It lent itself to being tinkered and tailored in a much more developer-friendly way than most OSes. The rexx scripting was pervasive. There were meaningful file object attributes that could be extended and used per app. The SOM model was a bit clunky but very powerful (I went on to work with its designer later in my career). It had a bunch of powerful compilers that left all the Windows competitors for dust.
And of course it looked great for a windowed GUI at the time.
To clarify my statement above about its code being a hot mess: my understanding was always that OS/2 achieved reliability through sheer force of will rather than it being built in a way that was architected to be intrinsically safe. Sort of how US nuclear subs have achieved an impressive safety record mainly due to extraordinary fastidiousness with respect to process and procedure.
I think both of these, combined with the fact that OS/2 is not a moving target anymore (unlike Windows), can make it much easier to make an open source OS/2 replacement by replacing its systems bit by bit. If anything, if you check eComStation (and i also guess Arca OS), it has a lot of newer parts than what was originally in OS/2 at all stages of the "stack" - from the bootloader, to graphics drivers, to UI, many of which seem to be open source.
TBH i think the main reason there haven't been already a working OS/2 replacement is that there doesn't seem to be enough of a technical know-how in the current community and there doesn't seem to be any new interest for it by people.
Haiku was pretty much in the same position, but the difference that somehow Haiku managed to get enough knowledgeable people on board despite BeOS being released on PC only a couple of years after OS/2 Warp. I guess the release of BeOS Personal for free helped spread it much further and OS/2 never had something like that.
Based on other free software packages, e.g. the interesting Fiasko (from the L4 family) micro kernel, one could rather see it as an example application of those.
Apart from “hacks and glory” (which for the record is a fine justification), I don’t imagine that this project would have a lot of utility. While some old ATMs and some other legacy applications run on OS/2, I’d guess that attaining the degree of compatibility needed to satisfy those applications would be an extremely tall mountain to climb.
I think there are a lot of us who remember computing of the 1990s and early 2000s fondly as being a time where you could discover an operating system that was new or novel in some way. Combine that with the fact that computers felt less laggy, and mostly used the same UI across applications (looking at you electron), and I share the nostalgia.
I guess what I’m saying is that an OSS OS/2 is cool, and its authors should have songs sung about them, but this isn’t the right foundation to build a modern OS on.