Re: [RFC] rebase: Handle cases where format-patch fails

From: Junio C Hamano <gitster@pobox.com>
Date: Oct 5, 2012, 20:17 UTC
Message-ID: <7vipaou0zw.fsf@alter.siamese.dyndns.org>
In-Reply-To: <1349412790-6087-2-git-send-email-andrew.kw.w@gmail.com>

Andrew Wong <andrew.kw.w@gmail.com> writes:

Show 29 quoted lines

> 'format-patch' could fail due to reasons such as out of memory. Such
> failures are not detected or handled, which causes rebase to incorrectly
> think that it completed successfully and continue with cleanup. i.e.
> calling move_to_original_branch
>
> Since only the exit status of the last command in the pipeline is
> available, we rely on || to detect whether 'format-patch' has failed.
>
> Also print messages to help user with how to recover from such failures.
>
> Signed-off-by: Andrew Wong <andrew.kw.w@gmail.com>
> ---
>  git-rebase--am.sh | 37 +++++++++++++++++++++++++++++++++++--
>  1 file changed, 35 insertions(+), 2 deletions(-)
>
> diff --git a/git-rebase--am.sh b/git-rebase--am.sh
> index 392ebc9..8dae804 100644
> --- a/git-rebase--am.sh
> +++ b/git-rebase--am.sh
> @@ -26,10 +26,43 @@ then
>  	# makes this easy
>  	git cherry-pick --allow-empty "$revisions"
>  else
> -	git format-patch -k --stdout --full-index --ignore-if-in-upstream \
> +	( git format-patch -k --stdout --full-index --ignore-if-in-upstream \
>  		--src-prefix=a/ --dst-prefix=b/ \
> -		--no-renames $root_flag "$revisions" |
> +		--no-renames $root_flag "$revisions" ||
> +		echo $? > "$GIT_DIR"/format-patch-failed ) |

Please make sure there is no marker-file that was leftover from previous invocation or whatever reason, e.g.

	rm -f "$GIT_DIR/format-patch-failed"
        (
		git format-patch -k --stdout --full-index --ignore-if-in-upstream \
			--src-prefix=a/ --dst-prefix=b/ \
			--no-renames $root_flag "$revisions" ||
		echo $? >"$GIT_DIR"/format-patch-failed
	) |
  	git am $git_am_opt --rebasing --resolvemsg="$resolvemsg"

But when format-patch dies for whatever reason, it is likely that the partial output will cause "am" to barf on the last part of it (either "missing patch text" if it stops in the middle of commit log message, or "corrupt patch" if it stops in the middle of a hunk). It may make sense to make this all-or-none, i.e. when format-patch fails, you do not even start "am", something like...

	rm -f "$GIT_DIR/patch-input"
        if ! git format-patch -k --stdout >"$GIT_DIR/patch-input" \
	        --full-index --ignore-if-in-upstream \
		--src-prefix=a/ --dst-prefix=b/ \
		--no-renames $root_flag "$revisions"
	then
		... format-patch barfed, here is how to deal with it...
	else
        	git am <"$GIT_DIR/patch-input" $git_am_opt ...
	fi
	rm -f "$GIT_DIR/patch-input"

but I wonder what the performance implication would be for normal cases.

Show 35 quoted lines

> +	ret=$?
> +	if test -f "$GIT_DIR"/format-patch-failed
> +	then
> +		ret=1
> +		rm -f "$GIT_DIR"/format-patch-failed
> +		if test -d "$state_dir"
> +		then
> +			echo
> +			echo "'git format-patch' seems to have failed in the middle of 'git am'."
> +			echo "If you continue rebasing, you will likely be losing some commits."
> +			echo "It is recommended that you abort rebasing by running:"
> +			echo
> +			echo "    git rebase --abort"
> +			echo
> +		else
> +			echo
> +			echo "'git format-patch' seems to have failed before 'git am' started."
> +			echo "It is impossible to continue or abort rebasing."
> +			echo "You have to use the following to return to your original head:"
> +			echo
> +			case "$head_name" in
> +			refs/*)
> +				echo "    git checkout $head_name"
> +				;;
> +			*)
> +				echo "    git checkout $orig_head"
> +				;;
> +			esac
> +			echo
> +		fi
> +	fi
> +	test 0 != $ret && false
>  fi && move_to_original_branch
>  
>  ret=$?