Why re-invent the wheel? Far more economical - in time and resources - for them to have used the shell of what's already out there.

Of course, these days, using the Windows Preinstallation Environment means it's that much easier to shoehorn in some adverts and kick off the data-grab nice and early. Y'know, start as you mean to go on and all that ;-)

It explained it all already…

…and it was a perfectly cromulent explanation.

Including the bits about starting a DOS graphical installer from within win 3.1 if one upgraded in place - even if usually not a good idea with windows way back then, but hey :-)

Windows 95 was competing with OS/2 at the time.

OS/2's claim to fame was that it was a new operating system, unlike Windows, which was just a graphic user interface that sat on top of DOS.\

Microsoft countered that Windows 95 was a standalone operating system that, unlike Windows 3.1, did not require DOS to run.

The fact was that Windows 95 did did run on top of a DOS boot loader and kernel. However, unlike Windows 3.1, it was not sold separately from DOS. Windows 95 was sold as a packaged that included the DOS boot loader and the GUI.

So while at a technical level, Windows 95 required a DOS kernel to boot, at a marketing level, it was an integrated package, so MS could claim that it did not require DOS to run, since users did not need to buy it separately.

After going through those conniptions to convince the world that Windows 95 was not just a "clown suit for DOS" (as the joke went), having it boot to DOS to install would contradict the messaging. So, it used WIndows 3.1 instead.

At least it didn't require several floppies to boot the installer, as you had to do with OS/2 Warp, even if it came on a CD...

Windows NT 4.0 (released a year after Win95 and far superior in every respect) needed 4 boot floppies, before it would read from the CD.

"Windows NT 4.0 (released a year after Win95 and far superior in every respect) needed 4 boot floppies, before it would read from the CD."

Windows NT 4.0 CD was El Torito compatible from the beginning - you can boot from it and install it from there.

If however your BIOS didn't support booting from CD drive, then you needed the boot disks - all three of them. They included the IDE ATAPI driver and few usual Adaptec drivers a well, but if your system had some unorthodox SCSI / sound card / parallel port CDROM interfaces for which the boot disks didn't carry drivers, then yes - you needed a separate driver diskette for that one as well.

I seem to remember being able to boot OS/2 from CD. But that was only after they were using IDE or SCSI connections, until then you were dependent upon hardware-specific drivers and they needed space. Same was true for lots of audio and video devices at the time: Windows 95 was mainly a graphical refresh of Windows 3.11 and, as such, not much of an OS, but it was good enough for many and manufacturers found it useful to have only one OS to program for.

> : Windows 95 was mainly a graphical refresh of Windows 3.11 and, as such, not much of an OS,

Well, apart from being a 32-bit pre-emptively scheduled kernel, instead of 16-bit co-operative.

But aside from the graphical refresh, the extra 16 bits for consistent flat memory addressing and the pre-emptive scheduler, what did '95 ever do for us?

Following that logic, Novell netware (3.11 and 4.1 in particular) ran atop of DOS, because they needed DOS to boot, and you could exit from them to DOS.

The reality was that Win1.o depended heavily on MS-DOS. Win 2.0 took some of the things DOS did, and started doing those itself in 286 (or 386) protected mode. Win3.x Took even more things away from MS-DOS, ditto for Win95(SE), Win98(SE) and WinMe. So, each iteration sheded more and more of MS-DOS

With the advent of WinXP, MS-DOS was completely exorcised from the consumer OS.

Whilst that was one of OS/2's marketing points, it really wasn't the major one. Having a more advanced interface, multithreading, a much better file system, the ability to multitask DOS and 16bit Windows programs next to new OS/2 apps without affecting stability was the draw. Plus it was 32 bit.

By the time '95 arrived it was good enough for most people. Long filename support hacked on, FAT32 a couple of revisions later. Multitasking that was better than Windows 3.1. DOS compatibility that was also an improvement. A vastly improved API and interface. We can argue just how much better OS/2 or 9x were, but one thing you also have to give Windows is introducing plug and play - which had some support in OS/2 but frankly was much better implemented in 9x (again you could argue about whether it's better to have present but rather flaky PnP, over largely absent PnP but a stable system it was necessary to configure yourself. It was only with Windows 2000 that PnP stopped being horrific).

OS/2's fate was sealed even before 2.0 was released, but even so it was a great OS at least until 94-95. The release of 95 killed it off (not to mention all the resource wasted on OS/2 PowerPC), and by the time Warp 4 was released in 1996 it was very clear it was at the end of the road. Windows NT 4.0 was released that year, featuring the 95 interface and a load of new features. Whilst its lack of USB support and later versions of DirectX would really start to bite in the late 90s, a lot of the OS/2 vendors started developing NT releases of their software, and the userbase switched mostly to that, with a few Unix transfers.

Whilst that was one of OS/2's marketing points, it really wasn't the major one.

It depends on where you were. I was working at IBM at the time on an OS/2 1.x product (part of AD/Cycle, if you remember that), and at all the trade shows, the IBM presenters pushed the "you don't need DOS" and "it doesn't sit on top of DOS" lines constantly .

OS/2 was 90% of the way there to being better than Windows for a huge number of users. What was really annoying was that instead of putting in the effort to address that 10% (the SIQ in particular), IBM just blamed the press, the retailers, and the end users for not appreciating the fact that it was better. They finally fixed the SIQ in Merlin in 1996, four years later, but if they'd done that, and addressed a number of other issues with the WPS in 1992, it would have lasted a lot longer.

OS/2's biggest problem wasn't competition from Microsoft, it was (mis)management within IBM. Some of the internal communications I saw were just jaw dropping in their delusions.

I didn't really start switching away from OS/2 until late in nt4 lifetime and fully with windows 2000. I had a dual boot for games for a few more years though.

Last time I tried OS/2 I couldn't even hardly remember how to use the things.

Debunked several times over, Windows 95 wasn't "a GUI that sat on top of DOS"

https://forums.theregister.com/forum/all/2024/11/19/chen_windows_95_setup/#c_4968643

Linux at the time needed LILO to boot.

Apart from the fact that DOS was maybe a little more capable than LILO and Linux (+GNU+XFree86 etc) far more capable than Windows, was there really any difference?

These days, installing Windows is a convoluted mess. Out of interest, I did a clean install on a spare Desktop today. It took nearly an hour. Then, of course, you have to de-bloat all the ads and other unnecessary crap out of it before you can think to load programs and actually use it. I then clean installed Linux Mint 21.1 Cinnamon. That took about 20 minutes, including running 2 scripts to run uppdates and download all the software I use. No competition in my book. Never going back to ANY Microsoft products.

I recall installing SunOS 4 using exactly the same philosophy. Boot off CD, load mini root into memory, and run that to install to disk. Plus ça change …..

You could boot and install SunOS from a tape drive. On a SCSI controller it had no driver for, on to a disk on that same controller. OpenPROM and Open Firmware were designed from the outset to be extensible. No driver for that device? Well the firmware driver for it is written in Forth, we'll just carry on rolling with that in our own hosted implementation when the main OS goes up, the user can install a native driver later.

Where is the boundary of the operating system exactly, especially since the very same firmware was also the boot manager and kernel loader? The endless debates over when Windows became an operating system (as above) are utterly meaningless when considered against all the other alternatives.

I think some of these thousands of nearly identical Linux distros could pick up a trick or two here.

But then I realize I use a distro nearly identical to Debian, except for rejecting the heresy of the People's Republic of Systemd as espoused by that vile heathen Poettering.

I notice Windows 8, 10 and early 11 use Windows 7 during the installer (you can tell because of the Windows Basic theme). I guess its a similar reason.

IOW bootstrapping all the installs of all these systems listed above uses a bootstrapping approach.

Given Windows3.1 dialog boxes survived until Windows 10 (possibly 11?), it's unsurprising:

https://www.reddit.com/r/Windows10/comments/l85hsx/windows_31_just_called_they_want_their_dialog_back/