* Re: cg-clone produces "___" file and no working tree
From: Zack Brown @ 2006-04-20 20:08 UTC (permalink / raw)
To: Junio C Hamano; +Cc: git
In-Reply-To: <7vejzsywrq.fsf@assigned-by-dhcp.cox.net>
On Thu, Apr 20, 2006 at 10:36:25AM -0700, Junio C Hamano wrote:
> Zack Brown <zbrown@tumblerings.org> writes:
>
> > I just downloaded the latest versions of git and cogito from kernel.org:
> > cogito-0.17.2 and git-1.3.0; put their directories in my path, and ran "make" on
> > both of them. There's no other version in my path.
>
> Earlier, you were having this symptom:
>
> >> What do these command say?
> >>
> >> $ git --exec-path
> >> $ ls -l "`git --exec-path`/git-clone"
> >
> > 22:07:05 [zbrown] ~$ git --exec-path
> > /home/zbrown/bin
> > 07:10:34 [zbrown] ~$ ls -l "`git --exec-path`/git-clone"
> > ls: /home/zbrown/bin/git-clone: No such file or directory
> >
> > Does that mean it's looking in /home/zbrown/bin for the git binaries?
>
> If that is the case, you did not just (quote) "and ran "make"".
>
> You must have run "make frotz=xyzzy target", but you did not mention
> what frotz, xyzzy and target were.
Not true. I went into the git source directory, and ran "make". Nothing more.
I've been doing that for a long time, whenever I sync with the repository. I
didn't know the installation instructions had changed.
> It probably would help if you did this:
>
> make clean
> make bindir=$HOME/git/git gitexecdir=$HOME/git/git/
> make bindir=$HOME/git/git gitexecdir=$HOME/git/git/ install
OK, I did this. The first 2 commands worked fine. The third complained of
duplicate files, and exited with an error. Maybe because the source tree is also
$HOME/git/git
I then did a 'cd ..; mkdir tmp; cd tmp; git-init-db' as before, but there
is still no ".git/info" entry created.
Be well,
Zack
>
> As I said in a previous message, the first paragraph in INSTALL
> file explains this.
>
--
Zack Brown
^ permalink raw reply
* Re: cg-clone produces "___" file and no working tree
From: Junio C Hamano @ 2006-04-20 20:17 UTC (permalink / raw)
To: Zack Brown; +Cc: git
In-Reply-To: <20060420200849.GA3653@tumblerings.org>
Zack Brown <zbrown@tumblerings.org> writes:
> Not true. I went into the git source directory, and ran "make". Nothing more.
Ah, I misunderstood. You are trying to run it _without_
installing it.
Well, then probably you do not have templates installed
anywhere, especially not where git-init-db expects them to be
found.
^ permalink raw reply
* Re: cg-clone produces "___" file and no working tree
From: Petr Baudis @ 2006-04-20 20:19 UTC (permalink / raw)
To: Junio C Hamano; +Cc: Zack Brown, git
In-Reply-To: <7vslo8xaql.fsf@assigned-by-dhcp.cox.net>
Dear diary, on Thu, Apr 20, 2006 at 10:17:38PM CEST, I got a letter
where Junio C Hamano <junkio@cox.net> said that...
> Zack Brown <zbrown@tumblerings.org> writes:
>
> > Not true. I went into the git source directory, and ran "make". Nothing more.
>
> Ah, I misunderstood. You are trying to run it _without_
> installing it.
>
> Well, then probably you do not have templates installed
> anywhere, especially not where git-init-db expects them to be
> found.
Duh, but shouldn't git-init-db create .git/info at any rate, even when
no templates are installed?
--
Petr "Pasky" Baudis
Stuff: http://pasky.or.cz/
Right now I am having amnesia and deja-vu at the same time. I think
I have forgotten this before.
^ permalink raw reply
* Re: cg-clone produces "___" file and no working tree
From: Junio C Hamano @ 2006-04-20 20:23 UTC (permalink / raw)
To: Zack Brown; +Cc: git
In-Reply-To: <7vslo8xaql.fsf@assigned-by-dhcp.cox.net>
Junio C Hamano <junkio@cox.net> writes:
> Zack Brown <zbrown@tumblerings.org> writes:
>
>> Not true. I went into the git source directory, and ran "make". Nothing more.
>
> Ah, I misunderstood. You are trying to run it _without_
> installing it.
>
> Well, then probably you do not have templates installed
> anywhere, especially not where git-init-db expects them to be
> found.
(sorry for the short message sent unfinished by mistake).
Running things without installing is somewhat tricky, but test
framework needs to do that, so there are some things you would
need to do.
- "git init-db" takes --template argument; in the source area
before installing, they are built in templates/blt/.
- "git" and programs that need to invoke other git programs
(e.g. git-send-pack) expects things to be found in gitexecdir
you set when you build. If you are not installing, you need
to override that with GIT_EXEC_PATH environment variable.
There might be other things, but you should be able to find them
from what t/Makefile and t/test-lib.sh do.
Having said that, honestly I would not recommend "runnning
without installing" unless you know what you are doing ;-).
^ permalink raw reply
* Re: cg-clone produces "___" file and no working tree
From: Junio C Hamano @ 2006-04-20 20:26 UTC (permalink / raw)
To: Petr Baudis; +Cc: git
In-Reply-To: <20060420201915.GF27689@pasky.or.cz>
Petr Baudis <pasky@suse.cz> writes:
> Duh, but shouldn't git-init-db create .git/info at any rate, even when
> no templates are installed?
I do not think so. We tend to lazily create necessary
directories under .git/ these days, and absolute minimum git
should not need an empty .git/info directory.
If there is something that creates files in .git/info without
making sure that leading path exists, we should fix it (maybe
update-server-info forgets it? I haven't checked).
^ permalink raw reply
* Re: cg-clone produces "___" file and no working tree
From: Petr Baudis @ 2006-04-20 20:35 UTC (permalink / raw)
To: Junio C Hamano; +Cc: git
In-Reply-To: <7vk69kxabp.fsf@assigned-by-dhcp.cox.net>
Dear diary, on Thu, Apr 20, 2006 at 10:26:34PM CEST, I got a letter
where Junio C Hamano <junkio@cox.net> said that...
> Petr Baudis <pasky@suse.cz> writes:
>
> > Duh, but shouldn't git-init-db create .git/info at any rate, even when
> > no templates are installed?
>
> I do not think so. We tend to lazily create necessary
> directories under .git/ these days, and absolute minimum git
> should not need an empty .git/info directory.
>
> If there is something that creates files in .git/info without
> making sure that leading path exists, we should fix it (maybe
> update-server-info forgets it? I haven't checked).
Aww. cg-clone assumed that .git/info is canonical part of git repository
now and git-init-db will always creat eit, but now it seems to be the
case only for .git/objects/info.
I've fixed cg-clone. Thanks for the info.
--
Petr "Pasky" Baudis
Stuff: http://pasky.or.cz/
Right now I am having amnesia and deja-vu at the same time. I think
I have forgotten this before.
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Nicolas Pitre @ 2006-04-20 21:02 UTC (permalink / raw)
To: Junio C Hamano; +Cc: git
In-Reply-To: <7v8xq0yteb.fsf@assigned-by-dhcp.cox.net>
On Thu, 20 Apr 2006, Junio C Hamano wrote:
> Nicolas Pitre <nico@cam.org> writes:
>
> > On Thu, 20 Apr 2006, Shawn Pearce wrote:
> >
> >> The more that I think about it the more it seems possible that the
> >> pathname hashing is what may be causing the problem. Not only did
> >> bisect point to 1d6b38cc76c348e2477506ca9759fc241e3d0d46 but the
> >> directory which contains the bulk of the space has many files with
> >> the same name located in different directories:
> > [...]
> >
> > But the bad commit according to your bisection talks about "thin" packs
> > which are not involved in your case. So something looks fishy with that
> > commit which should not have touched path hashing in the non-thin pack
> > case... I think...
>
> I think this explains it. The new code hashes full-path, but
> places bins for the paths with the same basename next to each
> other, so before Makefile and doc/Makefile and t/Makefile were
> all in the same bin, but now they are in three different bins
> next to each other.
That is fine. In fact I did try with a tweaked name_hash() that
completely ignored all directory components and the resulting pack was
even bigger, much bigger, when repacking Shawn's repo.
> I originally thought, with one single notable exception of
> Makefile, having the identically named file in many different
> directories is not common nor sane,
I'd tend to disagree with that but...
> and the new code favors to
> delta with the exact same path for deeper history over wasting
> delta window for making delta with objects with the same name in
> different places in more recent history. I think I benched this
> with kernel repository (git.git was too small for that).
This is obviously fine. And if a file in a given directory has few
revisions then the delta window will consider objects for a file with
the same name in other directories as well, which is also sensible. So
if files of the same name are located in different directories they
should delta well against each other if they're similar enough. This
should cover Shawn's repo layout.
> But I suspect we have a built-in "we sort bigger to smaller, and
> we cut off when we switch bins" somewhere in find_delta() loop,
> which I do not recall touching when I did that change, so that
> may be interfering and preventing 0-11-AdjLite.deg from all over
> the place to delta against each other.
I just cannot find something that would do that in the code. When
--no-reuse-delta is specified, the only things that will break the loop
in find_delta() is when try_delta() returns -1, and that happens only
when changing object type or when the size difference is too big, but
nothing looks at the name hash.
It is also hard to corelate it with commit 1d6b38cc which is the one
that introduced the regression.
Nicolas
^ permalink raw reply
* [PATCH] fix pack-object buffer size
From: Nicolas Pitre @ 2006-04-20 21:25 UTC (permalink / raw)
To: git
The input line has 40 _chars_ of sha1 and no 20 _bytes_. It should also
account for the space before the pathname, and the terminating \n and \0.
Signed-off-by: Nicolas Pitre <nico@cam.org>
---
I doubt anyone has ever used a repository with paths long enough to hit
the limit, but better make it right nevertheless.
diff --git a/pack-objects.c b/pack-objects.c
index 09f4f2c..3c2767b 100644
--- a/pack-objects.c
+++ b/pack-objects.c
@@ -1231,7 +1231,7 @@ static void setup_progress_signal(void)
int main(int argc, char **argv)
{
SHA_CTX ctx;
- char line[PATH_MAX + 20];
+ char line[40 + 1 + PATH_MAX + 2];
int window = 10, depth = 10, pack_to_stdout = 0;
struct object_entry **list;
int num_preferred_base = 0;
^ permalink raw reply related
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Junio C Hamano @ 2006-04-20 21:31 UTC (permalink / raw)
To: Shawn Pearce; +Cc: git, Nicolas Pitre, Linus Torvalds
In-Reply-To: <20060420173131.GF31738@spearce.org>
Shawn Pearce <spearce@spearce.org> writes:
> I just spent some time bisecting this issue and it looks like the
> following change by Junio may be the culprit:
>
> commit 1d6b38cc76c348e2477506ca9759fc241e3d0d46
> Author: Junio C Hamano <junkio@cox.net>
> Date: Wed Feb 22 22:10:24 2006 -0800
>
> pack-objects: use full pathname to help hashing with "thin" pack.
>
> This uses the same hashing algorithm to the "preferred base
> tree" objects and the incoming pathnames, to group the same
> files from different revs together, while spreading files with
> the same basename in different directories.
>
> Signed-off-by: Junio C Hamano <junkio@cox.net>
>
Unfortunately, that is not the same hash we use in v1.3.0, so we
need to look elsewhere for interactions.
v1.2.3 hash was base-name only. doc/Makefile and t/Makefile
were thrown in the same bin and sorted by size. When the
history you are packing is deep, and doc/Makefile and t/Makefile
are not related at all, this made effective size of delta window
1/N where N is the number of such duplicates.
The one you found above uses a hash that is fully full-path.
The two are in completely different bins, and bins are totally
random. This was not a good strategy.
v1.3.0 hash is base-name hash concatenated with leading-path
has. t/Makefile and doc/Makefile go in separate bins, but the
bins are close to each other; this avoids the problem in v1.2.3
when you have deep history, but at the same time if you do not
have many many versions of t/Makefile to overflow the delta
window, it gives t/Makefile a chance to delta with doc/Makefile.
The earlier observation by Linus on reverting eeef7135 is
consistent with it; that commit was the one that introduced
v1.3.0 hash.
You could try this patch to resurrect the hash used in v1.2.3,
and you may get better packing for your particular repository;
but I am not sure if it gives better results in the general
case. I am running the test myself now while waiting for my
day-job database to load X-<.
NOTE NOTE NOTE. The hash in v1.2.3 was done with the basename
(relying on rev-list --objects to only show the basename) and
hashed from front to back. The current one uses the same hash
scrambling function, but it hashes from back to front, and it
knows rev-list --objects gives it a full path.
What this patch does is to stop the hashing after we are done
with the basename part. So it still gives different hash value
to the same path from v1.2.3 version, but the distribution
should be equivalent.
NOTE 2. Feeding output from the current "rev-list --objects" to
v1.2.3 pack-object is the same as "hash full path and spread
things out" intermediate version, which is the worst performer.
-- >8 --
git diff
diff --git a/pack-objects.c b/pack-objects.c
index 09f4f2c..e58e169 100644
--- a/pack-objects.c
+++ b/pack-objects.c
@@ -492,6 +492,8 @@ static unsigned name_hash(struct name_pa
name_hash = hash;
hash = 0;
}
+ return name_hash;
+
for (p = path; p; p = p->up) {
hash = hash * 11 + '/';
n = p->elem + p->len;
^ permalink raw reply related
* Re: cg-clone produces "___" file and no working tree
From: Zack Brown @ 2006-04-20 21:34 UTC (permalink / raw)
To: Junio C Hamano; +Cc: git
In-Reply-To: <7vodywxago.fsf@assigned-by-dhcp.cox.net>
Hi,
> honestly I would not recommend "runnning
> without installing" unless you know what you are doing ;-).
OK, you're right, the problem was that I was not doing a proper install. I
followed the directions and it worked. Thanks!
Be well,
Zack
On Thu, Apr 20, 2006 at 01:23:35PM -0700, Junio C Hamano wrote:
> Junio C Hamano <junkio@cox.net> writes:
>
> > Zack Brown <zbrown@tumblerings.org> writes:
> >
> >> Not true. I went into the git source directory, and ran "make". Nothing more.
> >
> > Ah, I misunderstood. You are trying to run it _without_
> > installing it.
> >
> > Well, then probably you do not have templates installed
> > anywhere, especially not where git-init-db expects them to be
> > found.
>
> (sorry for the short message sent unfinished by mistake).
>
> Running things without installing is somewhat tricky, but test
> framework needs to do that, so there are some things you would
> need to do.
>
> - "git init-db" takes --template argument; in the source area
> before installing, they are built in templates/blt/.
>
> - "git" and programs that need to invoke other git programs
> (e.g. git-send-pack) expects things to be found in gitexecdir
> you set when you build. If you are not installing, you need
> to override that with GIT_EXEC_PATH environment variable.
>
> There might be other things, but you should be able to find them
> from what t/Makefile and t/test-lib.sh do.
>
> Having said that, honestly I would not recommend "runnning
> without installing" unless you know what you are doing ;-).
>
--
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Junio C Hamano @ 2006-04-20 21:40 UTC (permalink / raw)
To: Nicolas Pitre; +Cc: git
In-Reply-To: <Pine.LNX.4.64.0604201630320.2215@localhost.localdomain>
Nicolas Pitre <nico@cam.org> writes:
>> But I suspect we have a built-in "we sort bigger to smaller, and
>> we cut off when we switch bins" somewhere in find_delta() loop,
>> which I do not recall touching when I did that change, so that
>> may be interfering and preventing 0-11-AdjLite.deg from all over
>> the place to delta against each other.
>
> I just cannot find something that would do that in the code. When
> --no-reuse-delta is specified, the only things that will break the loop
> in find_delta() is when try_delta() returns -1, and that happens only
> when changing object type or when the size difference is too big, but
> nothing looks at the name hash.
The list is sorted by type then hash then size (type_size_sort),
so if you have t/Makefile that are big medium small too-small
and then doc/Makefile that are big medium, once you see the
too-small t/Makefile it would not consider the big doc/Makefile
as a candidate X-<.
This _might_ improve things:
diff --git a/pack-objects.c b/pack-objects.c
index 09f4f2c..0c6abe9 100644
--- a/pack-objects.c
+++ b/pack-objects.c
@@ -1037,7 +1039,7 @@ static int try_delta(struct unpacked *cu
sizediff = oldsize > size ? oldsize - size : size - oldsize;
if (size < 50)
- return -1;
+ return 0;
if (old_entry->depth >= max_depth)
return 0;
@@ -1052,7 +1054,7 @@ static int try_delta(struct unpacked *cu
if (cur_entry->delta)
max_size = cur_entry->delta_size-1;
if (sizediff >= max_size)
- return -1;
+ return 0;
delta_buf = diff_delta(old->data, oldsize,
cur->data, size, &delta_size, max_size);
if (!delta_buf)
^ permalink raw reply related
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Shawn Pearce @ 2006-04-20 21:53 UTC (permalink / raw)
To: Junio C Hamano; +Cc: git, Nicolas Pitre, Linus Torvalds
In-Reply-To: <7vodywvsrq.fsf@assigned-by-dhcp.cox.net>
Junio C Hamano <junkio@cox.net> wrote:
> Shawn Pearce <spearce@spearce.org> writes:
>
> > I just spent some time bisecting this issue and it looks like the
> > following change by Junio may be the culprit:
> >
> > commit 1d6b38cc76c348e2477506ca9759fc241e3d0d46
[snip]
> Unfortunately, that is not the same hash we use in v1.3.0, so we
> need to look elsewhere for interactions.
Pity. Then either bisect goofed or there was a goof in meatspace
while using bisect. I honestly expected bisect to point at the
problem commit. I tried reverting 1d6b38cc but it didn't apply
cleanly and I didn't feel like working through all of the conflicts
at the time.
[snip]
> The earlier observation by Linus on reverting eeef7135 is
> consistent with it; that commit was the one that introduced
> v1.3.0 hash.
Yet reverting that didn't help either.
[snip]
> You could try this patch to resurrect the hash used in v1.2.3,
> and you may get better packing for your particular repository;
> but I am not sure if it gives better results in the general
> case. I am running the test myself now while waiting for my
> day-job database to load X-<.
[snip]
Nope. When applied to 'next' it didn't help very much:
Total 46391, written 46391 (delta 6466), reused 38662 (delta 0)
118M pack-7f766f5af5547554bacb28c0294bd562589dc5e7.pack
Just to note: the 1.3.0 packer is saving 1M in the GIT repository
over the 1.2.3 packer. So for a real project it does seem to have
some benefit. And if you benchmarked the 1.3.0 packer against
the Linux kernel and found it to be better than the 1.2.3 packer
that's even better.
I think this repository of mine may just be a degenerate case which
GIT doesn't pack very well. GIT can't be all things to all people!
--
Shawn.
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Jakub Narebski @ 2006-04-20 21:56 UTC (permalink / raw)
To: git
In-Reply-To: <7vodywvsrq.fsf@assigned-by-dhcp.cox.net>
Junio C Hamano wrote:
> v1.2.3 hash was base-name only. doc/Makefile and t/Makefile
> were thrown in the same bin and sorted by size. When the
> history you are packing is deep, and doc/Makefile and t/Makefile
> are not related at all, this made effective size of delta window
> 1/N where N is the number of such duplicates.
>
> The one you found above uses a hash that is fully full-path.
> The two are in completely different bins, and bins are totally
> random. This was not a good strategy.
>
> v1.3.0 hash is base-name hash concatenated with leading-path
> has. t/Makefile and doc/Makefile go in separate bins, but the
> bins are close to each other; this avoids the problem in v1.2.3
> when you have deep history, but at the same time if you do not
> have many many versions of t/Makefile to overflow the delta
> window, it gives t/Makefile a chance to delta with doc/Makefile.
[...]
> You could try this patch to resurrect the hash used in v1.2.3,
> and you may get better packing for your particular repository;
> but I am not sure if it gives better results in the general
> case. I am running the test myself now while waiting for my
> day-job database to load X-<.
Perhaps the packing code could check which version gives smaller pack, or at
least be instructed that one might want different packing heuristic for
specific repository? Surely 2x difference in size is worth considering (and
complication)...
--
Jakub Narebski
Warsaw, Poland
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Shawn Pearce @ 2006-04-20 22:02 UTC (permalink / raw)
To: Junio C Hamano; +Cc: Nicolas Pitre, git
In-Reply-To: <7vfyk8vscl.fsf@assigned-by-dhcp.cox.net>
Junio C Hamano <junkio@cox.net> wrote:
> Nicolas Pitre <nico@cam.org> writes:
>
> >> But I suspect we have a built-in "we sort bigger to smaller, and
> >> we cut off when we switch bins" somewhere in find_delta() loop,
> >> which I do not recall touching when I did that change, so that
> >> may be interfering and preventing 0-11-AdjLite.deg from all over
> >> the place to delta against each other.
> >
> > I just cannot find something that would do that in the code. When
> > --no-reuse-delta is specified, the only things that will break the loop
> > in find_delta() is when try_delta() returns -1, and that happens only
> > when changing object type or when the size difference is too big, but
> > nothing looks at the name hash.
>
> The list is sorted by type then hash then size (type_size_sort),
> so if you have t/Makefile that are big medium small too-small
> and then doc/Makefile that are big medium, once you see the
> too-small t/Makefile it would not consider the big doc/Makefile
> as a candidate X-<.
>
> This _might_ improve things:
>
> diff --git a/pack-objects.c b/pack-objects.c
> index 09f4f2c..0c6abe9 100644
> --- a/pack-objects.c
> +++ b/pack-objects.c
> @@ -1037,7 +1039,7 @@ static int try_delta(struct unpacked *cu
> sizediff = oldsize > size ? oldsize - size : size - oldsize;
>
> if (size < 50)
> - return -1;
> + return 0;
> if (old_entry->depth >= max_depth)
> return 0;
>
> @@ -1052,7 +1054,7 @@ static int try_delta(struct unpacked *cu
> if (cur_entry->delta)
> max_size = cur_entry->delta_size-1;
> if (sizediff >= max_size)
> - return -1;
> + return 0;
> delta_buf = diff_delta(old->data, oldsize,
> cur->data, size, &delta_size, max_size);
> if (!delta_buf)
Holy cow, it did:
Total 46391, written 46391 (delta 8391), reused 37774 (delta 0)
46M pack-7f766f5af5547554bacb28c0294bd562589dc5e7.pack
That's the smallest packing I've seen yet. And it doesn't have a
negative affect on repacking GIT either.
--
Shawn.
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Junio C Hamano @ 2006-04-20 22:35 UTC (permalink / raw)
To: Shawn Pearce; +Cc: Nicolas Pitre, git
In-Reply-To: <20060420220240.GB32748@spearce.org>
Shawn Pearce <spearce@spearce.org> writes:
> Junio C Hamano <junkio@cox.net> wrote:
>
>> The list is sorted by type then hash then size (type_size_sort),
>> so if you have t/Makefile that are big medium small too-small
>> and then doc/Makefile that are big medium, once you see the
>> too-small t/Makefile it would not consider the big doc/Makefile
>> as a candidate X-<.
>>
>> This _might_ improve things:
>>...
>
> Holy cow, it did:
>
> Total 46391, written 46391 (delta 8391), reused 37774 (delta 0)
> 46M pack-7f766f5af5547554bacb28c0294bd562589dc5e7.pack
>
> That's the smallest packing I've seen yet. And it doesn't have a
> negative affect on repacking GIT either.
Thanks for trying. Mind trying one more?
I suspect the test patch makes pack-objects a lot more
expensive.
The code before the test patch said "if the size is very small
or size difference is too great, do not consider this, and do
not consider any more objects in the delta window, because we
know they are either even smaller of the same path, they have
different names, or they are of different type". The test patch
you tried was a quick and dirty hack that said "under the
too-small condition, skip this one, but keep trying the rest of
the delta window".
Here is a cleaned up patch. What it does is "under the
too-small condition, see if the object has the same basename,
and if so keep going, but otherwise skip the rest as before".
If you have objects like this and are trying to pack the first
object (this list is sorted in the order pack-object tries):
(size) (path)
1000 t/0-11-AdjLite.deg
10 t/0-11-AdjLite.deg
800 s/0-11-AdjLite.deg
20 t/0-12-AdjLite.deg
the current code stops after checking t/0-11-AdjLite.deg. The
test patch tries all of them. This patch skips that file, but
tries "s/0-11-AdjLite.deg", and then stops at the next one.
-- >8 --
diff --git a/pack-objects.c b/pack-objects.c
index 09f4f2c..2173709 100644
--- a/pack-objects.c
+++ b/pack-objects.c
@@ -1036,8 +1036,6 @@ static int try_delta(struct unpacked *cu
oldsize = old_entry->size;
sizediff = oldsize > size ? oldsize - size : size - oldsize;
- if (size < 50)
- return -1;
if (old_entry->depth >= max_depth)
return 0;
@@ -1048,20 +1046,27 @@ static int try_delta(struct unpacked *cu
* more space-efficient (deletes don't have to say _what_ they
* delete).
*/
- max_size = size / 2 - 20;
- if (cur_entry->delta)
- max_size = cur_entry->delta_size-1;
- if (sizediff >= max_size)
- return -1;
- delta_buf = diff_delta(old->data, oldsize,
- cur->data, size, &delta_size, max_size);
- if (!delta_buf)
+ if (50 <= size) {
+ max_size = size / 2 - 20;
+ if (cur_entry->delta)
+ max_size = cur_entry->delta_size-1;
+ if (sizediff < max_size) {
+ delta_buf = diff_delta(old->data, oldsize,
+ cur->data, size,
+ &delta_size, max_size);
+ if (!delta_buf)
+ return 0;
+ cur_entry->delta = old_entry;
+ cur_entry->delta_size = delta_size;
+ cur_entry->depth = old_entry->depth + 1;
+ free(delta_buf);
+ return 0;
+ }
+ }
+ /* Keep going as long as the basename matches */
+ if (((cur_entry->hash ^ old_entry->hash) >>DIRBITS) == 0)
return 0;
- cur_entry->delta = old_entry;
- cur_entry->delta_size = delta_size;
- cur_entry->depth = old_entry->depth + 1;
- free(delta_buf);
- return 0;
+ return -1;
}
static void progress_interval(int signum)
^ permalink raw reply related
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Linus Torvalds @ 2006-04-20 22:59 UTC (permalink / raw)
To: Shawn Pearce; +Cc: Junio C Hamano, Nicolas Pitre, git
In-Reply-To: <20060420220240.GB32748@spearce.org>
On Thu, 20 Apr 2006, Shawn Pearce wrote:
> Junio C Hamano <junkio@cox.net> wrote:
> >
> > This _might_ improve things:
> >
> > diff --git a/pack-objects.c b/pack-objects.c
> > index 09f4f2c..0c6abe9 100644
> > --- a/pack-objects.c
> > +++ b/pack-objects.c
> > @@ -1037,7 +1039,7 @@ static int try_delta(struct unpacked *cu
> > sizediff = oldsize > size ? oldsize - size : size - oldsize;
> >
> > if (size < 50)
> > - return -1;
> > + return 0;
> > if (old_entry->depth >= max_depth)
> > return 0;
> >
> > @@ -1052,7 +1054,7 @@ static int try_delta(struct unpacked *cu
> > if (cur_entry->delta)
> > max_size = cur_entry->delta_size-1;
> > if (sizediff >= max_size)
> > - return -1;
> > + return 0;
> > delta_buf = diff_delta(old->data, oldsize,
> > cur->data, size, &delta_size, max_size);
> > if (!delta_buf)
>
> Holy cow, it did:
>
> Total 46391, written 46391 (delta 8391), reused 37774 (delta 0)
> 46M pack-7f766f5af5547554bacb28c0294bd562589dc5e7.pack
>
> That's the smallest packing I've seen yet. And it doesn't have a
> negative affect on repacking GIT either.
I think I know what's going on, and why your bisection claimed it was the
re-hashing change that was the problem, even though it really wasn't.
That
if (sizediff >= max_size)
return -1;
check is actually fairly _old_. It's from June 2005, ie it's from pretty
much the first two days of that packing thing existing in the first place.
(The initial repacking was done June 25th, with a lot of tweaking over the
next few days. That sizediff thing was part of the fairly early tweaking).
The thing is, that check made sense back then. Why? Because we sorted
things in decreasing size order back then (I think this was before we even
did any name-based heuristic sorting at all), so that when we tried the
delta algorithm, and the size diff was bigger than the last delta size, we
pretty much _knew_ the new delta would be bigger still, and there was no
point in continuing with try_delta.
HOWEVER. We have since changed the sorting to sort according to name
before it sorts according to size, so that old heuristic that depended on
the size being monotonically increasing simply doesn't make any sense any
more.
So I think at that second "return -1" really _should_ be changed to a
"return 0", and not just because it helps your particular case. It's
literally a bug these days, because the assumptions that caused it to
return -1 simply aren't true any more.
(It wasn't _strictly_ true even originally: even if the sizediff is huge,
the _delta_ may not be huge, since we can delete data with a small delta.
So it's quite likely that we should compare the "old is bigger than new"
and "new is bigger than old" separately and have different heuristics for
them. Again, that was simply not much of an issue back when we sorted just
by size).
So even the "return 0" might not be completely right. We might actually
want to look at how big the delta is, and return only once that fails.
Linus
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Junio C Hamano @ 2006-04-20 23:02 UTC (permalink / raw)
To: Nicolas Pitre; +Cc: git
In-Reply-To: <Pine.LNX.4.64.0604201630320.2215@localhost.localdomain>
Nicolas Pitre <nico@cam.org> writes:
>> I originally thought, with one single notable exception of
>> Makefile, having the identically named file in many different
>> directories is not common nor sane,
>
> I'd tend to disagree with that but...
I disagree with that myself now. The kernel tree has many
files with the same basename (e.g. arch/*/kernel/irq.c).
It is a different issue if they are good delta base candidates
with each other, though.
^ permalink raw reply
* Cogito bug on Debian
From: Martin Langhoff @ 2006-04-20 23:17 UTC (permalink / raw)
To: Git Mailing List, Petr Baudis
This was spotted circulating on Catalyst's IRC channel. Apparently,
the bug "causes non-serious data loss".
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=330031
cheers,
martin
^ permalink raw reply
* [cogito-0.17.2] Bug in cg-log selection
From: Blaisorblade @ 2006-04-21 0:08 UTC (permalink / raw)
To: Petr Baudis; +Cc: git
(Please CC me on replies as I'm not subscribed).
On a standard Linux tree:
$ cg-branch-ls
origin
git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-2.6.16.y.git
this command selects two commit instead than the right one (which is selected
by git-whatchanged):
$ cg-log -r v2.6.16:v2.6.16.9 kernel/power/process.c
shows commits 6b2467e45179a336f1e5b70d2b2ae1fe89a00133 and
1dd6f008de5a04251d9cbe4c1cf67e4c708f9fe9, but the latter doesn't touch that
file, and
$ git-whatchanged -p v2.6.16..v2.6.16.9 kernel/power/process.c
only shows the first commit.
Verbatim output of short listing:
$ cg-log -c -s -r v2.6.16:v2.6.16.9 kernel/power/process.c|cat
6b2467e45179a336f1e5b70d2b2ae1fe89a00133 Pavel Machek 2006-04-17 13:16
[PATCH] Fix suspend with traced tasks
1dd6f008de5a04251d9cbe4c1cf67e4c708f9fe9 Jeff Garzik 2006-03-27 22:47
[PATCH] sata_mv: fix irq port status usage
Bye
--
Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!".
Paolo Giarrusso, aka Blaisorblade (Skype ID "PaoloGiarrusso", ICQ 215621894)
http://www.user-mode-linux.org/~blaisorblade
___________________________________
Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB
http://mail.yahoo.it
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Nicolas Pitre @ 2006-04-21 0:52 UTC (permalink / raw)
To: Junio C Hamano; +Cc: git
In-Reply-To: <7vfyk8vscl.fsf@assigned-by-dhcp.cox.net>
On Thu, 20 Apr 2006, Junio C Hamano wrote:
> Nicolas Pitre <nico@cam.org> writes:
>
> >> But I suspect we have a built-in "we sort bigger to smaller, and
> >> we cut off when we switch bins" somewhere in find_delta() loop,
> >> which I do not recall touching when I did that change, so that
> >> may be interfering and preventing 0-11-AdjLite.deg from all over
> >> the place to delta against each other.
> >
> > I just cannot find something that would do that in the code. When
> > --no-reuse-delta is specified, the only things that will break the loop
> > in find_delta() is when try_delta() returns -1, and that happens only
> > when changing object type or when the size difference is too big, but
> > nothing looks at the name hash.
>
> The list is sorted by type then hash then size (type_size_sort),
> so if you have t/Makefile that are big medium small too-small
> and then doc/Makefile that are big medium, once you see the
> too-small t/Makefile it would not consider the big doc/Makefile
> as a candidate X-<.
Bingo! I didn't think it all through before.
Nicolas
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Shawn Pearce @ 2006-04-21 1:01 UTC (permalink / raw)
To: Junio C Hamano; +Cc: Nicolas Pitre, git
In-Reply-To: <7vy7xzvpsg.fsf@assigned-by-dhcp.cox.net>
Junio C Hamano <junkio@cox.net> wrote:
[snip]
> I suspect the test patch makes pack-objects a lot more
> expensive.
Which patch are you talking about the previous patch or the one in
the message I'm now replying to?
> The code before the test patch said "if the size is very small
> or size difference is too great, do not consider this, and do
> not consider any more objects in the delta window, because we
> know they are either even smaller of the same path, they have
> different names, or they are of different type". The test patch
> you tried was a quick and dirty hack that said "under the
> too-small condition, skip this one, but keep trying the rest of
> the delta window".
>
> Here is a cleaned up patch. What it does is "under the
> too-small condition, see if the object has the same basename,
> and if so keep going, but otherwise skip the rest as before".
[snip]
The patch below does not help very much:
Total 46391, written 46391 (delta 6686), reused 37979 (delta 0)
129M pack-7f766f5af5547554bacb28c0294bd562589dc5e7.pack
> diff --git a/pack-objects.c b/pack-objects.c
> index 09f4f2c..2173709 100644
> --- a/pack-objects.c
> +++ b/pack-objects.c
> @@ -1036,8 +1036,6 @@ static int try_delta(struct unpacked *cu
> oldsize = old_entry->size;
> sizediff = oldsize > size ? oldsize - size : size - oldsize;
>
> - if (size < 50)
> - return -1;
> if (old_entry->depth >= max_depth)
> return 0;
>
> @@ -1048,20 +1046,27 @@ static int try_delta(struct unpacked *cu
> * more space-efficient (deletes don't have to say _what_ they
> * delete).
> */
> - max_size = size / 2 - 20;
> - if (cur_entry->delta)
> - max_size = cur_entry->delta_size-1;
> - if (sizediff >= max_size)
> - return -1;
> - delta_buf = diff_delta(old->data, oldsize,
> - cur->data, size, &delta_size, max_size);
> - if (!delta_buf)
> + if (50 <= size) {
> + max_size = size / 2 - 20;
> + if (cur_entry->delta)
> + max_size = cur_entry->delta_size-1;
> + if (sizediff < max_size) {
> + delta_buf = diff_delta(old->data, oldsize,
> + cur->data, size,
> + &delta_size, max_size);
> + if (!delta_buf)
> + return 0;
> + cur_entry->delta = old_entry;
> + cur_entry->delta_size = delta_size;
> + cur_entry->depth = old_entry->depth + 1;
> + free(delta_buf);
> + return 0;
> + }
> + }
> + /* Keep going as long as the basename matches */
> + if (((cur_entry->hash ^ old_entry->hash) >>DIRBITS) == 0)
> return 0;
> - cur_entry->delta = old_entry;
> - cur_entry->delta_size = delta_size;
> - cur_entry->depth = old_entry->depth + 1;
> - free(delta_buf);
> - return 0;
> + return -1;
> }
>
> static void progress_interval(int signum)
>
--
Shawn.
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Shawn Pearce @ 2006-04-21 1:20 UTC (permalink / raw)
To: Junio C Hamano; +Cc: Nicolas Pitre, git
In-Reply-To: <7vfyk8vscl.fsf@assigned-by-dhcp.cox.net>
Based on Linus' comment I changed your patch to just the following.
It still produced the 46M pack file, so the first hunk apears to
not have had much of an affect with this data.
>From a running time perspective it appears as though this patch is
making things slightly better, not worse. I ran it a few times
for each case always using the 46M pack as input for
"git-repack -a -d -f".
'next' 137.13 real 95.82 user 25.24 sys
'next'+patch 131.62 real 89.35 user 28.56 sys
but even if the running time was an extra 6 seconds I'd still rather
spend 4% more running time to use 1/2 the storage space.
diff --git a/pack-objects.c b/pack-objects.c
index 09f4f2c..f7d6217 100644
--- a/pack-objects.c
+++ b/pack-objects.c
@@ -1052,7 +1052,7 @@ static int try_delta(struct unpacked *cu
if (cur_entry->delta)
max_size = cur_entry->delta_size-1;
if (sizediff >= max_size)
- return -1;
+ return 0;
delta_buf = diff_delta(old->data, oldsize,
cur->data, size, &delta_size, max_size);
if (!delta_buf)
^ permalink raw reply related
* PATCH RESEND: git-svnimport memory leak plug
From: Auke Kok @ 2006-04-21 2:12 UTC (permalink / raw)
To: git
I was unable with todays git-svnimport to convert a modest 140mb svn repository to git using svn-git-import. the process would bomb out after 2000 revisions and I have 20000. It consumed over 1.5gb vm space on my 1gb machine, and died with a 'cannot fork error'. this also killed my project server today after I discivered it the wrong way.
The original patch send earlier by Santi Bejar fixes the problem. I include the patch again here so it can be merged. my mailer probably will nuke it so here's the link to the archive post:
http://marc.theaimsgroup.com/?l=git&m=114345884526971&w=2
Please include this patch.
Cheers,
Auke
On 3/24/06, Santi Béjar <sbejar@gmail.com> wrote:
> Jan-Benedict Glaw <jbglaw@lug-owl.de> writes:
>
> > On Wed, 2006-03-22 14:33:37 +0100, Jan-Benedict Glaw <jbglaw@lug-owl.de> wrote:
> >
> > Since it seems nobody looked at the GCC import run (which means to use
> > the svnimport), I ran it again, under strace control:
> >
> >> GCC
> >> ~~~
> >> $ /home/jbglaw/bin/git svnimport -C gcc -v svn://gcc.gnu.org/svn/gcc
> >
> >> Committed change 3936:/ 1993-03-31 05:44:03)
> >> Commit ID ceff85145f8671fb2a9d826a761cedc2a507bd1e
> >> Writing to refs/heads/origin
> >> DONE: 3936 origin ceff85145f8671fb2a9d826a761cedc2a507bd1e
> >> ... 3937 trunk/gcc/final.c ...
> >> Can't fork at /home/jbglaw/bin/git-svnimport line 379.
> >
>
> I have the same (?) problem with one of my svn repository. It worked
> before (I've redone the import with the -r flag), so I bisected it.
> The problematic commit seems to be:
>
> diff-tree 4802426... (from 525c0d7...)
> Author: Karl Hasselström <kha@treskal.com>
> Date: Sun Feb 26 06:11:27 2006 +0100
>
> svnimport: Convert executable flag
>
> Convert the svn:executable property to file mode 755 when converting
> an SVN repository to GIT.
>
> Signed-off-by: Karl Hasselström <kha@treskal.com>
> Signed-off-by: Junio C Hamano <junkio@cox.net>
>
> :100755 100755 ee2940f... 6603b96... M git-svnimport.perl
>
> I think it has a memory leak, it used up to 140m of memory.
>
> $ git reset --hard 4802426^
> $ time ../git-svnimport.perl file:///path/
> Use of uninitialized value in string eq at ../git-svnimport.perl line 463.
> Use of uninitialized value in substitution (s///) at ../git-svnimport.perl line 466.
> real 0m55.801s
> user 0m30.578s
> sys 0m23.084s
>
> $ git reset --hard 4802426
> $ time ../git-svnimport.perl file:///path/
> Use of uninitialized value in string eq at ../git-svnimport.perl line 463.
> Use of uninitialized value in substitution (s///) at ../git-svnimport.perl line 466.
> Can't fork at /home/santi/usr/src/scm/git/git-svnimport.perl line 331.
> real 6m2.163s
> user 0m20.332s
> sys 0m50.180s
>
> and it didn't finished. Hope it helps.
And this patch fixes my problems.
---
Introduced in 4802426.
Signed-off-by: Santi Béjar <sbejar@gmail.com>
---
git-svnimport.perl | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/git-svnimport.perl b/git-svnimport.perl
index 639aa41..f2cf062 100755
--- a/git-svnimport.perl
+++ b/git-svnimport.perl
@@ -135,8 +135,10 @@
print "... $rev $path ...\n" if $opt_v;
my (undef, $properties);
+ my $pool = SVN::Pool->new();
eval { (undef, $properties)
- = $self->{'svn'}->get_file($path,$rev,$fh); };
+ = $self->{'svn'}->get_file($path,$rev,$fh,$pool); };
+ $pool->clear;
if($@) {
return undef if $@ =~ /Attempted to get checksum/;
die $@;
^ permalink raw reply related
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Nicolas Pitre @ 2006-04-21 2:28 UTC (permalink / raw)
To: Shawn Pearce; +Cc: Junio C Hamano, git
In-Reply-To: <20060421012029.GB819@spearce.org>
On Thu, 20 Apr 2006, Shawn Pearce wrote:
> Based on Linus' comment I changed your patch to just the following.
> It still produced the 46M pack file, so the first hunk apears to
> not have had much of an affect with this data.
>
> From a running time perspective it appears as though this patch is
> making things slightly better, not worse. I ran it a few times
> for each case always using the 46M pack as input for
> "git-repack -a -d -f".
>
> 'next' 137.13 real 95.82 user 25.24 sys
> 'next'+patch 131.62 real 89.35 user 28.56 sys
>
> but even if the running time was an extra 6 seconds I'd still rather
> spend 4% more running time to use 1/2 the storage space.
>
>
> diff --git a/pack-objects.c b/pack-objects.c
> index 09f4f2c..f7d6217 100644
> --- a/pack-objects.c
> +++ b/pack-objects.c
> @@ -1052,7 +1052,7 @@ static int try_delta(struct unpacked *cu
> if (cur_entry->delta)
> max_size = cur_entry->delta_size-1;
> if (sizediff >= max_size)
> - return -1;
> + return 0;
> delta_buf = diff_delta(old->data, oldsize,
> cur->data, size, &delta_size, max_size);
> if (!delta_buf)
I can confirm this is indeed the best fix so far. Any "smarter"
solution I could think of did increase the size of the final pack quite
spectacularly and rather unexpectedly with Shawn's repository.
Of course removing the if (sizediff >= max_size) entirely does produce a
smaller pack (39MB) but it takes about twice the CPU.
With the patch above the Linux kernel pack is 0.3% smaller with 1% more
CPU usage. But like for the diff-delta hash list limiting code this
small overhead is certainly a good compromize to avoid big degradations
in some other cases.
Nicolas
^ permalink raw reply
* Re: 1.3.0 creating bigger packs than 1.2.3
From: Shawn Pearce @ 2006-04-21 2:32 UTC (permalink / raw)
To: Junio C Hamano; +Cc: Nicolas Pitre, git
In-Reply-To: <20060421012029.GB819@spearce.org>
I just tried the patch below on a couple-month-old Linux 2.6
repository from Linus (last commit: Feb 14 2006). It did not
decrease the pack file size by much despite the higher delta:
'next' Total 189435, written 189435 (delta 142093), reused 44057 (delta 0)
'next'+patch Total 189435, written 189435 (delta 142712), reused 43954 (delta 0)
'next' 104464297 bytes
'next'+patch 104092920 bytes (99.6% of 'next')
'next' 328.98 real 206.02 user 93.60 sys
'next'+patch 363.06 real 218.98 user 94.72 sys
So it looks like the patch is taking longer to run, and by about 10%.
An expensive price to pay for what amounts to only a 0.4% reduction
in pack size on the kernel.
Shawn Pearce <spearce@spearce.org> wrote:
> Based on Linus' comment I changed your patch to just the following.
> It still produced the 46M pack file, so the first hunk apears to
> not have had much of an affect with this data.
>
> From a running time perspective it appears as though this patch is
> making things slightly better, not worse. I ran it a few times
> for each case always using the 46M pack as input for
> "git-repack -a -d -f".
>
> 'next' 137.13 real 95.82 user 25.24 sys
> 'next'+patch 131.62 real 89.35 user 28.56 sys
>
> but even if the running time was an extra 6 seconds I'd still rather
> spend 4% more running time to use 1/2 the storage space.
>
>
> diff --git a/pack-objects.c b/pack-objects.c
> index 09f4f2c..f7d6217 100644
> --- a/pack-objects.c
> +++ b/pack-objects.c
> @@ -1052,7 +1052,7 @@ static int try_delta(struct unpacked *cu
> if (cur_entry->delta)
> max_size = cur_entry->delta_size-1;
> if (sizediff >= max_size)
> - return -1;
> + return 0;
> delta_buf = diff_delta(old->data, oldsize,
> cur->data, size, &delta_size, max_size);
> if (!delta_buf)
--
Shawn.
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox