Difference between 'git rebase' and 'git rebase -m'

We ran into an interesting issue yesterday during a rebase.

User Bob had added a line at the same place in a (C++) .h file as User Joe, who had already pushed his one line change.  The natural expectation would be a conflict during User Bob's rebase operation, but, in fact, the conflicting patch applied without difficulty.

After an investigation, we discovered this particular file had a near duplicate of its C++ class definition immediately following the original but renamed into a thread-safe version; the rename of the class was the only change to its contents.  The default  'git rebase' operation looked some 150 lines into the file, discovered the patch didn't match due to User Joe's added line, continued scanning the file until it found another matching location some 300 lines later (the near duplicate thread-safe version of the class), and applied the patch there.

While coming up with a repro to post here, we cherry picked the individual changes and found the cherry pick properly conflicted.

We further discovered 'git rebase -m' (or 'git rebase -s recursive') also conflicted at the User Bob/Joe modification.

I can post a repro here, if needed.

I'm curious as to why 'git rebase -m' isn't the default and what the real difference is between 'git rebase' and 'git rebase -m'.  Additionally, is there any documentation describing how the patch algorithm determines context?

Thanks!

Josh

Joshua Jensen wrote:

I'm curious as to why 'git rebase -m' isn't the default and what the
real difference is between 'git rebase' and 'git rebase -m'.

git rebase is faster. :)

git rebase works by formatting a patch series and applying it.
git rebase -m uses repeated 3-way merge and should be more robust.

On Wed, Nov 17, 2010 at 1:27 AM, Jonathan Nieder [off-list ref] wrote:

Joshua Jensen wrote:

quoted

I'm curious as to why 'git rebase -m' isn't the default and what the
real difference is between 'git rebase' and 'git rebase -m'.

git rebase is faster. :)

Perhaps a config option to let people to default to -m. Or perhaps
they can just make an alias.

By the way, can we make rebase even faster? Rebasing ~20 patches or
more seems slow to me. I don't know much how patching works though.
-- 
Duy

Nguyen Thai Ngoc Duy venit, vidit, dixit 17.11.2010 12:43:

On Wed, Nov 17, 2010 at 1:27 AM, Jonathan Nieder [off-list ref] wrote:

quoted

Joshua Jensen wrote:

quoted

I'm curious as to why 'git rebase -m' isn't the default and what the
real difference is between 'git rebase' and 'git rebase -m'.

git rebase is faster. :)

Perhaps a config option to let people to default to -m. Or perhaps
they can just make an alias.

By the way, can we make rebase even faster? Rebasing ~20 patches or
more seems slow to me. I don't know much how patching works though.

git sequencer...

Michael

On Wed, Nov 17, 2010 at 8:36 PM, Michael J Gruber
[off-list ref] wrote:

Nguyen Thai Ngoc Duy venit, vidit, dixit 17.11.2010 12:43:

quoted

On Wed, Nov 17, 2010 at 1:27 AM, Jonathan Nieder [off-list ref] wrote:

quoted

Joshua Jensen wrote:

quoted

I'm curious as to why 'git rebase -m' isn't the default and what the
real difference is between 'git rebase' and 'git rebase -m'.

git rebase is faster. :)

Perhaps a config option to let people to default to -m. Or perhaps
they can just make an alias.

By the way, can we make rebase even faster? Rebasing ~20 patches or
more seems slow to me. I don't know much how patching works though.

git sequencer...

Sequencer is definitely an improvement. But does it improve
performance? I don't know rebase internals but I think that it needs
to update worktree/index for every commit, which might slow things
down a bit.
-- 
Duy