Re: [egit-dev] Gerrit Commits

[Robin Rosenberg <robin.rosenberg@xxxxxxxxxx>]

måndagen den 28 december 2009 00.30.26 skrev  Shawn O. Pearce:
> Gunnar Wagenknecht <gunnar@xxxxxxxxxxxxxxx> wrote:
> > I looked at the new Git URLs via HTTP. I like that way of quickly
> > browsing what's going on in a repository. But something that really
> > bugged me were a few commits made by a "Code Review" user. I tend to
> > call them anonymous commits because there is no indication where it came
> > from (just a cryptic id) nor who approved and committed it.
> 
....
> Well, we could do a few things.df
> 
> I could change our Gerrit configuration to cherry-pick the change
> onto the branch, rather than merge it.  That means that in the
> above case we would instead wind up with:
> 
>                          +-- master
> 
>    +------------ R' ---- S'
>   /
>   B --------- R
>    \
>     +--- S
> 
> Where R' and S' are the same delta as R and S, but have a different
> commit SHA-1 than the author created on their system.  It also loses
> the fact that S was actually develoepd against B, not against R'.
> As such I usually don't prefer this style of submitting changes,
> and that is why it isn't the default.

That would be a nice option as some commits are usable on their own,
such as the stream sequence you submitted recently in the middle of a
larger patch set (hypotetical example). But it's not the solution to the
particular problem here.

> The other thing we could do is change Gerrit's code to try and
> attribute those merge commits differently.  Gerrit could look for
> the submitter data of the change(s) being merged and try to use that
> as the author of the merge commit, instead of the generic build user.

Indeed it should. 

> The problem with that style is, some people at my $DAY_JOB disagreetur
> and say these commits are false, they didn't make them, they just
> pushed the submit button, the software made the commit on their
> behalf.

It's like saying, "git merged", not me and blame Linus for everything.

> Although with both an author and a committer line available in the
> header we could assign the submitter as the author line, and the
> generic user identity as the committer line, to make it clear that
> this commit was made by Gerrit on behalf of the user who clicked
> the submit button, vs. being a commit the user made on their desktop
> and uploaded.

This solution gets my vote.

> 
> For the latter, this is probably an existing issue in the Gerrit
> issue tracker http://code.google.com/p/gerrit/issues/detail?id=162
> 

Looks that way.

A slightly disturbing aspect of these automatic merges is that they
are not tested. How do we fix those? IMHO broken merges shoul be
fixed as an "evil" merge, shouldn't they?

-- robin