On Thu, 14 Sep 2017, Theodore Ts'o wrote:
On Thu, Sep 14, 2017 at 12:15:32PM -0400, Steve
Nickolas wrote:
Maybe
some are deluded like that. But the more typical case (and I saw
this personally not just on "the Internet") are those who actively and
consciously disdain Unix, and want Penguin kernel based systems to move
to a completely new and different userland, free from any links to Unix
history.
And we should stop assuming they're kidding when they say so openly.
Isn't that pretty much just Lennart Poettering and his fan club?
It's mostly Lennart Poettering and his fan club, but it's also
important to remember that Unix was not perfect.
For years, I've been ranting about the telldir/seekdir interface, for
which JFS has three b-trees that have to be updated for every
directory operation --- one of which was added *just* because of
telldir/seekdir. Other file systems make other design choices or go
through other bits of hell just because of telldir/seekdir, but
assuming a 32-bit cookie which must survive potentially indefinitely,
with the readdir "will return file names exactly zero or one times"
guarantees required by POSIX, is rather hellish.
Or, say the atime update requirement, which can be a performance
killer, and for which the default on Linux violates Posix, and so I
suppose it technically isn't allowed to use the Unix trademark anyway.
I'm sure the Plan 9 folks could come up with other Unix shortcomings. :-)
- Ted
I never managed to pull it off, but I tried creating a full live Linux
environment based on musl, clang, Heirloom Toolchest and OpenBSD/NetBSD
sources. The idea was that I wanted to make a "Real Unix" that happened
to have Linux as its kernel. (It also would have run the CDE as its
default desktop.)
One thing I did come up with was, if I were to pull it off, it would be a
Linux distribution that rightfully could NOT be called, by any means,
"GNU/Linux" - and some heads would explode.
(I still want to do it, but I remain at a loss as to execution.)
-uso.