IA64 bitkeeper trees (again)
After I posted my description of how I'm using Bitkeeper trees to manage the ia64 part of Linux I got some feedback. Len Brown kindly talked me through how he runs a 2-phase/2-stage set of trees for ACPI ... and after a while it began to make sense (the layout of the trees, not ACPI :-). So I've set up a similar set of four trees:
+--------------+ +-----------------+
| test-2.6.9 | | release-2.6.9 |
+--------------+ +-----------------+
^ ^
| |
+--------------+ +-----------------+
| test-2.6.8.1 | -> | release-2.6.8.1 |
+--------------+ +-----------------+Work begins at the bottom left, which is a clone of Linus's tree when he applied the v2.6.8.1 label. I apply patches there, and do whatever compile/boot tests I need to make me feel comfortable about the patches. I can then push/pull them up to the test-2.6.9 tree which is based on Linus's latest (by pulling each day) and re-compile there to make sure that the changes don't conflict with other changes to the base. Andrew Morton can then pull from the upper left tree to feed into "-mm" expose the changes to a wider test audience.
Once the changes are suitably aged, I'll pull them over to the "release" trees, which are the staging area for Linus to pull from.
The actual names all have "linux-ia64-" prepended. E.g.
http://lia64.bkbits.net/linux-ia64-test-2.6.9
The names of the trees all change when Linus makes a release.
Len also gave me a script to create plain patches for non-BK users ... I'll make the tweaks to the pathnames and get those running soon.
-Tony
Quoted 5 lines · Tony Luck
The actual names all have "linux-ia64-" prepended. E.g.
http://lia64.bkbits.net/linux-ia64-test-2.6.9
The names of the trees all change when Linus makes a release.
I don't know how or if it's even possible, but is there some way to make a linux-ia64-test-latest symbolic link (or something) that you can pull from to always have the latest testing tree?
The initial reason I wanted this was because we clone the tree to a local server and then do local clones from that to save bandwidth, and I'd have to write a script to automatically figure out when the tree changed over and re-clone it every time. But I think it would be just generally handy.
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Tony Luck
Tony> The names of the trees all change when Linus makes a release.
I don't see the point of this. It makes it painful for me to stay current with the latest ia64 tree. Why not just use tags to mark releases?
--david
Quoted 9 lines · Tony Luck
The actual names all have "linux-ia64-" prepended. E.g.
http://lia64.bkbits.net/linux-ia64-test-2.6.9The names of the trees all change when Linus makes a release.
Len also gave me a script to create plain patches for non-BK users ... I'll make the tweaks to the pathnames and get those running soon.
Maybe I'm just used to the old method, but doesn't this make it harder to just do a 'pull' on an existing tree to merge one's changes up to the latest code? It also seems like it makes it more confusing if you ask Linus to pull multiple times in a release cycle. I liked your first message about trees better. :) In particular, this part:
Quoted 10 lines
I've set up two bitkeeper trees too:
http://lia64.bkbits.net/to-base-2.6is my holding area for patches that I want Linus to pull.
http://lia64.bkbits.net/linux-ia64-2.6will be a place for me to stash changesets that I'm not ready to push (or for any non-ia64 specific changes that I want to play with). At the moment there is nothing in this tree that isn't also queued in the to-base-2.6 tree.
Quoted 4 lines
Summary: For 99% of uses, you can clone a tree from Linus and use it on ia64. If you are sending a sequence of related patches and know that I've taken some of them, then either of my trees should work for you.
But maybe I'm missing the advantages of the scheme Len is using?
Thanks,
Jesse
Quoted 10 lines · Ian Wienand
Quoted 5 lines · Tony Luck
The actual names all have "linux-ia64-" prepended. E.g.
http://lia64.bkbits.net/linux-ia64-test-2.6.9
The names of the trees all change when Linus makes a release.
I don't know how or if it's even possible, but is there some way to make a linux-ia64-test-latest symbolic link (or something) that you can pull from to always have the latest testing tree?
It wasn't possible last week ... but "ask and ye shall receive". Larry McVoy added a symlink feature to the bkbits.net admin interface for us (Thanks Larry).
So what would you like the names of the symlinks to be? I'm thinking of:
linux-ia64-release-2.6.8.1 -> release-2.6
linux-ia64-test-2.6.8.1 -> test-2.6 linux-ia64-release-2.6.9 -> release-2.6.next linux-ia64-test-2.6.9 -> test-2.6.next
but I'd be happy to entertain better suggstions.
-Tony
Quoted 9 lines · Tony
So what would you like the names of the symlinks to be? I'm thinking of:
linux-ia64-release-2.6.8.1 -> release-2.6 linux-ia64-test-2.6.8.1 -> test-2.6 linux-ia64-release-2.6.9 -> release-2.6.next linux-ia64-test-2.6.9 -> test-2.6.nextbut I'd be happy to entertain better suggstions.
Just an idea (and I may have misunderstood the process); why not name the trees for the people they are going to, for example
test-2.6 -> linux-ia64-tony (the main IA64 development area)
test-2.6.next -> linux-ia64-to-mm (patches waiting to go to -mm series)
release-2.6.next -> linux-ia64-to-linus (patches waiting to go to official kernel)
I'm still not sure where the other release tree fits in all of this though.
Under this my thinking was a developer sends (non-trivial) patch that first goes into -tony where it makes the core ia64 developers happy, gets moved into -mm where it is picked up and used by a wider audience, eventually it gets put into -linus where it is moved into the main kernel. Smaller patches, like a compile fix or whatever, go straight from -tony to -linus?
-i