Re: [PATCH v3 06/12] bisect--helper: finish porting `bisect_start()` to C

[Johannes Schindelin <Johannes.Schindelin@gmx.de>]

Hi Miriam,

On Thu, 23 Apr 2020, Miriam Rubio wrote:

> From: Pranit Bauva <pranit.bauva@gmail.com>
>
> Add the subcommand to `git bisect--helper` and call it from
> git-bisect.sh.
>
> With the conversion of `bisect_auto_next()` from shell to C in a
> previous commit, `bisect_start()` can now be fully ported to C.
>
> So let's complete the `--bisect-start` subcommand of
> `git bisect--helper` so that it fully implements `bisect_start()`,
> and let's use this subcommand in `git-bisect.sh` instead of
> `bisect_start()`.
>
> This removes the signal handling we had in `bisect_start()` as we
> don't really need it. While at it, "trap" is changed to "handle".

I do not see any "handle" in the diff...

> As "trap" is a reference to the shell "trap" builtin, which isn't
> used anymore.

Otherwise, the commit message looks good to me!

>
> Mentored-by: Christian Couder <chriscool@tuxfamily.org>
> Mentored-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
> Signed-off-by: Pranit Bauva <pranit.bauva@gmail.com>
> Signed-off-by: Tanushree Tumane <tanushreetumane@gmail.com>
> Signed-off-by: Miriam Rubio <mirucam@gmail.com>
> ---
>  builtin/bisect--helper.c | 54 +++++++++++++++++++++++++++++++---------
>  git-bisect.sh            | 26 ++-----------------
>  2 files changed, 44 insertions(+), 36 deletions(-)
>
> diff --git a/builtin/bisect--helper.c b/builtin/bisect--helper.c
> index c6aaa8eb15..f2ec7f89e4 100644
> --- a/builtin/bisect--helper.c
> +++ b/builtin/bisect--helper.c
> @@ -588,11 +588,12 @@ static enum bisect_error bisect_auto_next(struct bisect_terms *terms, const char
>  	return bisect_next(terms, prefix);
>  }
>
> -static int bisect_start(struct bisect_terms *terms, int no_checkout,
> +static enum bisect_error bisect_start(struct bisect_terms *terms, int no_checkout,
>  			const char **argv, int argc)
>  {
>  	int i, has_double_dash = 0, must_write_terms = 0, bad_seen = 0;
> -	int flags, pathspec_pos, res = 0;
> +	int flags, pathspec_pos;
> +	enum bisect_error res = BISECT_OK;
>  	struct string_list revs = STRING_LIST_INIT_DUP;
>  	struct string_list states = STRING_LIST_INIT_DUP;
>  	struct strbuf start_head = STRBUF_INIT;
> @@ -645,9 +646,12 @@ static int bisect_start(struct bisect_terms *terms, int no_checkout,
>  			return error(_("unrecognized option: '%s'"), arg);
>  		} else {
>  			char *commit_id = xstrfmt("%s^{commit}", arg);
> -			if (get_oid(commit_id, &oid) && has_double_dash)
> -				die(_("'%s' does not appear to be a valid "
> -				      "revision"), arg);
> +			if (get_oid(commit_id, &oid) && has_double_dash) {
> +				error(_("'%s' does not appear to be a valid "
> +					"revision"), arg);
> +				free(commit_id);
> +				return BISECT_FAILED;
> +			}

Makes sense.

>
>  			string_list_append(&revs, oid_to_hex(&oid));
>  			free(commit_id);
> @@ -725,12 +729,12 @@ static int bisect_start(struct bisect_terms *terms, int no_checkout,
>  	 * Get rid of any old bisect state.
>  	 */
>  	if (bisect_clean_state())
> -		return -1;
> +		return BISECT_FAILED;
>
>  	/*
> -	 * In case of mistaken revs or checkout error, or signals received,
> +	 * In case of mistaken revs or checkout error,
>  	 * "bisect_auto_next" below may exit or misbehave.
> -	 * We have to trap this to be able to clean up using
> +	 * We have to handle this to be able to clean up using
>  	 * "bisect_clean_state".
>  	 */
>
> @@ -746,7 +750,7 @@ static int bisect_start(struct bisect_terms *terms, int no_checkout,
>  		}
>  		if (update_ref(NULL, "BISECT_HEAD", &oid, NULL, 0,
>  			       UPDATE_REFS_MSG_ON_ERR)) {
> -			res = -1;
> +			res = BISECT_FAILED;
>  			goto finish;
>  		}
>  	}
> @@ -758,25 +762,51 @@ static int bisect_start(struct bisect_terms *terms, int no_checkout,
>  	for (i = 0; i < states.nr; i++)
>  		if (bisect_write(states.items[i].string,
>  				 revs.items[i].string, terms, 1)) {
> -			res = -1;
> +			res = BISECT_FAILED;
>  			goto finish;
>  		}
>
>  	if (must_write_terms && write_terms(terms->term_bad,
>  					    terms->term_good)) {
> -		res = -1;
> +		res = BISECT_FAILED;
>  		goto finish;
>  	}
>
>  	res = bisect_append_log_quoted(argv);
>  	if (res)
> -		res = -1;
> +		res = BISECT_FAILED;
>
>  finish:
>  	string_list_clear(&revs, 0);
>  	string_list_clear(&states, 0);
>  	strbuf_release(&start_head);
>  	strbuf_release(&bisect_names);
> +	if (res)
> +		return res;
> +
> +	res = bisect_auto_next(terms, NULL);
> +	/*
> +	 * In case of mistaken revs or checkout error, or signals received,
> +	 * "bisect_auto_next" below may exit or misbehave.

This "below" was true in the shell script version, but now it should be
"above".

> +	 * We have to handle this to be able to clean up using
> +	 * "bisect_clean_state".
> +	 * return code BISECT_INTERNAL_SUCCESS_MERGE_BASE is special code
> +	 * that indicates special success.

Maybe we should have a helper function and use it here:

	static int is_bisect_success(enum bisect_error res)
	{
		return !res ||
			res == BISECT_INTERNAL_1ST_BAD_FOUND ||
			res == BISECT_INTERNAL_SUCCESS_MERGE_BASE;
	}

(Note that I added the `_1ST_BAD_FOUND` case here.)

> +	 *	-> bisect_start()
> +	 *	   . res = bisect_auto_next()
> +	 *	    -> bisect_auto_next()
> +	 *	       . return bisect_next()
> +	 *	       -> bisect_next()
> +	 *		  . res = bisect_next_all()
> +	 *		  -> bisect_next_all()
> +	 *		     . res = check_good_are_ancestors_of_bad()
> +	 *		     -> check_good_are_ancestors_of_bad()
> +	 *			. res = check_merge_bases()
> +	 *			-> check_merge_bases()
> +	 *			   . res = BISECT_INTERNAL_SUCCESS_MERGE_BASE

While this might be an interesting tidbit for the commit message, I do not
think that it is a good idea to keep this as a code comment; It is prone
to become stale/incorrect over time.

> +	 */
> +	if (res && res != BISECT_INTERNAL_SUCCESS_MERGE_BASE)
> +		bisect_clean_state();
>  	return res;
>  }
>
> diff --git a/git-bisect.sh b/git-bisect.sh
> index 897825b675..049ffacdff 100755
> --- a/git-bisect.sh
> +++ b/git-bisect.sh
> @@ -49,27 +49,6 @@ bisect_head()
>  	fi
>  }
>
> -bisect_start() {
> -	git bisect--helper --bisect-start $@ || exit
> -
> -	#
> -	# Change state.
> -	# In case of mistaken revs or checkout error, or signals received,
> -	# "bisect_auto_next" below may exit or misbehave.
> -	# We have to trap this to be able to clean up using
> -	# "bisect_clean_state".
> -	#
> -	trap 'git bisect--helper --bisect-clean-state' 0
> -	trap 'exit 255' 1 2 3 15
> -
> -	#
> -	# Check if we can proceed to the next bisect state.
> -	#
> -	git bisect--helper --bisect-auto-next || exit
> -
> -	trap '-' 0
> -}
> -
>  bisect_skip() {
>  	all=''
>  	for arg in "$@"
> @@ -162,8 +141,7 @@ bisect_replay () {
>  		get_terms
>  		case "$command" in
>  		start)
> -			cmd="bisect_start $rev"
> -			eval "$cmd" ;;
> +			eval "git bisect--helper --bisect-start $rev" ;;

The original code _had_ to use `eval` because `bisect_start` could cause
the code to `exit` and we did _not_ want to let that exit the `replay`
process.

However, we are calling `git bisect--helper` here, which runs in its own
process anyway, so the `eval` should be dropped now.

Otherwise, the patch looks fine to me,
Dscho

>  		"$TERM_GOOD"|"$TERM_BAD"|skip)
>  			git bisect--helper --bisect-write "$command" "$rev" "$TERM_GOOD" "$TERM_BAD" || exit;;
>  		terms)
> @@ -262,7 +240,7 @@ case "$#" in
>  	help)
>  		git bisect -h ;;
>  	start)
> -		bisect_start "$@" ;;
> +		git bisect--helper --bisect-start "$@" ;;
>  	bad|good|new|old|"$TERM_BAD"|"$TERM_GOOD")
>  		bisect_state "$cmd" "$@" ;;
>  	skip)
> --
> 2.25.0
>
>