mailing list mirror (one of many)
 help / color / mirror / code / Atom feed
From: Jeff King <>
To: Junio C Hamano <>
Cc: John Cai <>, git <>,
	Christian Couder <>
Subject: Re: [PATCH 1/3] fsck: free tree buffers after walking unreachable objects
Date: Thu, 22 Sep 2022 14:58:17 -0400	[thread overview]
Message-ID: <> (raw)
In-Reply-To: <xmqqa66rz20q.fsf@gitster.g>

On Thu, Sep 22, 2022 at 11:40:05AM -0700, Junio C Hamano wrote:

> > diff --git a/builtin/fsck.c b/builtin/fsck.c
> > index f7916f06ed..34e575a170 100644
> > --- a/builtin/fsck.c
> > +++ b/builtin/fsck.c
> > @@ -228,6 +228,8 @@ static void mark_unreachable_referents(const struct object_id *oid)
> >  
> >  	options.walk = mark_used;
> >  	fsck_walk(obj, NULL, &options);
> > +	if (obj->type == OBJ_TREE)
> > +		free_tree_buffer((struct tree *)obj);
> >  }
> Unlike codepaths like mark_object(), which uses the REACHABLE bit to
> avoid the walker coming into an already marked objects, we have no
> protection that says "this tree is already marked as USED, so lets
> not go into its contents" (it would be a disaster if we free tree
> buffer here and then later end up calling the function on the same
> tree), but it is OK because this is an unreachable object nobody
> points at and we will never come back?

I do think it is true that this is the final time we'd look at these
objects. But I don't think it would be a disaster if somebody did. The
free_tree_buffer() function clears the "parsed" flag on the struct. And
anybody wishing to look at the tree contents would need to call
parse_tree(), which would then re-load the contents.

In general, that's _possibly_ less efficient if we visit the same tree
twice, but it would be balanced against not holding all of the tree data
in RAM at once. And as I said, that doesn't happen for this use case

As a side note, IMHO having tree->buffer at all is a mistake, because it
leads to exactly this kind of confusion about when the buffer should be
discarded. We'd be better off having all callers parse directly into a
local buffer, and then clean up when they're done. It effectively ends
up the same, except then it's obvious when a tree is "leaked" because
the local buffer goes out of scope, rather than hanging around in the
global struct and just wasting memory. But that's obviously a much
bigger change.


  reply	other threads:[~2022-09-22 18:59 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-09-20 19:27 [INVESTIGATION] why is fsck --connectivity-only so much more expensive than rev-list --objects --all? John Cai
2022-09-20 20:41 ` Jeff King
2022-09-22 10:09   ` [PATCH 0/3] reducing fsck memory usage Jeff King
2022-09-22 10:11     ` [PATCH 1/3] fsck: free tree buffers after walking unreachable objects Jeff King
2022-09-22 18:40       ` Junio C Hamano
2022-09-22 18:58         ` Jeff King [this message]
2022-09-22 19:27           ` Junio C Hamano
2022-09-22 22:16             ` Jeff King
2022-09-22 10:13     ` [PATCH 2/3] fsck: turn off save_commit_buffer Jeff King
2022-09-22 10:15     ` [PATCH 3/3] parse_object_buffer(): respect save_commit_buffer Jeff King

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:

  List information:

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \ \ \ \ \ \ \

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
Code repositories for project(s) associated with this public inbox

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).