Re: Subtree in Git

[Herman van Rink <rink@initfour.nl>]

On 05/05/2012 06:25 AM, Junio C Hamano wrote:
> greened@obbligato.org writes:
>
>>> I basically did a: git subtree merge --prefix=contrib/subtree <my
>>> git-subtree branch>
>>>
>>> The work in progress in on: https://github.com/helmo/git (the
>>> subtree-updates branch)
>> This branch seems to have a bunch of commits from master or some other
>> branch:
> Isn't the confusing shape of the history a direct result of what Herman
> said he did above, i.e. use of "subtree merge"?  I thought that we agreed
> not to do any more subtree merges for further updates when we slurped the
> subtree history to contrib/ early in this cycle, so if that is the case,
> Herman needs to rebase his work so that the integration will not need any
> "subtree merge" into git.git, perhaps?
>
> I looked at various branches found with ls-remote in that repository but I
> couldn't quite tell which is what, with too many cross merges, among which
> there are unnecessary duplicated commits (e.g. 90275824 and b9a745f7 seems
> to be two equivalent commits) and questionable changes from the overall
> project's point of view.
>
> For example, it renames git-subtree.txt to README.md at a4416ee; while I
> find the idea of departing from asciidoc somewhat attractive (perhaps this
> is only because I haven't been burned by markdown yet), if "git subtree"
> wants to live in the git.git repository, that change is a regression.
> Later the file is renamed back to git-subtree.txt (README.md is lost) at
> 9ffdeb, a commit with a single-liner "fixing typo" log message adds the
> README.md file with full contents of git-subtree.txt again at d9ccd03b,
> and then later merge of the branch at 8861de28 finally decides to revert
> that to have a shorter README.md that the history originally had, or
> something.  In short, it is a mess.
>
> Not very impressed, but I have this suspition that the history I was
> looking at was not what was meant to be sent to me and an older
> incarnation of the project before Herman cleaned it up for public
> consumption, or something.
>
> Confused...

I agree that it's a messy history.  It the result of the many painful
merges I did. In various stages a conflicting indentation and other
changes made it painful to get a clean merge.
In an attempt to get through this in a pragmatic way the history has
taken some damage.

Before  starting this latest subtree merge I actually tried to rebase.
However this failed very quickly, on the I think third commit out of 60,
landing me in conflict resolution as I had already been through.
I'd love to improve git but this was just taking too mush effort.
When I saw the quick result from subtree merge that seemed like a good
thing.

Wouldn't a good rebase have almost just as messy a history as the
subtree merge?

As an alternative I've now applied a patch with all changes on a clean
master branch.
In the commit message I've named all committers from the original history.
Would that be acceptable?
Its now available as https://github.com/helmo/git/tree/subtree-updates
The subtree merge version is still available as
https://github.com/helmo/git/tree/subtree-updates-merged

-- 

Met vriendelijke groet / Regards,

Herman van Rink
Initfour websolutions