Magit slow on Windows

Magit slow on Windows

[Bastian Bechtold]

Hi,

My installation of Magit is dog slow on WIndows. For example, I want to see the changes I made to some file in Magit. On Linux and OSX, this is as simple as putting point on top of the file name, then hit tab. Magit shows the diff instantly. On Windows, it takes about three seconds (on the first try. Subsequent tabs on the same file are fast). Same for any other action within Magit. 

Funny enough, doing the same thing through eshell is plenty fast.

As some friendly people on the internet pointed out, this apparently did not happen in earlier versions of Emacs/Magit.
(http://www.reddit.com/r/emacs/comments/tebll/magit_and_others_slow_on_windows/) 

I am using the latest official Emacs 24 pretest and the latest Magit from MELPA/github together with Git for Windows (aka msysgit afaik). I have a Cygwin installation, but `which git` in eshell claims that it is using the msysgit version.

Any hints or ideas would be greatly appreciated!

Basti

Re: Magit slow on Windows

[Eli Zaretskii]

> From: Bastian Bechtold <basti.bechtold@googlemail.com>
> Date: Thu, 10 May 2012 14:06:14 +0200
> 
> My installation of Magit is dog slow on WIndows. For example, I want to see the changes I made to some file in Magit. On Linux and OSX, this is as simple as putting point on top of the file name, then hit tab. Magit shows the diff instantly. On Windows, it takes about three seconds (on the first try. Subsequent tabs on the same file are fast). Same for any other action within Magit. 
> 
> Funny enough, doing the same thing through eshell is plenty fast.
> 
> As some friendly people on the internet pointed out, this apparently did not happen in earlier versions of Emacs/Magit.
> (http://www.reddit.com/r/emacs/comments/tebll/magit_and_others_slow_on_windows/) 
> 
> I am using the latest official Emacs 24 pretest and the latest Magit from MELPA/github together with Git for Windows (aka msysgit afaik). I have a Cygwin installation, but `which git` in eshell claims that it is using the msysgit version.

And what does 'M-: (executable-find "git")' tells you about which git
version is invoked by Emacs?

Re: Magit slow on Windows

[Kevin Yu]

Eli Zaretskii <eliz@gnu.org> writes:

>> My installation of Magit is dog slow on WIndows. For example, I want
>> to see the changes I made to some file in Magit. On Linux and OSX,
>> this is as simple as putting point on top of the file name, then hit
>> tab. Magit shows the diff instantly. On Windows, it takes about
>> three seconds (on the first try. Subsequent tabs on the same file
>> are fast). Same for any other action within Magit.

it's really slow on windows. :(

> And what does 'M-: (executable-find "git")' tells you about which git
> version is invoked by Emacs?

  c:/git/cmd/git.cmd


by the way. vc-git commit doesn't work on windows now.

Re: Magit slow on Windows

[Eli Zaretskii]

> From: Kevin Yu <yujie052@gmail.com>
> Date: Fri, 11 May 2012 13:35:03 +0800
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> > And what does 'M-: (executable-find "git")' tells you about which git
> > version is invoked by Emacs?
> 
>   c:/git/cmd/git.cmd

Which is what? msysgit or the Cygwin git?

Re: Magit slow on Windows

[Kevin Yu]

Eli Zaretskii <eliz@gnu.org> writes:

> Which is what? msysgit or the Cygwin git?

msysgit. 

Re: Magit slow on Windows

[Eli Zaretskii]

> From: Kevin Yu <yujie052@gmail.com>
> Date: Fri, 11 May 2012 13:46:53 +0800
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> > Which is what? msysgit or the Cygwin git?
> 
> msysgit. 

Then I think this is wrong: you should see

  c:/git/bin/git.exe

Invoking git through a Windows shell script incurs an unnecessary
slowdown.  You should start Emacs from the shell that knows about the
msysgit installation, to get correct behavior.

But I'm not sure this will solve your problem.  The Windows filesystem
is slower than Linux filesystems, especially with a cold cache, and
nothing, not even Emacs, can change that.

If indeed invoking git from eshell is much faster (and not an artifact
of your testing, since a warm cache makes a world of difference), then
I suggest to try to find out what does eshell do differently from
magit.

Re: Magit slow on Windows

[Kevin Yu]

>
> Then I think this is wrong: you should see
>
>   c:/git/bin/git.exe

If you do not want to export all the unix fileutils commands comming
with msys, the msys installation will wrap the git.exe command in
git.cmd, and put it in a seperated directory.

>
> Invoking git through a Windows shell script incurs an unnecessary
> slowdown.  You should start Emacs from the shell that knows about the
> msysgit installation, to get correct behavior.

here's the test from eshell:

    $ which git
    c:/git/cmd/git.cmd
    
    $ time git status
    0.156 secs

If you invoke M-x magit-status. It takes about 3 seconds to update the
status. I don't why :(.

Magit is really faster on Linux. I tested it in my virtual machine
(Debian 6.0) on the same repo(shared by virtual box shared
directory). It takes less than 1 second to show the repo status.

Re: Magit slow on Windows

[Yann Hodique]

>>>>> "Kevin" == Kevin Yu <yujie052@gmail.com> writes:

> here's the test from eshell:

>     $ which git
>     c:/git/cmd/git.cmd
    
>     $ time git status
>     0.156 secs

> If you invoke M-x magit-status. It takes about 3 seconds to update the
> status. I don't why :(.

Unfortunately this is no accurate comparison, since `magit-status'
invokes multiple git commands (and actually none of them is "git
status"). Some of them are calls to "git diff" and "git ls-files" that
can be pretty heavy and might be particularly vulnerable to the cold
cache effect.

That said, calling everything through the git.cmd wrapper will
definitely slow down things even further. According to "some guy on the
internet" (https://github.com/dahlbyk/posh-git/issues/21 :)) it has
a cost that might explain your observations (since `magit-status'
performs about 15 calls to git).
So I'd say tweaking your PATH would definitely be worth a try.

Yann.

-- 
Like many culinary delicacies, revenge is a dish best savored slowly,
after long and delicate preparation.

  -- EMPEROR ELROOD IX, Deathbed Insights

Re: Magit slow on Windows

[Kevin Yu]

Yann Hodique <yann.hodique@gmail.com> writes:

>>>>>> "Kevin" == Kevin Yu <yujie052@gmail.com> writes:
> That said, calling everything through the git.cmd wrapper will
> definitely slow down things even further. According to "some guy on the
> internet" (https://github.com/dahlbyk/posh-git/issues/21 :)) it has
> a cost that might explain your observations (since `magit-status'
> performs about 15 calls to git).
> So I'd say tweaking your PATH would definitely be worth a try.

Thanks. it makes magit-status run faster.

Re: Magit slow on Windows

[Bastian Bechtold]

On 11.05.2012, at 07:53, Eli Zaretskii wrote:

>> From: Kevin Yu <yujie052@gmail.com>
>> Date: Fri, 11 May 2012 13:46:53 +0800
>> 
>> Eli Zaretskii <eliz@gnu.org> writes:
>> 
>>> Which is what? msysgit or the Cygwin git?
>> 
>> msysgit. 
> 
> Then I think this is wrong: you should see
> 
>  c:/git/bin/git.exe
> 
> But I'm not sure this will solve your problem.  The Windows filesystem
> is slower than Linux filesystems, especially with a cold cache, and
> nothing, not even Emacs, can change that.
> 

Changing my path totally solved my problem. Magit is still slower than on my unixes, but not aggravatingly so.

Thanks a lot! You have made my day!

Re: Magit slow on Windows

[Bastian Bechtold]

Here is the best way to change the path without breaking any non-emacs stuff:

(if (eq system-type 'windows-nt)
	(progn
	  (setq exec-path (add-to-list 'exec-path "C:/Program Files (x86)/Git/bin"))
	  (setenv "PATH" (concat "C:\\Program Files (x86)\\Git\\bin;" (getenv "PATH")))))

exec-path is important for Magit, setenv is used by eshell.

Re: Magit slow on Windows

[Kevin Yu]

Bastian Bechtold <basti.bechtold@googlemail.com> writes:

> Here is the best way to change the path without breaking any non-emacs stuff:
>
> (if (eq system-type 'windows-nt)
> 	(progn
>
> exec-path is important for Magit, setenv is used by eshell.
  
set magit-git-executable is better. :)

Re: Magit slow on Windows

[Richard Riley]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Kevin Yu <yujie052@gmail.com>
>> Date: Fri, 11 May 2012 13:46:53 +0800
>> 
>> Eli Zaretskii <eliz@gnu.org> writes:
>> 
>> > Which is what? msysgit or the Cygwin git?
>> 
>> msysgit. 
>
> Then I think this is wrong: you should see
>
>   c:/git/bin/git.exe
>
> Invoking git through a Windows shell script incurs an unnecessary
> slowdown.  You should start Emacs from the shell that knows about the
> msysgit installation, to get correct behavior.
>
> But I'm not sure this will solve your problem.  The Windows filesystem
> is slower than Linux filesystems, especially with a cold cache, and
> nothing, not even Emacs, can change that.

Not in anyway that would cause any human recognisable delay.  Suggesting
that speed of the file system is in any way contributing to this issue
is somewhat silly and misleading and introducing fud.

Re: Magit slow on Windows

[Eli Zaretskii]

> From: Richard Riley <rileyrg@gmail.com>
> Date: Fri, 11 May 2012 12:37:17 +0200
> 
> > The Windows filesystem is slower than Linux filesystems,
> > especially with a cold cache, and nothing, not even Emacs, can
> > change that.
> 
> Not in anyway that would cause any human recognisable delay.  Suggesting
> that speed of the file system is in any way contributing to this issue
> is somewhat silly and misleading and introducing fud.

That the Windows filesystem is slower is a fact, not FUD in any way.

Re: Magit slow on Windows

[Óscar Fuentes]

Richard Riley <rileyrg@gmail.com> writes:

>> But I'm not sure this will solve your problem.  The Windows filesystem
>> is slower than Linux filesystems, especially with a cold cache, and
>> nothing, not even Emacs, can change that.
>
> Not in anyway that would cause any human recognisable delay.  Suggesting
> that speed of the file system is in any way contributing to this issue
> is somewhat silly and misleading and introducing fud.

git is heavily optimized for GNU/Linux. It exploits the APIs and use
cases that performs better there. Msysgit is not optimized for
MS-Windows and, on addition, MSYS/Cygwin, which git depends on,
introduces an intermediate layer that slows down certain operations
quite a bit.

So it is not that NTFS is slower than ext4/whatever for all relevant use
cases, but that git is fine-tuned for the GNU/Linux filesystems. Plus,
creating processes is much slower on Windows than on GNU/Linux (that's
an undeniable fact.) As magit runs several commands for several
often-used operations, that makes a big difference.

Last but not least, there are problems with certain configurations that
dramatically slows down git on Windows7/Vista. Take a look at the
msysgit bug tracker for details.

Re: Magit slow on Windows

[Bastian Bechtold]

On 11.05.2012, at 17:02, Óscar Fuentes <ofv@wanadoo.es> wrote:

> Richard Riley <rileyrg@gmail.com> writes:
> 
>>> But I'm not sure this will solve your problem.  The Windows filesystem
>>> is slower than Linux filesystems, especially with a cold cache, and
>>> nothing, not even Emacs, can change that.
>> 
>> Not in anyway that would cause any human recognisable delay.  Suggesting
>> that speed of the file system is in any way contributing to this issue
>> is somewhat silly and misleading and introducing fud.
> 
> git is heavily optimized for GNU/Linux. It exploits the APIs and use
> cases that performs better there. Msysgit is not optimized for
> MS-Windows and, on addition, MSYS/Cygwin, which git depends on,
> introduces an intermediate layer that slows down certain operations
> quite a bit.
> 
> So it is not that NTFS is slower than ext4/whatever for all relevant use
> cases, but that git is fine-tuned for the GNU/Linux filesystems. Plus,
> creating processes is much slower on Windows than on GNU/Linux (that's
> an undeniable fact.) As magit runs several commands for several
> often-used operations, that makes a big difference.

However, its performance is pretty fine on OSX, even though HFS is probably the worst filesystem that is still in non-legacy (FAT) use. I mean, a global write-lock, seriously?

Re: Magit slow on Windows

[Óscar Fuentes]

Bastian Bechtold <basti@bbheim.de> writes:

>> So it is not that NTFS is slower than ext4/whatever for all relevant use
>> cases, but that git is fine-tuned for the GNU/Linux filesystems. Plus,
>> creating processes is much slower on Windows than on GNU/Linux (that's
>> an undeniable fact.) As magit runs several commands for several
>> often-used operations, that makes a big difference.
>
> However, its performance is pretty fine on OSX, even though HFS is
> probably the worst filesystem that is still in non-legacy (FAT) use. I
> mean, a global write-lock, seriously?

The case of OSX (or any unix-like system, for that matter) is more
related to GNU/Linux than to MS-Windows: fast process creation, no
intermediate emulation layer, C system library closely tied to the OS
API, etc.

Re: Magit slow on Windows

[Miles Bader]

Richard Riley <rileyrg@gmail.com> writes:
> Not in anyway that would cause any human recognisable delay.  Suggesting
> that speed of the file system is in any way contributing to this issue
> is somewhat silly and misleading and introducing fud.

Git uses the filesystem heavily, and tends to rely on it being fast
(and in particular on operations with a "hot cache" being very fast).

The negative impact of Window's much slower FS is a well-known issue
for git.  A system developed with Windows in mind might, for instance,
do much more caching in memory instead.

-miles

-- 
Quack, n. A murderer without a license.

Re: Magit slow on Windows

[Richard Stallman]

    My installation of Magit is dog slow on WIndows. For example, I
    want to see the changes I made to some file in Magit. On Linux and
    OSX, this is as simple as putting point on top of the file name,
    then hit tab.

Since you're talking about something comparable to Windows and MacOS,
you must mean an entire operating system.  In other words, not Linux.
Linux is a kernel, just one component of an operating system.

What you mean must be GNU/Linux -- that is an entire system.
I launched it in 1984, and GNU Emacs is part of it.

To call that system "Linux" is to attribute it to someone else,
which is practically slapping all of us in the face.

I appreciate the effort you've gone to in reporting a bug, since that
may help us fix it.  But it is hard to hold on to the appreciative
feeling when you credit our work to someone else.  Please call the
system "GNU/Linux", not "Linux".

See http://www.gnu.org/gnu/linux-and-gnu.html and
http://www.gnu.org/gnu/gnu-linux-faq.html, plus
http://www.gnu.org/gnu/the-gnu-project.html for the full history of
GNU and GNU/Linux.


--
Dr Richard Stallman
President, Free Software Foundation
51 Franklin St
Boston MA 02110
USA
www.fsf.org  www.gnu.org
Skype: No way! That's nonfree (freedom-denying) software.
  Use Ekiga or an ordinary phone call

Re: Magit slow on Windows

[Bohdan Makohin]

Richard Stallman <rms <at> gnu.org> writes:

> Since you're talking about something comparable to Windows and MacOS,
> you must mean an entire operating system.  In other words, not Linux.
> Linux is a kernel, just one component of an operating system.
> 
> What you mean must be GNU/Linux -- that is an entire system.
> I launched it in 1984, and GNU Emacs is part of it.
> 
> To call that system "Linux" is to attribute it to someone else,
> which is practically slapping all of us in the face.
> 
> I appreciate the effort you've gone to in reporting a bug, since that
> may help us fix it.  But it is hard to hold on to the appreciative
> feeling when you credit our work to someone else.  Please call the
> system "GNU/Linux", not "Linux".
> 
> 

Totally agree with Richard. 
But even knowing the fact that the operating system You use is GNU/Linux - it is 
human habit to reduce words and say "Linux" instead of "GNU/Linux", "X Server" 
instead of "Xorg Server", "ssh" instead of "openssh" and etc. It means that 
GNU/Linux became a "standard" for most of Linux-based OS.

Re: Magit slow on Windows

[Richard Stallman]

    But even knowing the fact that the operating system You use is
    GNU/Linux - it is human habit to reduce words and say "Linux"
    instead of "GNU/Linux", "X Server" instead of "Xorg Server", "ssh"
    instead of "openssh" and etc.

Yes, that is a human tendency, especially when other humans encourage
it.  But people do not have to give way to that tendency.  It is a
human capacity to decide to follow a path other than the easiest one.

Would you please make an effort to say "GNU/Linux" and not treat GNU
as if it were part of Linux?  This kind of help will make a big
difference to us, and it won't take much work once you change the
habit.

--
Dr Richard Stallman
President, Free Software Foundation
51 Franklin St
Boston MA 02110
USA
www.fsf.org  www.gnu.org
Skype: No way! That's nonfree (freedom-denying) software.
  Use Ekiga or an ordinary phone call