Another svn-git oddity

[Ed Maste]

On Wed, 4 Mar 2020 at 03:33, Ed Maste <emaste at freebsd.org> wrote:
>
> On Fri, 22 Nov 2019 at 08:52, Ed Maste <emaste at freebsd.org> wrote:
> >
> > As mentioned before there are some oddities in our git history - svn
> > commits translated to git in a slightly strange way. These have been
> > addressed and if/when we run the conversion from the beginning they'll
> > disappear.
> >
> > I've just found another oddity that affected attempts to use git
> > subtree for prototyping a method of handling vendor code in our tree.
> > I'm not sure yet if the changes to the converter would address this
> > case or not.
> >
> > The affected commits:
> >   svn      git                   Headline
> > r300828 00d578928e Create branch for bhyve graphics import.
> > r300829 b1cfc58d46 Initial bhyve native graphics support.
> > r302332 6775e12557 Import bhyve_graphics into CURRENT. Thanks to all who
>
> These commits still exist in uqs's latest prototype conversion, with hashes:
> 953b4a972bec
> 8efdfe5ee821
> 6d7144acee3e

r300828 branched (only) usr.sbin/bhyve/ to the top level directory,
which is represented in the git conversion as deleting everything
else. It's this that is the root cause of the git subtree trouble, I
believe. The actual changes come in r300829, merged in r302332.

Uli, do you think we could just omit r300828 (and r300829) from the
svn2git conversion? This is a slight revision to the actual history,
but I think no information is really lost -- the merge in r302332
includes the full commit message of the only change that was made on
the branch.

I do not believe there are other branches made in a similar way to
r300828 (copying just a single subdirectory) but am not certain and am
trying to verify that now.