Re: Unreliable 'git rebase --onto'

[SZEDER Gábor <szeder.dev@xxxxxxxxx>]

On Wed, Jan 08, 2020 at 10:43:49PM +0100, Eugeniu Rosca wrote:
> Hello Git community,
> 
> Below is a simple reproduction scenario for what looks to be a bug (?)
> in 'git rebase --onto' (v2.25.0-rc1-19-g042ed3e048af).
> 
> I would appreciate your confirmation of the misbehavior.
> If the behavior is correct/expected, I would appreciate some feedback
> how to avoid it in future, since it occurs with the default parameters.
> 
> 1. git clone https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
> 
> 2. ### Cherry pick an upstream commit, to contrast the results with
>    'git rebase --onto':
>    $ git checkout -b v4.18-cherry-pick v4.18
>    $ git cherry-pick 463fa44eec2fef50
>    Auto-merging drivers/input/touchscreen/atmel_mxt_ts.c
>    warning: inexact rename detection was skipped due to too many files.
>    warning: you may want to set your merge.renamelimit variable to at least 7216 and retry the command.
>    [v4.18-cherry-pick bd142b45bf3a] Input: atmel_mxt_ts - disable IRQ across suspend
>     Author: Evan Green <evgreen@xxxxxxxxxxxx>
>     Date: Wed Oct 2 14:00:21 2019 -0700
>     1 file changed, 4 insertions(+)
> 
> 3. ### In spite of the warning, the result matches the original commit:
>    $ vimdiff <(git show 463fa44eec2fef50) <(git show v4.18-cherry-pick)
> 
> 4. ### Now, backport the same commit via 'git rebase --onto'
>    $ git rebase --onto v4.18 463fa44eec2fef50~ 463fa44eec2fef50
>    First, rewinding head to replay your work on top of it...
>    Applying: Input: atmel_mxt_ts - disable IRQ across suspend
> 
> 5. ### The result is different:
>    $ git branch v4.18-rebase-onto
>    $ git diff v4.18-cherry-pick v4.18-rebase-onto
> 
> diff --git a/drivers/input/touchscreen/atmel_mxt_ts.c b/drivers/input/touchscreen/atmel_mxt_ts.c
> index b45958e89cc5..2345b587662b 100644
> --- a/drivers/input/touchscreen/atmel_mxt_ts.c
> +++ b/drivers/input/touchscreen/atmel_mxt_ts.c
> @@ -3139,8 +3139,6 @@ static int __maybe_unused mxt_suspend(struct device *dev)
>  
>  	mutex_unlock(&input_dev->mutex);
>  
> -	disable_irq(data->irq);
> -
>  	return 0;
>  }
>  
> @@ -3162,6 +3160,8 @@ static int __maybe_unused mxt_resume(struct device *dev)
>  
>  	mutex_unlock(&input_dev->mutex);
>  
> +	disable_irq(data->irq);
> +
>  	return 0;
>  }
> 
> 
> In a nutshell, purely from user's perspective:
>  - I get a warning from 'git cherry pick', with perfect results
>  - I get no warning from 'git rebase --onto', with wrong results
> 
> Does git still behave expectedly? TIA!

This is a known issue with the 'am' backend of 'git rebase'.

The good news is that work is already well under way to change the
default backend from 'am' to 'merge', which will solve this issue.
>From the log message of aa523de170 (rebase: change the default backend
from "am" to "merge", 2019-12-24):

  The am-backend drops information and thus limits what we can do:
  [...]
    * reduction in context from only having a few lines beyond those
      changed means that when context lines are non-unique we can apply
      patches incorrectly.[2]
  [...]
  [2] https://lore.kernel.org/git/CABPp-BGiu2nVMQY_t-rnFR5GQUz_ipyEE8oDocKeO+>

Alas, there is unexpected bad news: with that commit the runtime of
your 'git rebase --onto' command goes from <1sec to over 50secs.
Cc-ing Elijah, author of that patch...