diff mbox series

[v3,4/4] commit: don't lazy-fetch commits

Message ID 5924a5120bc8e0bf529fc1cde5c23724550f72a4.1670532905.git.jonathantanmy@google.com (mailing list archive)
State Superseded
Headers show
Series Don't lazy-fetch commits when parsing them | expand

Commit Message

Jonathan Tan Dec. 8, 2022, 8:57 p.m. UTC
When parsing commits, fail fast when the commit is missing or
corrupt, instead of attempting to fetch them. This is done by inlining
repo_read_object_file() and setting the flag that prevents fetching.

This is motivated by a situation in which through a bug (not necessarily
through Git), there was corruption in the object store of a partial
clone. In this particular case, the problem was exposed when "git gc"
tried to expire reflogs, which calls repo_parse_commit(), which triggers
fetches of the missing commits.

(There are other possible solutions to this problem including passing an
argument from "git gc" to "git reflog" to inhibit all lazy fetches, but
I think that this fix is at the wrong level - fixing "git reflog" means
that this particular command works fine, or so we think (it will fail if
it somehow needs to read a legitimately missing blob, say, a .gitmodules
file), but fixing repo_parse_commit() will fix a whole class of bugs.)

Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
---
 commit.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

Comments

Ævar Arnfjörð Bjarmason Dec. 9, 2022, 2:14 p.m. UTC | #1
On Thu, Dec 08 2022, Jonathan Tan wrote:


> diff --git a/commit.c b/commit.c
> index 572301b80a..a02723f06b 100644
> --- a/commit.c
> +++ b/commit.c
> @@ -508,6 +508,17 @@ int repo_parse_commit_internal(struct repository *r,
>  	enum object_type type;
>  	void *buffer;
>  	unsigned long size;
> +	struct object_info oi = {
> +		.typep = &type,
> +		.sizep = &size,
> +		.contentp = &buffer,
> +	};
> +	/*
> +	 * Git does not support partial clones that exclude commits, so set
> +	 * OBJECT_INFO_SKIP_FETCH_OBJECT to fail fast when an object is missing.
> +	 */
> +	int flags = OBJECT_INFO_LOOKUP_REPLACE | OBJECT_INFO_SKIP_FETCH_OBJECT |
> +		OBJECT_INFO_DIE_IF_CORRUPT;
>  	int ret;
>  
>  	if (!item)
> @@ -516,8 +527,8 @@ int repo_parse_commit_internal(struct repository *r,
>  		return 0;
>  	if (use_commit_graph && parse_commit_in_graph(r, item))
>  		return 0;
> -	buffer = repo_read_object_file(r, &item->object.oid, &type, &size);
> -	if (!buffer)
> +
> +	if (oid_object_info_extended(r, &item->object.oid, &oi, flags) < 0)

Style: you're adding another \n here, usually we'd prefer it, but here
the function already has all these checks bundled together without a \n,
and this "if" is followed by one without it.

But then again those two "if"'s have to do with populating the "oi" and
then reading out the "type", so it's probably fine & OK.
diff mbox series

Patch

diff --git a/commit.c b/commit.c
index 572301b80a..a02723f06b 100644
--- a/commit.c
+++ b/commit.c
@@ -508,6 +508,17 @@  int repo_parse_commit_internal(struct repository *r,
 	enum object_type type;
 	void *buffer;
 	unsigned long size;
+	struct object_info oi = {
+		.typep = &type,
+		.sizep = &size,
+		.contentp = &buffer,
+	};
+	/*
+	 * Git does not support partial clones that exclude commits, so set
+	 * OBJECT_INFO_SKIP_FETCH_OBJECT to fail fast when an object is missing.
+	 */
+	int flags = OBJECT_INFO_LOOKUP_REPLACE | OBJECT_INFO_SKIP_FETCH_OBJECT |
+		OBJECT_INFO_DIE_IF_CORRUPT;
 	int ret;
 
 	if (!item)
@@ -516,8 +527,8 @@  int repo_parse_commit_internal(struct repository *r,
 		return 0;
 	if (use_commit_graph && parse_commit_in_graph(r, item))
 		return 0;
-	buffer = repo_read_object_file(r, &item->object.oid, &type, &size);
-	if (!buffer)
+
+	if (oid_object_info_extended(r, &item->object.oid, &oi, flags) < 0)
 		return quiet_on_missing ? -1 :
 			error("Could not read %s",
 			     oid_to_hex(&item->object.oid));