threads / patch / 63819

patchblame: only parse detailed commit info if needed

Subject: [PATCH] blame: only parse detailed commit info if needed

## tl;dr

4 messages between Jul 22, 2025 and Jul 23, 2025. Diffs are folded; open one to read it.

replies: 3people: 3as markdown or json

Han Young· Jul 22, 2025, 11:42 UTC · lore

In commit cee7f245d (git-pickaxe: blame rewritten., 2006-10-19), The function get_commit_info can terminate commit parsing early if only the author information is needed. This ability is not used by callers who do not require detailed commit information. Stop requesting detailed commit information for these callers.

Signed-off-by: Han Young <hanyang.tony@bytedance.com>
---
 builtin/blame.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Show changes to builtin/blame.c +2 −2

diff --git a/builtin/blame.c b/builtin/blame.c
index 91586e685..b6a38e530 100644
--- a/builtin/blame.c
+++ b/builtin/blame.c
@@ -471,7 +471,7 @@ static void emit_other(struct blame_scoreboard *sb, struct blame_entry *ent, int
 	int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
 	const char *default_color = NULL, *color = NULL, *reset = NULL;
 
-	get_commit_info(suspect->commit, &ci, 1);
+	get_commit_info(suspect->commit, &ci, 0);
 	oid_to_hex_r(hex, &suspect->commit->object.oid);
 
 	cp = blame_nth_line(sb, ent->lno);
@@ -665,7 +665,7 @@ static void find_alignment(struct blame_scoreboard *sb, int *option)
 		if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
 			struct commit_info ci = COMMIT_INFO_INIT;
 			suspect->commit->object.flags |= METAINFO_SHOWN;
-			get_commit_info(suspect->commit, &ci, 1);
+			get_commit_info(suspect->commit, &ci, 0);
 			if (*option & OUTPUT_SHOW_EMAIL)
 				num = utf8_strwidth(ci.author_mail.buf);
 			else

-- 
2.50.1.321.gea4e667e7

Patrick Steinhardt· Jul 22, 2025, 14:15 UTC · re: Han Young · lore

Re: [PATCH] blame: only parse detailed commit info if needed

On Tue, Jul 22, 2025 at 07:42:20PM +0800, Han Young wrote:

Show 5 quoted lines

> In commit cee7f245d (git-pickaxe: blame rewritten., 2006-10-19),
> The function get_commit_info can terminate commit parsing early if only
> the author information is needed. This ability is not used by callers
> who do not require detailed commit information. Stop requesting detailed
> commit information for these callers.

Okay. I think there's two important pieces of information missing here:

  - What does this buy us? I guess the answer is performance, but it
    would be sure to quantify in which scenarios and how much of a
    speedup this buys us.

  - Any reasoning why those two callers don't need the information.
    Reviewers can try to piece it together manually, but it would be
    nice to hold their hand and lead them through the change.

Thanks!

Patrick

Junio C Hamano· Jul 22, 2025, 17:08 UTC · re: Patrick Steinhardt · lore

Re: [PATCH] blame: only parse detailed commit info if needed

Patrick Steinhardt <ps@pks.im> writes:

Show 16 quoted lines

> On Tue, Jul 22, 2025 at 07:42:20PM +0800, Han Young wrote:
>> In commit cee7f245d (git-pickaxe: blame rewritten., 2006-10-19),
>> The function get_commit_info can terminate commit parsing early if only
>> the author information is needed. This ability is not used by callers
>> who do not require detailed commit information. Stop requesting detailed
>> commit information for these callers.
>
> Okay. I think there's two important pieces of information missing here:
>
>   - What does this buy us? I guess the answer is performance, but it
>     would be sure to quantify in which scenarios and how much of a
>     speedup this buys us.
>
>   - Any reasoning why those two callers don't need the information.
>     Reviewers can try to piece it together manually, but it would be
>     nice to hold their hand and lead them through the change.

Good questions to ask.

If the answer to the first question is "well, not really?", then another thing to consider would be if we want to remove that short-cut as conditionally grabbing only just some pieces of information without getting others is not helping.

Thanks.

Han Young· Jul 23, 2025, 06:02 UTC · re: Patrick Steinhardt · lore

Re: [External] Re: [PATCH] blame: only parse detailed commit info if needed

On Tue, Jul 22, 2025 at 10:16 PM Patrick Steinhardt <ps@pks.im> wrote:

>   - What does this buy us? I guess the answer is performance, but it
>     would be sure to quantify in which scenarios and how much of a
>     speedup this buys us.

I actually ran some performance tests before submitting this patch, On a 5000-line file with a fairly long history, running "git blame --porcelain FILE" for 100 times, the speedup is less than 1 second. Considering the total run time is 180 seconds, I think the performance gain is negligible (the speed increase could even be due to system noise).

> - Any reasoning why those two callers don't need the information.
> Reviewers can try to piece it together manually, but it would be
> nice to hold their hand and lead them through the change.

These two callers only access the author information part of the commit_info struct, before discarding the commit_info object. Sorry for not including this information in the patch description.

On Wed, Jul 23, 2025 at 1:08 AM Junio C Hamano <gitster@pobox.com> wrote:

> If the answer to the first question is "well, not really?", then
> another thing to consider would be if we want to remove that
> short-cut as conditionally grabbing only just some pieces of
> information without getting others is not helping.

All callers of the get_commit_info function request detailed commit information. Removing the shortcut would reduce the complexity of the codebase. I will send another patch to remove the "detailed" param from the get_commit_info.

Thanks