29-Nov-2020 00:45 GMT.
UNDER CONSTRUCTION
Anonymous, there are 10 items in your selection
[Forum] Matt Dillon's Dragonfly pre-releaseANN.lu
Posted on 02-Jul-2004 09:17 GMT by Elwood10 comments
View flat
View list
Some of you probably remember Matt Dillon. He released DragonFly, an operating system and environment designed to be the logical continuation of the FreeBSD-4.x OS series. Read everything here
Matt Dillon's Dragonfly pre-release : Comment 1 of 10ANN.lu
Posted by Anonymous on 02-Jul-2004 09:35 GMT
Will there be a PPC version?
Matt Dillon's Dragonfly pre-release : Comment 2 of 10ANN.lu
Posted by Joe "Floid" Kanowitz on 02-Jul-2004 12:56 GMT
In reply to Comment 1 (Anonymous):
Will there be a PPC version?

Probably not for a bit. I shudder to go into all the reasons right now (just waking up), but let's say that the project is about 'not being stupid' on x86, first. Once's that's done, and the kernel architecture has been shaken down by real-world use, the time may come time to investigate how portable it could be... not unlike what happened to Linux, but to DBSD's advantage, its developers already have the awareness that things like endianness would then be an issue. ;)

Since G5s are arguably very 'x86-like' now (in the sense that both architectures have become postmodern, superscalar, out-of-order chips with 'large' on-die caches), and are coming mostly as SMP systems (with *roughly* similar issues overall, though not yet the NUMA DBSD might be targeting), I'm going to take a wild guess and say someone, somewhere will try to tackle it around 2.0.

---

If you're an experienced BSD user, or perhaps an experienced Debian user looking to try something new, 1.0 should be a lot of fun... Where 'experienced' means 'can survive at the command line long enough to set XFree86 up,' and 'fun' implies helping to uncover (and surviving through) any growing pains not caught in the Release Candidate.

If you're more comfortable with something like Mandrake or OS X, but have absorbed the enthusiasm and want to feel the snazziness at work on your x86, you might want to wait for Silver, which will be the first commercial DragonFly distro. [I have to admit, I have some doubts there... if only because, if coming soon, it'll be a 1.0 of a distribution of a 1.0 of an OS. But if it funds and motivates development of some of the missing miracles -- the packaging system, for instance, is not quite nailed down -- then it'll have done a service. Why yes, those are a lot of ifs.]


(Dreams for the packaging system here, if you'd like to know why it's been twice-patched FreeBSD Ports in the near term, and why it looks like Debian APT in the... medium-term. APT might even prove 'good enough' for the long haul, if people are still seriously considering it.)

---

Disclaimer: I'm still lacking the hardware (or at least, the disk space) to actually run the tree myself. Opinions are formulated based on what I've caught of dragonfly-kernel, and more recently, only what's filtered down to the DBSD blawg. It is, however, clear that this release (or as happened, the release of the Release Candidate) was timed for USENIX, where it's long-standing tradition to 'come bearing gifts.'* So, while it should, of course, be quite nice as it is (and probably has a 'better' first Release Candidate than some recent OSes have had -RELEASEs), remember also that grain of salt.


*As it's put normally, "have something to present." If you want to be crass, you could say the USENIX cycle may be the reason anyone dare claimed "release" status on FreeBSD 5. But as such, it's also what keeps these guys from holing up in their ivory towers and forces them to toss their work out for review. You could say the whole thing is a 'stability test' of how well an architecture holds up to deadline pressure, a very real-world scenario. [This is said as a FreeBSD user; however bumpy the ride, other OSes would've tried even more of my patience, and it sure has been.. educational.]

floid@mustelid:/usr/obj/usr/src/sys/MUSTELID i386, FreeBSD mustelid 5.2.1-RC2 FreeBSD 5.2.1-RC2 #1: Fri Feb 20 12:13:38 EST 2004, 10:48AM up 123 days, 6:37, 39 users, load averages: 0.51, 0.41, 0.41 signing off.
Matt Dillon's Dragonfly pre-release : Comment 3 of 10ANN.lu
Posted by Darth_X on 02-Jul-2004 14:29 GMT
I find it ironic Matt Dillon did a BSD remake instead of an AmigaOS remake.
Matt Dillon's Dragonfly pre-release : Comment 4 of 10ANN.lu
Posted by sutro on 02-Jul-2004 14:54 GMT
In reply to Comment 2 (Joe "Floid" Kanowitz):
> Since G5s are arguably very 'x86-like' now ...

I thought it was the opposite. Still we can't argue about the technical superiotity of PPC.
Matt Dillon's Dragonfly pre-release : Comment 5 of 10ANN.lu
Posted by Kolbjørn Barmen on 02-Jul-2004 15:08 GMT
In reply to Comment 4 (sutro):
We can't?
Matt Dillon's Dragonfly pre-release : Comment 6 of 10ANN.lu
Posted by Elwood on 02-Jul-2004 17:21 GMT
And Dmouse works on OS4 !! (well sort of, it has an unpredictable behaviour)
Matt Dillon's Dragonfly pre-release : Comment 7 of 10ANN.lu
Posted by Oliver B. Warzecha on 03-Jul-2004 07:29 GMT
In reply to Comment 3 (Darth_X):
> I find it ironic Matt Dillon did a BSD remake instead of an AmigaOS remake.

In fact it is more of an "BSD goes AmigaOS" kind of thing if you take a closer look at the planned specs. Could be interesting to tack AROS userland onto a matured DragonflyBSD kernel...

reentering lurk mode again,
OBW
Matt Dillon's Dragonfly pre-release : Comment 8 of 10ANN.lu
Posted by Anonymous on 04-Jul-2004 16:50 GMT
excuse me, this is VERY old. why are we mentioning DragonFlyBSD months after it was first rpereleased??

I've been running it on an x86 box (PII-500) for a few months now - this was a straight forward install from the CD image burnt. Previusly I ran it on another system - which was a side upgrade from a FreeBSD 4.8 system.

trivial to migrate, trvial to install from scratch. and a thorough joy to use!
Matt Dillon's Dragonfly pre-release : Comment 9 of 10ANN.lu
Posted by Joe "Floid" Kanowitz on 04-Jul-2004 22:39 GMT
In reply to Comment 8 (Anonymous):
excuse me, this is VERY old. why are we mentioning DragonFlyBSD months after it was first rpereleased??

Because this is 1.0 (or a RC thereof; actually, RC2 just came out), the first version the project will guarantee "normal humans" some stability with.

Again, since it was pretty smooth sailing (except for a few data-loss bugs, a few crashes, a few days when the tree might not build at all, and a few drastic architectural changes) from founding to 'release,' a long-time BSD user will ask "What's the big deal?" ... To which an Amiga or Windows or binary-only Linux user will respond, isolated from the realities of software development, "There are days when your operating system won't compile?!"

Think of the HEAD of the tree as an internal 'alpha.' "Release Candidates" (not unlike the OS4 Developer's Preview) are effectively 'betas' (though it's fair to say the DP isn't a beta, it's a... erm, "Preview," composed only of components that tested stable in beta... a conservative practice when Apple was selling untested code), in the sense that the developers trust them not to set anyone's printers on fire. In DragonFly CVS, the development process on that 'tag' is then frozen except for fixes to issues uncovered in the candidates. New work can continue to pile up on the HEAD of the tree, as this article might explain.

(The tag-vs.-branch thing is pretty esoteric, and doing it this way is just a convenience that fits with the development model... DragonFly won't be 'forking' itself with each release, so fixes can be 'captured' in a new tag -- the WINDOWS_98 tag might equal the WINDOWS_95 tag but for a new version of EXPLORER.EXE ;) -- without doing explicit copies and so on. FreeBSD, on the other hand, rarely expects new commits to work cleanly with old versions, so they branch to keep things clean as files are deleted or moved around; RELENG_5 doesn't need to preserve memory that some file once existed in 2.x, while it'll still be archived in the RELENG_2 branch if you need it. None of this *directly* relates to the need for 'code freezes,' either; it's only that DragonFly is smaller and more focused, so it's easy for Matt to know what's going on in his HEAD... and not wake up one morning to discover he's grown three complete threading libraries* and a rash of regressions.)

Anyway, to get the real truth of the matter, you can read this as "DragonFly has gone into what a commercial project would call 'final beta.' You can expect 1.0 by the end of the month, which the developers, for calling it '1.0,' will then consent to try to give you support on if you use it in a 'production environment.'"

In DragonFly's case, 1.0 should also have been the first 'snapshot' of the API and ABI, as plans for backwards-compatibility are concerned; however, same seems to have been pushed off to 2.0... Making this, again, sort of the 'special case, get people to run it' release, but certainly no worse than FreeBSD for it.

---

*Yes, some guys would kill to have three, but do any of them stay up when working with GIMPs or giant lizards?
Matt Dillon's Dragonfly pre-release : Comment 10 of 10ANN.lu
Posted by hammer on 04-Jul-2004 22:50 GMT
In reply to Comment 4 (sutro):
With PPC 970, PPC ISA instructions gets decoded into FRISC...
Anonymous, there are 10 items in your selection
Back to Top