choose any point in time, in the era of OS/2 1.x - 1987..92.

My point is: OS/2 1.x was designed for 286 with 4+ MB of RAM, which... did never exist on the market.
Prove me wrong!

Hm, still tricky..

So far, I've merely found a real world application of such a configuration.

"The Library has started its automation activity in the year 1989, by procuring a PC-AT Intel 80286 with 4mb RAM, 80mb Hard disk, MS-DOS operating system."

Source: https://library.uohyd.ac.in/automation.html

Grzyb wrote on 2023-04-25, 12:08:

You're free to choose any point in time, in the era of OS/2 1.x - 1987..92. […]
Show full quote

Jo22 wrote on 2023-04-25, 10:55:

This is tricky, because at this point in time, the 80286 PCs were considered to be lower end.

You're free to choose any point in time, in the era of OS/2 1.x - 1987..92.

My point is: OS/2 1.x was designed for 286 with 4+ MB of RAM, which... did never exist on the market.
Prove me wrong!

I think the problem here is that DOS itself was limited to about 640KB at the time and that end-user PCs shipped with DOS or were intended to run on DOS.
Installing up to a megabyte made sense to manufacturers/sellers, because it allowed for Shadow Memory (640KB to 1MB) in addition to DOS base memory.
That improved performance of the System BIOS, as well as many option ROMs, like EGA/VGA BIOSes, SCSI controller firmware etc.

Edit: My points about EMS, printer spooler software and RAM disk still stand. Though that was likely not being considered by manufactures in first place.
Third-party manufacturers did address those problems, though. There were plenty of ads in PC magazines about memory boards (ISA) that featured these things.
Manufacturers like AST, QuadRAM, Bocca etc. They sold those memory expansions, for XTs, ATs, AT 386es etc.

That's why many advertisements offered a 1MB base model, while simultaneously provided memory expansion as an add-on (they made a big fuss about it).
I mean, let's think about it. Many VGA cards had 256KB stock, but sockets for memory expansions up to 512KB or 1MB.
That doesn't mean that users had stuck to those 256KB. Many installed video RAM themselves, later on.

If PC setups with +1MB were so incredible rare, then why did most 286 PCs have SIMM/SIPP sockets, at all?
- They already shipped with about 1MB installed (okay, cheapo models had 512KB). They could have soldered them in, like in some XTs and saving money.

Personally, I think that the OS did play a big role here and that the PC market of its time was unsure about what the future brings.
In addition to the memory shortage that happened shortly before, which shook the market quite a lot. This may caused sellers to be extra conservative here.

Windows /386 capable PCs, like 386 PCs, were being advertised as 2MB model - even in the late 80s.
Simply because Windows /386 was good at multitasking DOS applications and simulating EMS without special EMS hardware.

Unfortunately, OS/2 didn't have that ability at the time. It was the exact reverse to Windows /386.
Whereas Windows /386 did a lot for DOS applications, it did little for native applications.
OS/2, on the other hand, did a lot for native applications, but little for DOS applications.

That's why the situation was like in this story here:

"In October 1986, a version of PageMaker was made available for Hewlett-Packard's HP Vectra computers.
In 1987, PageMaker was available on Digital Equipment's VAXstation computers. Aldus PageMaker 2 was released in 1987.
Until May 1987, the initial Windows release of PageMaker was bundled with a full version of Windows 1.0.3;
after that date, a "Windows-runtime" with no task-switching capabilities was included.
Thus, users who did not have Windows could run the application from MS-DOS.

Aldus PageMaker 3 for Macintosh was shipped in April 1988.
PageMaker 3.0 for the PC was shipped in May 1988 and required Windows 2.0,
which was bundled as a run-time version. Version 3.01 was available for OS/2 and
took extensive advantage of multithreading for improved user responsiveness.

In November 1989, Aldus PageMaker was released for OS/2 v1.1. Running under Presentation Manager, it cost $795.
It required quite a hefty PC at the time with 6 MB of RAM and a 40 MB hard disk.
In a first review of the product, it stated that "all of the extravagant claims for the Presentation Manager are suddenly vindicated.
Aldus Corp.'s Presentation Manager desktop publishing program doesn't do much that you can't do with the Windows version of PageMaker - it simply does it faster.
At virtually every point where Windows users must sit and stare at the hourglass icon, PM [Presentation Manager] users barely have time to blink their eyes before moving on.
The difference is so great that those accustomed to the delays in PageMaker for Windows may find themselves waiting
for the hourglass to disappear - and then suddenly realize it was never there at all.
The PM version allows you to open multiple documents simultaneously and let one document reformat while you work on another.
Even in a single document, printing, text importing, and text flow between columns and pages can all proceed in the
background while you continue with your work in the foreground. Text loads and flows at about twice the speed of the Windows version."

Source: http://www.dosdays.co.uk/topics/Software/aldus_pagemaker.php

I know, this story is getting old already, but it simply explains the situation so well!
OS/2 was a real operating system, something that Amiga, Atari ST and Mac users had used for years.
But with the exception to QNX, Xenix, Minix and some other systems like PC-MOS, there was none for PC.

By contrast to those, DOS was a glorified bootloader - up until the failed DOS 4 (-the Win Me/Vista of DOS essentially-), which tried to make DOS a real OS:
Things like an installable file system (IFS) were steps in the right direction to a modular operating system with external modules.
https://en.wikipedia.org/wiki/Installable_File_System

Unfortunately, what broke DOS 4's neck was the same what slowed down OS/2's acceptance - memory consumption.
Compared to DOS 3.3, DOS 4 was a memory hog. It required so much conventional memory that older DOS applications no longer fit in memory.
That's why MS-DOS 5 went back to an older design (3.3 kernal), while simultaneously introducing support for HMA and UMB.

Edit: What also must be taken into consideration here: OS/2 had a bigger memory consumption, but didn't require a more expensive 386 or tricks like EMS.
If we take this into consideration, then OS/2 wasn't that expensive. All your old IBM AT Model 5170 needed was a HDD and a memory upgrade.
On the Windows platform, a memory upgrade was required too, because the data that was being processed was equally big on either platform.

You needed either a more expensive 386 PC with 2 to 4MB of RAM for running Windows /386, or an old 286 PC with hardware-EMS running plain Windows 2.x.
But those EMS memory boards weren't dirt cheap, either. That's what I mean. 640KB were not enough, either way. Lucky those 286 PC owners with EMS support in the chipset..
OS/2 had at least the ability to use memory above 1MB directly and swap memory to the hard disk, a feature that Windows /386 didn't have - virtual memory.
Unlike its successor, Windows 3.0, which even looked like the current 16-Bit OS/2 of the time (v1.2/1.3)..

If IBM had sold OS/2 for a similar low price as Windows /386, it might have been more successful, maybe.
The high cost of OS/2 was in parts purely based on politics. IBM had to save face, so they couldn't make it dirt cheap.
If we consider the main customer base of IBM, then this makes sense. If it's too cheap, people think its not worth anything.

Edited. Edited. Edited. Edited.

Quick update. Found another cool 286 PC build at YT.
It has 5 MB RAM, Sound Blaster and a nice IBM PS/2 monitor.
It matches my 286 setup from the early-mid 90s a bit, even down to the games (that flight sim).
Especially the VGA monitor brings back memories, it's somewhat of a period-correct piece in my opinion..

https://www.youtube.com/watch?v=fVtkRpq2-cA

Quick Update. Here are some pictures of the little pizza box 286 I'm working on.
The PSU is repaired, more or less and the Gotek emulator got it's piezo speaker.
I'm planning on using this PC for experiments in the future. An OS/2 driver for the Trident 9000 is available, too.