NetBSD with git so far
Low memory hosts:
git appears to have slightly different memory characteristics depending on the protocol used. Over http I am able to get a full clone with all history on a 256 + 256 raspberri pi. If you bump up the memory to 512 + 256 it makes ssh possible, which means writes are possible.
The link above has some tuning I used to get memory requirements way down.
It should be noted that git support a "shallow" clone (--depth 1) which ignores most history but allows commits and full development. A shallow clone works on very small systems; I would guess 128MB + a little swap is enough.
git is slow during 'status' by default since it searches the entire tree for a change. It will produce a warning with tunable options if the command runs slowly.
After some complaining on the git@ mailing list a patch has been produced which drops the memory requirements down quite a bit. I can now, without much tuning, work on my 512 system. I'm pretty sure a 256 + swap without any special tuning would also work.
CVS in parallel
I do not think this is a good idea and do not plan to advocate for it. Git does have a cvs server built-in but I have not taken the time to set it up for testing because it is slightly involved and I don't see the purpose.
One-shot to create the new True Source. I don't think there will be many cvs hold-outs.
See above for CVS server provided if ongoing conversion is really desired.
existing cvs dependencies
How should NetBSD be setup
High level- private box for write master using ssh, any number of additional systems with read-only mirrors over http:// and git://
how to install
git should fit into NetBSD src/tools easily. I have not personally tested cross compilation.
See DragonflyBSD examples given above
There are many many workflows supported in git. For the most part I think NetBSD developers would follow the "feature branch" workflow from the main repo (or private/semi-private clones before merge).
Public collaboration is a big feature of git since it can format patches into at least two different email formats and they can be submitted to a bug report or to a mailing list, which should allow clean apply.
A non-developer could also post a pull request to github or host his git repo for a friendly developer to add as an origin and pull his branch.
(git origin add future-developer http://example.com/~greatguy/src.git)
log message formats
Try to references named branches/tags instead of sha-1's Also using the dates for commits instead of commit id's
how to convert
I am unable to anticipate the next generation of SCM. Don't do anything weird like change history and we should be fine.
Maybe when we have 30 years of project history it will be time to consider restructuring the project.
I think this is less a function of the tool and more a function of the project not allowing non-"standard" actions.
Who, When, and How Long?
- ESR/IIJ/Joerg - convert
- sometime, eventually, maybe
Assuming conversion starting from date(x) to freeze(y) is relatively easy, the refinements of Joerg/ESR conversion can continue to run in read-only mode as they do today. This means the "switch" is a few hours only for:
- cvs goes read only
- history from last git conversion pull until now is appended
- cvs is turned off
- git is made available over ssh