the plan for stable

[Eric W. Biederman]

Ronald G Minnich <rminnich at lanl.gov> writes:

> A short sketch.
> 
> What I want to do, but have not had time to do. For each mainboard, we
> designate an owner. The owner is responsible for letting us know that
> their mainboard works. Mainboards are in one of 3 states: (stable,
> unstable, unsupported)
> 
> Mainboards start out in the unstable or unsupported state.
> 
> We pick a date (1/1/03?) and say we want all owners of all mainboards to
> tell us that their mainboard is stable. We freeze the tree one month
> ahead of that time and the only changes that go in are for stabilization.

1 January 2003 is a bad date for me as I have plans to be far
away from computers over christmas.

> If nobody steps up for a board, it goes to unsupported state. Boards with
> owners start out in the unstable state.

The challenge is for a lot of boards we do not get active feedback
after a port has been completed.  So for any ongoing work we need
to very very careful not to make changes to the core that break ports.
I am probably the worst offender, except for the various bits of debug
code that come and go but still.
 
> Mainboards move to the stable state when the owner confirms stability.
> When patches are made for a problem, ALL stable mainboards revert to
> unstable. We iterate until we get it solid, then freeze it.

For the first round this looks o.k, it really depends on what
kind of feedback we have. 
 
> This information is maintained by a file in each mainboard directory
> called STATUS, which consists of name/value paris.

One file in the root directory called STATUS should do it..
 
> Will this work?

Sounds like a good rough draft.  The very important thing
about the stable series is that nothing happens to the core
code that could possibly break a motherboard port.  That
way within a stable series we can get more but not fewer
boards working.

Then whenever a new port gets working we can do another release.
Of course if the come in fast enough we can delay...

> thanks (I'm off email for a bit -- at a workshop)

Speaking of which we probably should put together
some kind of conference/workshop for LinuxBIOS.  So the developers can
get together and talk face to face. 

Eric