git@vger.kernel.org mailing list mirror (one of many)
 help / Atom feed
From: Derrick Stolee <stolee@gmail.com>
To: Jeff King <peff@peff.net>
Cc: Constantine <hi-angel@yandex.ru>,
	Junio C Hamano <gitster@pobox.com>,
	Christian Couder <christian.couder@gmail.com>,
	Mike Hommey <mh@glandium.org>, git <git@vger.kernel.org>
Subject: Re: git-clone causes out of memory
Date: Fri, 13 Oct 2017 10:25:10 -0400
Message-ID: <42cbcb4f-7f9d-df69-f55e-0ba42b931957@gmail.com> (raw)
In-Reply-To: <20171013142004.ocxpdkkbcxpi52yv@sigill.intra.peff.net>

On 10/13/2017 10:20 AM, Jeff King wrote:
> On Fri, Oct 13, 2017 at 10:10:18AM -0400, Jeff King wrote:
>
>> Hmm. So this patch makes it go fast:
>>
>> diff --git a/revision.c b/revision.c
>> index d167223e69..b52ea4e9d8 100644
>> --- a/revision.c
>> +++ b/revision.c
>> @@ -409,7 +409,7 @@ static void file_add_remove(struct diff_options *options,
>>   	int diff = addremove == '+' ? REV_TREE_NEW : REV_TREE_OLD;
>>   
>>   	tree_difference |= diff;
>> -	if (tree_difference == REV_TREE_DIFFERENT)
>> +	if (tree_difference & REV_TREE_DIFFERENT)
>>   		DIFF_OPT_SET(options, HAS_CHANGES);
>>   }
>>   
>>
>> But that essentially makes the conditional a noop (since we know we set
>> either NEW or OLD above and DIFFERENT is the union of those flags).
>>
>> I'm not sure I understand why file_add_remove() would ever want to avoid
>> setting HAS_CHANGES (certainly its companion file_change() always does).
>> This goes back to Junio's dd47aa3133 (try-to-simplify-commit: use
>> diff-tree --quiet machinery., 2007-03-14).
>>
>> Maybe I am missing something, but AFAICT this was always buggy. But
>> since it only affects adds and deletes, maybe nobody noticed? I'm also
>> not sure if it only causes a slowdown, or if this could cause us to
>> erroneously mark something as TREESAME which isn't (I _do_ think people
>> would have noticed that).
> Answering my own question a little, there is a hint in the comment
> a few lines above:
>
>    /*
>     * The goal is to get REV_TREE_NEW as the result only if the
>     * diff consists of all '+' (and no other changes), REV_TREE_OLD
>     * if the whole diff is removal of old data, and otherwise
>     * REV_TREE_DIFFERENT (of course if the trees are the same we
>     * want REV_TREE_SAME).
>     * That means that once we get to REV_TREE_DIFFERENT, we do not
>     * have to look any further.
>     */
>
> So my patch above is breaking that. But it's not clear from that comment
> why we care about knowing the different between NEW, OLD, and DIFFERENT.
>
> Grepping around for REV_TREE_NEW and REV_TREE_OLD, I think the answer is
> in try_to_simplify_commit():
>
>       case REV_TREE_NEW:
>                if (revs->remove_empty_trees &&
>                    rev_same_tree_as_empty(revs, p)) {
>                        /* We are adding all the specified
>                         * paths from this parent, so the
>                         * history beyond this parent is not
>                         * interesting.  Remove its parents
>                         * (they are grandparents for us).
>                         * IOW, we pretend this parent is a
>                         * "root" commit.
>                         */
>                        if (parse_commit(p) < 0)
>                                die("cannot simplify commit %s (invalid %s)",
>                                    oid_to_hex(&commit->object.oid),
>                                    oid_to_hex(&p->object.oid));
>                        p->parents = NULL;
>                }
>      /* fallthrough */
>      case REV_TREE_OLD:
>      case REV_TREE_DIFFERENT:
>
> So when --remove-empty is not in effect (and it's not by default), we
> don't care about OLD vs NEW, and we should be able to optimize further.
>
> -Peff

This does appear to be the problem. The missing DIFF_OPT_HAS_CHANGES is 
causing diff_can_quit_early() to return false. Due to the corner-case of 
the bug it seems it will not be a huge performance improvement in most 
cases. Still worth fixing and I'm looking at your suggestions to try and 
learn this area better.

It will speed up natural cases of someone adding or renaming a folder 
with a lot of contents, including someone initializing a repository from 
an existing codebase. This git bomb case happens to add all of the 
folders at once, which is why the performance is so noticeable. I use a 
version of this "exponential file growth" for testing that increases the 
folder-depth one commit at a time, so the contents are rather small in 
the first "add", hence not noticing this issue.

Thanks,
-Stolee

  reply index

Thread overview: 23+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-10-13  9:51 Constantine
2017-10-13 10:06 ` Mike Hommey
2017-10-13 10:26   ` Christian Couder
2017-10-13 10:37     ` Mike Hommey
2017-10-13 10:44       ` Christian Couder
2017-10-13 12:04         ` Junio C Hamano
2017-10-13 12:12           ` Constantine
2017-10-13 12:44             ` Jeff King
2017-10-13 13:15               ` Derrick Stolee
2017-10-13 13:39                 ` Derrick Stolee
2017-10-13 13:50                   ` Jeff King
2017-10-13 13:55                     ` Derrick Stolee
2017-10-13 13:56                       ` Jeff King
2017-10-13 14:10                         ` Jeff King
2017-10-13 14:20                           ` Jeff King
2017-10-13 14:25                             ` Derrick Stolee [this message]
2017-10-13 14:26                               ` Jeff King
2017-10-13 14:30                                 ` Derrick Stolee
2017-10-13 15:27                                 ` [PATCH] revision: quit pruning diff more quickly when possible Jeff King
2017-10-13 15:37                                   ` Derrick Stolee
2017-10-13 15:44                                     ` Jeff King
2017-10-14  2:43                                   ` Junio C Hamano
2017-10-13 12:35   ` git-clone causes out of memory Jeff King

Reply instructions:

You may reply publically 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:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

  List information: http://vger.kernel.org/majordomo-info.html

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

  git send-email \
    --in-reply-to=42cbcb4f-7f9d-df69-f55e-0ba42b931957@gmail.com \
    --to=stolee@gmail.com \
    --cc=christian.couder@gmail.com \
    --cc=git@vger.kernel.org \
    --cc=gitster@pobox.com \
    --cc=hi-angel@yandex.ru \
    --cc=mh@glandium.org \
    --cc=peff@peff.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

git@vger.kernel.org mailing list mirror (one of many)

Archives are clonable:
	git clone --mirror https://public-inbox.org/git
	git clone --mirror http://ou63pmih66umazou.onion/git
	git clone --mirror http://czquwvybam4bgbro.onion/git
	git clone --mirror http://hjrcffqmbrq6wope.onion/git

Newsgroups are available over NNTP:
	nntp://news.public-inbox.org/inbox.comp.version-control.git
	nntp://ou63pmih66umazou.onion/inbox.comp.version-control.git
	nntp://czquwvybam4bgbro.onion/inbox.comp.version-control.git
	nntp://hjrcffqmbrq6wope.onion/inbox.comp.version-control.git
	nntp://news.gmane.org/gmane.comp.version-control.git

 note: .onion URLs require Tor: https://www.torproject.org/
       or Tor2web: https://www.tor2web.org/

AGPL code for this site: git clone https://public-inbox.org/ public-inbox