gnusol-devel  

Re: [osol-discuss] GNU libc on OpenSolaris

Alex Ross
Fri, 19 Sep 2008 15:49:52 -0700

Back in '05 (years fly) we actually started from trying to port glibc and a bunch of core libraries outside glibc. We then relatively quickly, having spent about 3 weeks on it, realized the size and complexity of the exercise. At this time, as well as at any other time since then, we were extremely pressed in time, and so we then decided to use Sun libc. Needless to say, the latter worked. The sky's the limit, obviously, but still, there are two always valid perspectives to look at it: pure practical end-user-value perspective (what would end user get from this C library versus that other one..), and some kind of a technical estimate of the effort. Both perspectives pose their corresponding questions.

I'd also worry about the focus. There was Michael's email "A review of the base system" a few days ago, with an excellent (detailed, exact) summary of what needs to be done. What needs to be done amounts to quite a bit.

The argument that "OpenSolaris-specific extensions are trivial wrappers around syscalls" works both ways: in a great majority of cases porting a GNU application on top of Sun libc is a trivial exercise, and there's now certainly ample evidence for that. In fact, one thing that would make a lot of sense to do is to (finally!) come up with a header file containing all those wrappers, make it standard for the port, and make sure that this header file is always included at compile time.

--
Alex

Michael Casadevall wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I don't have a problem with two separate ports. Like for people who
want Solaris based system for stability and ZFS, and a solaris based
one. A nice and practical upshot of this is the possibility of a
kopensolaris-amd64 port which has been a bit of an issue with the
current ON based system. The only question is if we ever became an
offical Ubuntu port, which one would/should be accepted upstream. If
we're legitimentally going to set up a second port, then I'll install
dak (not mini-dak), and configure it for this adventure (mini-dak is
great for single ports, not so much on multiple ones in my
experience).

As a second benefit, its likely the base system will not require the
same amount of work to get buildds working, so you can probably
leverage the existing Debian autobuilder system, and get hardy built
much faster than we can since we need to work on improving the ON
base.
Michael

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: http://getfiregpg.org

iEYEARECAAYFAkjUHccACgkQpblTBJ2i2psiEwCfRh5CVJD4XE9VxQ70f8xrz7Az
lBwAnRdGRj2RhRskq4ElXCIIqdgAlARi
=esqL
-----END PGP SIGNATURE-----

On Fri, Sep 19, 2008 at 5:08 PM, Per Lundberg <[EMAIL PROTECTED]> wrote:
It is not Solaris, but it is GNU/kOpenSolaris. :-)

If I might state my opinion, I believe diversity is a strength and choice is
a good thing. If some people want to go for Solaris libc, let them do so;
likewise for those who prefer an even more GNU-styled userland (with GNU
libc being the cornerstone). What we should note though is that Nexenta
(CP2) is already much more GNU-like than Solaris has ever been, in my
experience... Which is one the reasons I like Nexenta more than I've ever
come to like Solaris, after working with a Solaris-based Perl web
application for around one year. Of course, my background is much more
GNU/Linux-based so I'm biased...

Anyway, I think GNU Solaris should be able to "umbrella" both of these two
"branches". They can probably not be combined in the same "distribution",
because we are talking about such core pieces of the system that it would be
weird having GNU libc installed when packages have been compiled against
Sun's libc, and vice versa. Of course, we could have double packages
available for each and every package - one compiled against GNU libc and one
compiled against Sun libc. But that would really be a kind of weird
operating system... It is much better (IMO) to let the branches be branches.
They can share the same infrastructure; both of them can have their
autobuilders (when they are ready) hosted on the same machine, but in
different zones. And so forth.

For the time being though, it might be best to hold some kind of
"referendum" among the core developers (which I am not a part of myself) of
GNU Solaris as to which of these branches that should be emphasized. It's
not like we have 1000 developers just sitting around and waiting for more
work to be done, so a bit of focus (with the clear allowance of letting
people with different opinions "do their own thing", within the same
infrastructure) might be a good thing. Does this sound good?

On Fri, Sep 19, 2008 at 11:29 PM, Michael Casadevall
<[EMAIL PROTECTED]> wrote:
This poses an interesting question then. With this, we could, in
theory dump the ON userland, and go pure GNU, more inline with the
other Debian/Ubuntu ports. That being said, I still feel diversity is
a strength, and is it still Solaris if we dump the userland (and with
it, binary and script compability?)
Michael

--
Best regards,
Per Lundberg

_______________________________________________
gnusol-devel mailing list
gnusol-devel@lists.sonic.net
http://lists.sonic.net/mailman/listinfo/gnusol-devel


_______________________________________________
gnusol-devel mailing list
gnusol-devel@lists.sonic.net
http://lists.sonic.net/mailman/listinfo/gnusol-devel