From: David Gudeman
To: [email protected]
Cc: Philip Oakley
Sent: Sunday, June 16, 2013 11:31 PM
Subject: Re: [git-users] what is the point of rebase?
Thanks for taking the time for such a detailed reply, Philip, but everything
you say seems to be true of merge as well so I still don't get what is special
about rebase. Is it just merge+throw away history? And if so, why throw away
the history rather than just hide it?
On Thursday, June 13, 2013 12:31:07 PM UTC-7, Philip Oakley wrote:
From: David Gdeman
Sent: Thursday, June 13, 2013 7:12 PM
Subject: [git-users] what is the point of rebase?
I've been getting to know git recently as part of some work I needed to
do, and I love the easy branching and merging, but I have to admit that rebase
is confusing me. I'm not confused about what it does (I don't think); I'm
confused about what the point is. As far as I can tell, rebase is just the same
as doing a merge and then throwing away some history. Why throw away the
history? All of the documentation I can find just refers to making history
"clean", but the history doesn't have to actually BE clean to LOOK clean. So
why?
Or here is another way to ask the question. Suppose git repositories had
a special flag that you could put on a branch that said, "don't show this
branch in histories or push it or allow it to be pulled unless the user does
something special to ask for hidden branches". Then suppose git had a
merge-and-hide command that would do a merge and then set this flag on the
topic branch to effectively hide it unless the user really wants to see it.
What does rebase accomplish that merge-and-hide would not?
I keep feeling that there must be something else, otherwise why deal with
the potential negative consequences of deleting history unrecoverably? All of
the documentation on rebase talks about the problem of rebasing a branch out
from under someone who is still working on it. So what is the advantage of
rebase over merge-and-hide that makes it worth dealing with this problem.
Rebase is a way, within a distributed system, of noticing that the world
has moved on since you started on your magic topic branch.
So you want your magic topic branch to cleanly apply to the New World Order
of the upstream repo, hence you 'rebase', or transfer, your series of nice
small changes (commits) to the latest upstream reference point (often tip of
master/production), and while doing the rebase, you fix all the conflicts (with
the NWO changes) as you go to create a clean sequence, and then you will have a
commit/patch series that can either be fast forward applied, or merged without
conflict.
Obviously (?), having done your rebase you will have to style check,
compile and test you new magic code to see if it actually works as planned
without any regressions (and fix a few minor mistakes). By which time the NWO
has moved on again, so another quick rebase gets you to the head of the queue,
and you are awarded tea and medals.
The other scenario is that you are working on say git itself and you send
in patches for review, which end up with fixups and comments, so you have to
continue to rebase your own work forward to the relevant start point,
especially if there is an overlap with other patch series.
The "problem" doesn't exist in other, non-distributed, systems because of
the restrictive check-out - check-in process which nails both feet to the floor
to stop you going round in circles (and limits progress!). Rebase lets you leap
ahead and catch up with the latest and greatest code release, or a
collaborators code, or wherever.
Philip
Hi david,
Interesting question. The distinction is about 'who does what', and how is the
hsitory to be presented.
If it is just you on your own, then a rebase is likely to be rare. You would
use it if you started a piece of work from the wrong start point and wanted to
make that commit series start from the right point. Otherwise you can merge
your way forward anyway you please.
In a multi-developer situation there are often rules and conventions that mean
you have to 'keep up' with the work of others, so you have to sort out the
conflicts *before* your commits are merged into the main workflow (with the
merge having minimal conflicts), so you rebase your work, in your local repo,
onto the tip of the fetched mainline (which had moved on), fixing the conflicts
as the commmits are rebased, leaving you with a clean series that obviously can
be either fast forwarded, or merged without conflict, into the main line.
Summary. In a rebase you fix conflicts before the merge, while in a regular
merge you usually have conflicts to fixup afterwards (after the auto-merge
step).
Philip
--
You received this message because you are subscribed to the Google Groups "Git
for human beings" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.
