svn commit: r331461 - in user/markj/netdump/sys: kern netinet/netdump sys vm

[John Baldwin]

On Monday, March 26, 2018 12:26:25 PM Ian Lepore wrote:
> On Mon, 2018-03-26 at 10:12 -0700, John Baldwin wrote:
> > On Saturday, March 24, 2018 08:40:24 AM Rodney W. Grimes wrote:
> > > 
> > > > 
> > > > On Sat, Mar 24, 2018 at 02:17:02PM +0200, Andriy Gapon wrote:
> > > > > 
> > > > > On 24/03/2018 04:46, Rodney W. Grimes wrote:
> > > > > > 
> > > > > > I know this is on a private branch, but when/if it
> > > > > > is merged this becomes part of the main line.
> > > > > Not with svn, I think.
> > > > > At least, the way we use it.
> > > > Indeed, I have no intention to merge the branch directly. I'm using an
> > > > svn branch so that it's marginally easier for others to test.
> > > None the less as stated in:
> > > 	https://svnweb.freebsd.org/base/projects/GUIDELINES.txt?view=markup
> > > 
> > > 12 	General guidelines:
> > > 13 	
> > > 14 	* Should be relevant to FreeBSD.
> > > 15 	* Should be at least conceivably of interest to somebody else.
> > > 16 	* Should be in a format that is suitable to merge into the base tree.
> > > 17 	* Should be something that is worth people's time to read commit mail for.
> > > 18 	* Write decent commit messages!
> > > 
> > > Thanks,
> > We generally don't do that for user, etc. branches.  Merging from a
> > projects/user branch into head in svn is often a disaster due to svn's
> > limitations, so normally a projects/user branch is treated as a work area
> > and the resulting diff is then hand-applied to head with a suitable commit
> > message that describes the entire change.  This is similar to using something
> > like 'git rebase' to rewrite history and compress a long tail of changes
> > down to a small number of commits prior to merging to head.
> > 
> > You generally don't see these work branches in svn as most developers do them
> > outside of svn in git, p4, hg, etc. due to svn's limitations.
> > 
> > For things that live permanently in user/projects (e.g. the code for core
> > elections or the patches for freebsd-update), we do want standard commit
> > messages.  However, I don't think we want to impose that on WIP branches
> > that are later compressed down before merging.
> > 
> 
> The support for armv6 happened this way... lots of out-of-tree and
> project-branch work, with nothing much useful in the way of commit
> messages, followed by a single massive import with a commit summary of
> something like "import armv6 support".
> 
> Now when you try to search svn history for how something came to be,
> all you can find out is that it was part of the incomprehensibly-huge
> initial commit.  You can't even figure out who to ask about something,
> let alone why something was done.

That isn't fixed by having detailed commits in a WIP branch for compile
fixes.  To me that particular issue should be addressed by doing a better job
importing things into HEAD including pulling out pieces of long-running
branches and merging them as multiple commits each with useful commit
messages.  We should not rely on being able to find the history behind
commits to HEAD because that history (if it exists) lives in random places.

-- 
John Baldwin