Just one confusion I have: wasn't NT Win 4.0? The first NT release IIRC was 3.5, which had the same UI style as 3.0-3.2, but with the new kernel. This of course muddles the waters, having two parallel OSs with almost the same APIs but different kernel being sold at the same time with easy to confuse versioning (all the way up to Me and 2000 and until XP finally finished merging the lineages).
NT's first version released as 3.1, to align with the traditional Windows 3.1. Windows NT 3.5 (Code name 'Daytona') was a performance and stability improvement release. You can think of NT 3.1 as the "1.0" release and 3.5 as the "1.1" release.
Windows NT was natively 32-bit, but this was also the same time that the traditional Windows kernel was developing its own story for running 32-bit code. Windows/386 introduced the 386-specific 32-bit underpinnings and Windows 3.1 added something called Win32s. Win32s was an extension to the Windows executable loader and a set of 32-bit libraries that mapped a subset of Win32 calls to the underling Win16 API. Win32s made it possible to take the same 32-bit binary and run it unmodified on both Windows NT and on Windows 3.1. The caveat (and it was a big one) was that you only had access to a very restrictive subset of the Win32 API.
For Windows 95, Microsoft's strategy was mainly to build out more of Win32s. This built on the Windows 3.1/Win32s ability to run 32-bit binaries and added more of the API's that Win32 developers would've been used to from NT.
> The first NT release IIRC was 3.5, which had the same UI style as 3.0-3.2, but with the new kernel.
It's Windows NT 4 that unified the new UI from Windows 95 and the NT kernel. Even then, the system requirements and compatability issues were demanding enough that it took a few releases to make the NT kernel suitable for the mainstream.
Small correction if I may: Windows 3.1 did not add Win32S.
Win32S was an optional extra: a separate download and was never included with any version of Windows 3.x.
Win32S came along sometime later -- I think approximately in the time frame of Windows for Workgroups 3.11.
There were two purposes for Win32S: an official public one, and a secret one.
The official one was that it allowed developers to experiment with the new Win32 API, develop applications for Windows NT but which could be deployed onto Windows 3.1.
The secret reason behind Win32S was that it didn't work very well if Windows 3 was being hosted on top of a different 32-bit operating system. In other words, to be specific, it gave IBM a really hard time supporting Win32S in WinOS2 on top of 32-bit OS/2. So these new Win32S applications wouldn't run on top of IBM's OS/2 2.x. IBM kept finding ways to implement the additional Apis in Win32S, and Microsoft get changing them and putting out new point revisions of Win32S which broke those Apis again on OS/2.
Eventually the final version did this so effectively that as for as I know IBM never managed to get it entirely working.
By the way Win32S is still out there and you can still download it. There are not many reasons that you would want to do that, but there is one: it includes the original demonstration app for Win32S -- Microsoft Freecell.
This 30-year-old binary runs perfectly on Windows 10 64-bit and Windows 11. You can download Win32S, unpack the zip file, and just copy the files called `freecell.*` to wherever you want. Bingo: the original tiny Freecell Game, in place of the bloated advertisement filled one on the Windows store, for nothing. Enjoy.
> Small correction if I may: Windows 3.1 did not add Win32S.
>
> Win32S was an optional extra: a separate download and was never included with any version of Windows 3.x.
Andrew Schulman (in Unauthorized Windows 95) and Matt Pietrek (in Windows Internals) both describe the Windows 3.1 EXE loader as shipping with special case code for detecting PE executables. Win32s was an add on to Windows 3.1, but Windows 3.1 shipped out of the box with the necessary hook.
Whether or not that qualifies as Windows 3.1 _adding_ Win32s is a matter of packaging. (But it is why I used the wording I did.)
> IBM a really hard time supporting Win32S in WinOS2 on top of 32-bit OS/2.
Interesting, I never imagined Win32s working on OS/2 at all. I didn't use it much, but I always envisioned the Windows on OS/2 2.x as being closer to Standard mode (286) windows, since the Windows VMM probably didn't run on OS/2. But like I said... I didn't use it much.
Related to this: do you happen to remember Microsoft's earlier solution for 32-bit code on Windows? Predating Win32s, I remember there being a library that allowed allocation of 32-bit memory and jumping to 32-bit code. This code wasn't like Win32s code that had API access (it didn't), but it was a rudimentary way to access 32-bit offsets.
There's a bit of nuance in what I was saying.... I agree (and agreed) that Windows 3.1 didn't include Win32s. What it _did_ include, though, is direct support for Win32s in the form of PE executable detection and a hook for the (separately installable) Win32s PE loader.
Windows 4.0 was the original planned name for the product code-named Chicago, which was eventually released as Windows 95.
The first-ever version of Windows NT was called version 3.1. The public reason that it was called 3.1 is is to keep it parity with the then current version of 16-bit Windows which was at that time Windows version 3.1.
Industry rumour at the time was that the real reason it was called Version 3.1 was that Microsoft had a contract in place with Novell which gave them some confidential internal details of how the Novell NetWare IPX/SPX network protocol and the NetWare network client worked.
At the time that NT was in beta, Novell still did not have a client for the new operating system. Microsoft had to write its own client so that the new NT operating system would be able to connect to Novell NetWare servers over Novell's own network protocol. So the first several versions of Windows NT what bundle with a Microsoft written Novell network client, and subsequently a third-party client became available from Novell itself which implemented additional functionality such as support for the Novell NDS directory system.
The licensing agreement between Microsoft and Novell only covered up to and including Windows version 3.1. Therefore in order to use that information, that was the version number that Microsoft had to use for the new product.
I also add to that that you include another red herring that Microsoft's marketing department put out in your comment. The Windows 9X and Windows NT product lines were never merged.
Windows 2000 internally was Windows NT version 5.0. It changed to using plug and play device enumeration so that it could boot on p&p systems designed for Windows 9X. It was able to cope with not knowing the IRQ or DMA channels in use for its boot hard disk for example.
So what that means is that Windows 2000 was the version that incorporated support for the key technology that enabled Windows 9X but which could prevent NT 4 from running effectively on such machines.
Windows 2000 was meant to be the first consumer-focused version of the Windows NT product family... But Microsoft got cold feet, was afraid that it wasn't quite ready, and pushed out one more iteration of the 9X product line: Windows ME. Then it gave a bit of cosmetic facelift to Windows NT 5, added theming to the GUI, substantially increased the speed of boot up and suspend and resume, called it Windows NT 5.1, the marketing department branded it as "Windows XP", and the rest is history. But it doesn't actually include any technology from Windows 9X.
As I understand it some people from the DOS-based Windows product development team were moved over to the Windows NT development team, and the rest were laid off. Not many people were transferred and no technology whatsoever -- there never was a merger.
I don't think that qualifying XP as merely 2000 with a facelift is not entirely fair. XP benefited from having drivers produced for 2000 by the time it came out, but it also had additional compatibility shims to allow almost every Windows application in existence to work (in 32bits). 2000 was a great OS, that I liked, but it wasn't ready for consumers (not dissimilar to the Vista/7 situation). I believe that waiting until XP for the NT kernel big push was the right one, even if it gave us Me which was likely the worst release of the OS.
Just one confusion I have: wasn't NT Win 4.0? The first NT release IIRC was 3.5, which had the same UI style as 3.0-3.2, but with the new kernel. This of course muddles the waters, having two parallel OSs with almost the same APIs but different kernel being sold at the same time with easy to confuse versioning (all the way up to Me and 2000 and until XP finally finished merging the lineages).