Improving Emacs performance on Cygwin

Improving Emacs performance on Cygwin

[Ken Brown]

I would like to apply the following patch, which seems to give a 
substantial performance improvement on Cygwin (and obviously has no 
effect on other platforms).

=== modified file 'src/dired.c'
--- src/dired.c 2010-07-25 00:20:51 +0000
+++ src/dired.c 2010-07-26 20:49:34 +0000
@@ -72,8 +72,7 @@
  #endif /* not MSDOS */
  #endif /* not SYSV_SYSTEM_DIR */

-/* Some versions of Cygwin don't have d_ino in `struct dirent'.  */
-#if defined(MSDOS) || defined(__CYGWIN__)
+#ifdef MSDOS
  #define DIRENTRY_NONEMPTY(p) ((p)->d_name[0] != 0)
  #else
  #define DIRENTRY_NONEMPTY(p) ((p)->d_ino)

This reverses a change made by Eli on 2006-01-27.  Rationale: The time 
when Cygwin didn't have d_ino in dirent was a brief period around 
December 2005, and it applied only to Cygwin 1.5.19, which has long been 
obsolete.

Any objections?  If not, I would apply it to the emacs-23 branch and let 
it propagate to the trunk.

Ken

Re: Improving Emacs performance on Cygwin

[Stefan Monnier]

> Any objections?  If not, I would apply it to the emacs-23 branch and let it
> propagate to the trunk.

Sounds good to me,


        Stefan

Re: Improving Emacs performance on Cygwin

[Eli Zaretskii]

> Date: Mon, 26 Jul 2010 17:00:06 -0400
> From: Ken Brown <kbrown@cornell.edu>
> 
> This reverses a change made by Eli on 2006-01-27.  Rationale: The time 
> when Cygwin didn't have d_ino in dirent was a brief period around 
> December 2005, and it applied only to Cygwin 1.5.19, which has long been 
> obsolete.

Cygwin 1.5.19 is obsolete, but are you willing to make it completely
unsupported by Emacs?

If you don't want to drop it completely, then maybe condition on the
version of Cygwin that is installed.

Re: Improving Emacs performance on Cygwin

[Ken Brown]

On 7/26/2010 10:55 PM, Eli Zaretskii wrote:
>> Date: Mon, 26 Jul 2010 17:00:06 -0400
>> From: Ken Brown<kbrown@cornell.edu>
>>
>> This reverses a change made by Eli on 2006-01-27.  Rationale: The time
>> when Cygwin didn't have d_ino in dirent was a brief period around
>> December 2005, and it applied only to Cygwin 1.5.19, which has long been
>> obsolete.
>
> Cygwin 1.5.19 is obsolete, but are you willing to make it completely
> unsupported by Emacs?

I think that's OK.  There will never be another Emacs release for Cygwin 
1.5 (which was frozen at 1.5.25), so the change will only affect people 
who are running Cygwin 1.5.19, don't want to upgrade to 1.5.25, *and* 
want to build a modern Emacs from source.  This seems pretty unlikely, 
and any such person can just revert my change before building.

Ken

Re: Improving Emacs performance on Cygwin

[Eli Zaretskii]

> Date: Tue, 27 Jul 2010 07:07:42 -0400
> From: Ken Brown <kbrown@cornell.edu>
> Cc: "emacs-devel@gnu.org" <emacs-devel@gnu.org>
> 
> > Cygwin 1.5.19 is obsolete, but are you willing to make it completely
> > unsupported by Emacs?
> 
> I think that's OK.  There will never be another Emacs release for Cygwin 
> 1.5 (which was frozen at 1.5.25), so the change will only affect people 
> who are running Cygwin 1.5.19, don't want to upgrade to 1.5.25, *and* 
> want to build a modern Emacs from source.  This seems pretty unlikely, 
> and any such person can just revert my change before building.

They wouldn't know what to revert.

Anyway, it's your call, but maybe post something to the Cygwin list
announcing that Emacs 23.3 and later is dropping Cygwin 1.5 support.

Re: Improving Emacs performance on Cygwin

[Ken Brown]

On 7/27/2010 1:07 PM, Eli Zaretskii wrote:
>> Date: Tue, 27 Jul 2010 07:07:42 -0400
>> From: Ken Brown<kbrown@cornell.edu>
>> Cc: "emacs-devel@gnu.org"<emacs-devel@gnu.org>
>>
>>> Cygwin 1.5.19 is obsolete, but are you willing to make it completely
>>> unsupported by Emacs?
>>
>> I think that's OK.  There will never be another Emacs release for Cygwin
>> 1.5 (which was frozen at 1.5.25), so the change will only affect people
>> who are running Cygwin 1.5.19, don't want to upgrade to 1.5.25, *and*
>> want to build a modern Emacs from source.  This seems pretty unlikely,
>> and any such person can just revert my change before building.
>
> They wouldn't know what to revert.
>
> Anyway, it's your call, but maybe post something to the Cygwin list
> announcing that Emacs 23.3 and later is dropping Cygwin 1.5 support.

OK, I'll do that when Emacs 23.3 is released and packaged for Cygwin. 
Thanks for the suggestion.  BTW, it's only Cygwin 1.5.19 that's 
affected.  People running Cygwin 1.5.20 and later should still be able 
to build Emacs AFAIK.

Ken

Re: Improving Emacs performance on Cygwin

[Chad Brown]

On Jul 27, 2010, at 11:01 AM, Ken Brown wrote:
> 
> OK, I'll do that when Emacs 23.3 is released and packaged for Cygwin. Thanks for the suggestion.  BTW, it's only Cygwin 1.5.19 that's affected.  People running Cygwin 1.5.20 and later should still be able to build Emacs AFAIK.

Note in etc/problems, perhaps?

Re: Improving Emacs performance on Cygwin

[Ken Brown]

On 7/27/2010 3:12 PM, Chad Brown wrote:
> Note in etc/problems, perhaps?

Done.

Re: Improving Emacs performance on Cygwin

[Davis Herring]

> I would like to apply the following patch, which seems to give a
> substantial performance improvement on Cygwin (and obviously has no
> effect on other platforms).
> [...]
>   #define DIRENTRY_NONEMPTY(p) ((p)->d_name[0] != 0)
>   #else
>   #define DIRENTRY_NONEMPTY(p) ((p)->d_ino)

How can looking up d_name[0] be any slower than looking up d_ino?  They're
both already in memory (d_name is an array, not a pointer, even); are we
really seeing a "substantial performance improvement" from a cache effect
or something?

This is a question born of curiosity, not a criticism; so long as the
relevant version of Cygwin is properly handled (as addressed elsewhere in
the thread) simplifying the conditional is already a win.

Davis

-- 
This product is sold by volume, not by mass.  If it appears too dense or
too sparse, it is because mass-energy conversion has occurred during
shipping.

Re: Improving Emacs performance on Cygwin

[Ken Brown]

On 7/27/2010 4:01 PM, Davis Herring wrote:
>> I would like to apply the following patch, which seems to give a
>> substantial performance improvement on Cygwin (and obviously has no
>> effect on other platforms).
>> [...]
>>    #define DIRENTRY_NONEMPTY(p) ((p)->d_name[0] != 0)
>>    #else
>>    #define DIRENTRY_NONEMPTY(p) ((p)->d_ino)
>
> How can looking up d_name[0] be any slower than looking up d_ino?  They're
> both already in memory (d_name is an array, not a pointer, even); are we
> really seeing a "substantial performance improvement" from a cache effect
> or something?

I don't know.

> This is a question born of curiosity, not a criticism; so long as the
> relevant version of Cygwin is properly handled (as addressed elsewhere in
> the thread) simplifying the conditional is already a win.

My initial motivation was just to simplify the code and get rid of an 
unnecessary special case for Cygwin.  I then thought I observed a 
performance improvement, but I could be wrong.  I didn't make any 
measurements.

Ken

Bazar trouble

[Alessandro Madruga Correia]

Hi,

Have anyone place which I can download repository?
I waited for 4 hours and don't finish, so closed connection.

And now, I can't connect...

amcorreia@phoenix:t$ bzr branch http://bzr.savannah.gnu.org/r/emacs/trunk
bzr: ERROR: Connection error: while sending POST 
/r/emacs/trunk/.bzr/smart: [Errno 110] Connection timed out


Best regards.

PS: excuse my poor english.

-- 
   ,= ,-_-. =.   [<o>] Alessandro Madruga Correia
  ((_/)o o(\_))  Viaconnect -- Suporte Técnico +55 (54) 4009 3444
   `-'(. .)`-'   "A matemática é a arte de transformar uma coisa...
       \_/       ... na mesma coisa." (Demétrius Melo de Souza)

Re: Bazar trouble

[Óscar Fuentes]

Alessandro Madruga Correia <amcorreia@viaconnect.com.br> writes:

> Have anyone place which I can download repository?
> I waited for 4 hours and don't finish, so closed connection.

You can try using the mirror at LaunchPad:

bzr branch http://bazaar.launchpad.net/~vcs-imports/emacs/trunk emacs