Contributing to the kernel video drivers

[Gleb Popov]

My 2 cents about commit-by-commit approach. Because patch composition is an
associative operation (read: because we have rebase functionality), this
workflow can actually be parallelized.

Let LB be the last ported commit from Linux repo and FB the corresponding
commit in FreeBSD repo. The dev picks whatever child commit LCn of LB he is
interested in, ports it and places it (lets name it FCn) somewhere in
FreeBSD repo (let it be an individual branch). Until all LC(n-1) commits
between LB and LCn are also ported (read turn into FC(n-1), FCn remains in
this "detached" state, and when all FC(n-1) are in place we rebase this
chain on top of FB, advancing FB tag to the FCn.

This way we can still have 1:1 history correspondence between Linux and
FreeBSD changes as well as distribute work between developers. I even
imagine some script to keep track of "detached" commits and automatically
rebase them.

What do you think?