During the last year the D foundation was founded and is spinning up operations. The 2016 D conference had a visitor record again with over 100 people. Downloads are growing over the years [0]. The package manager dub is now fully integrated. You can find success stories on the blog [1] and companies using it [2]. There are books.
[0] http://erdani.com/d/downloads.daily.png [1] https://dlang.org/blog/ [2] https://dlang.org/orgs-using-d.html
The reference compiler backend is still only free as in beer, so there seems to be no way to include "DMD" in Debian. However, LDC (reference frontend + LLVM backend) is practically in sync with DMD today.
The advantage of DMD is faster compilation. LDC gives you faster programs. So, for development I prefer DMD for the faster iteration.
The backend comes from the Digital Mars C++ compiler, which is/was commercial. That is probably the source of the confusion.
Now that developers are ready for new languages, D is already an "old, failed" language.
It reminds me of selling a house. If you put your house during a hot market and go into contract right away, but then your buyer can't come up with the money and it falls out of contract three weeks later, bringing your house back on the market it's now no longer "newly available!", and everyone thinks that, because it didn't sell in the first week, there must be something wrong with it.
My personal entirely uninformed opinion of D is exactly that: It's an older, "failed" language. It's not on the upswing in popularity. It didn't catch peoples' hearts and minds. And if only because of network effects, therefore it's not worth learning, simply because no one else is learning it, and so therefore its ecosystem isn't expanding exponentially.
Success of technologies is far less merit than one might expect. See: VHS vs. Beta.