setting an optimal value of gc-cons-threshold

setting an optimal value of gc-cons-threshold

[Leo]

Hi list,

I found quite a few people in their .emacs setting gc-cons-threshold to
a different value than the default 400000.

I wonder if there is a guidance for setting this value.

Thanks,
-- 
Leo <sdl.web AT gmail.com>                         (GPG Key: 9283AA3F)

Re: setting an optimal value of gc-cons-threshold

[Xavier Maillard]

Hi Eli,

   >    I wonder if there is a guidance for setting this value.
   > 
   > So do I. I remember increasing it after several Gnus crashes but
   > I did it without reflexion on what value to put here :)
   > 
   > If you manage to get a valid answer here, I am really interested
   > in reading it.

   The answer, based on experience in both increasing and decreasing the
   value, is to use the default.  It is tuned quite well to strike the
   balance between too frequent GCs and ones that are far apart, but take
   an annoyingly long time.

Although I think you are right, I remember having been advised to
do so after having encountered errors such as with
`max-specpdl-size' being too small/high (depending on the case).
Does it sound possible to you ?

	Xavier
-- 
http://www.gnu.org
http://www.april.org
http://www.lolica.org

Re: setting an optimal value of gc-cons-threshold

[Eli Zaretskii]

> Date: Thu, 30 Aug 2007 01:41:59 +0200
> From: Xavier Maillard <xma@gnu.org>
> CC: help-gnu-emacs@gnu.org
> Jabber-ID: xma01@jabber.fr
> 
>    The answer, based on experience in both increasing and decreasing the
>    value, is to use the default.  It is tuned quite well to strike the
>    balance between too frequent GCs and ones that are far apart, but take
>    an annoyingly long time.
> 
> Although I think you are right, I remember having been advised to
> do so

To do what?

> after having encountered errors such as with `max-specpdl-size'
> being too small/high (depending on the case).  Does it sound
> possible to you ?

If you are asking whether it would make sense to change the GC limit
after max-specpdl-size errors, then no, I don't see any connection
between those two.

Re: setting an optimal value of gc-cons-threshold

[Xavier Maillard]

Hi Eli,

   > after having encountered errors such as with `max-specpdl-size'
   > being too small/high (depending on the case).  Does it sound
   > possible to you ?

   If you are asking whether it would make sense to change the GC limit
   after max-specpdl-size errors, then no, I don't see any connection
   between those two.

Then, I will revert such a change into my .emacs.

Thank you very much.

	Xavier
-- 
http://www.gnu.org
http://www.april.org
http://www.lolica.org

Re: setting an optimal value of gc-cons-threshold

[Stefan Monnier]

>    I found quite a few people in their .emacs setting gc-cons-threshold to
>    a different value than the default 400000.
> I did.
>    I wonder if there is a guidance for setting this value.
> So do I. I remember increasing it after several Gnus crashes but
> I did it without reflexion on what value to put here :)

There used to be some benefit to setting it higher for large Emacs sessions
(where GC can take a long time) since it would reduce the frequency of those
GCs.  In Emacs-22, this is not needed any more because of the new
`gc-cons-percentage' variable which automatically causes the limit to grow
as the Emacs process grows, thus preserving a more or less constant
GC overhead.


        Stefan

Re: setting an optimal value of gc-cons-threshold

[Xavier Maillard]

   >    I found quite a few people in their .emacs setting gc-cons-threshold to
   >    a different value than the default 400000.
   > I did.
   >    I wonder if there is a guidance for setting this value.
   > So do I. I remember increasing it after several Gnus crashes but
   > I did it without reflexion on what value to put here :)

   There used to be some benefit to setting it higher for large Emacs sessions
   (where GC can take a long time) since it would reduce the frequency of those
   GCs.  In Emacs-22, this is not needed any more because of the new
   `gc-cons-percentage' variable which automatically causes the limit to grow
   as the Emacs process grows, thus preserving a more or less constant
   GC overhead.

Thank you Stefan for your clarifications.

	Xavier
-- 
http://www.gnu.org
http://www.april.org
http://www.lolica.org

Re: setting an optimal value of gc-cons-threshold

[Stefan Monnier]

> I found quite a few people in their .emacs setting gc-cons-threshold to
> a different value than the default 400000.

> I wonder if there is a guidance for setting this value.

Better not touch it.  In Emacs-22 we introduced gc-cons-percentage which
provides the same benefit as increasing gc-cons-threshold but without
the drawbacks.  And without having to fiddle with it.

I.e. I'd recommend users to remove any gc-cons-threshold settings from their
.emacs.


        Stefan

Re: setting an optimal value of gc-cons-threshold

[Leo]

----- Stefan Monnier (2007-06-10) wrote:-----

>> I found quite a few people in their .emacs setting gc-cons-threshold to
>> a different value than the default 400000.
>
>> I wonder if there is a guidance for setting this value.
>
> Better not touch it.  In Emacs-22 we introduced gc-cons-percentage which
> provides the same benefit as increasing gc-cons-threshold but without
> the drawbacks.  And without having to fiddle with it.
>
> I.e. I'd recommend users to remove any gc-cons-threshold settings from their
> .emacs.

Many thanks.

-- 
Leo <sdl.web AT gmail.com>                         (GPG Key: 9283AA3F)

Re: setting an optimal value of gc-cons-threshold

[Stephen Berman]

On Sat, 09 Jun 2007 21:04:01 -0400 Stefan Monnier <monnier@iro.umontreal.ca> wrote:

>> I found quite a few people in their .emacs setting gc-cons-threshold to
>> a different value than the default 400000.
>
>> I wonder if there is a guidance for setting this value.
>
> Better not touch it.  In Emacs-22 we introduced gc-cons-percentage which
> provides the same benefit as increasing gc-cons-threshold but without
> the drawbacks.  And without having to fiddle with it.
>
> I.e. I'd recommend users to remove any gc-cons-threshold settings from their
> .emacs.

Perhaps this recommendation should be mentioned in the manual and
docstrings.

Steve Berman

Re: setting an optimal value of gc-cons-threshold

[Leo]

----- Stephen Berman (2007-06-10) wrote:-----

> On Sat, 09 Jun 2007 21:04:01 -0400 Stefan Monnier <monnier@iro.umontreal.ca> wrote:
>
>>> I found quite a few people in their .emacs setting gc-cons-threshold to
>>> a different value than the default 400000.
>>
>>> I wonder if there is a guidance for setting this value.
>>
>> Better not touch it.  In Emacs-22 we introduced gc-cons-percentage which
>> provides the same benefit as increasing gc-cons-threshold but without
>> the drawbacks.  And without having to fiddle with it.
>>
>> I.e. I'd recommend users to remove any gc-cons-threshold settings from their
>> .emacs.
>
> Perhaps this recommendation should be mentioned in the manual and
> docstrings.

I second this.

-- 
Leo <sdl.web AT gmail.com>                         (GPG Key: 9283AA3F)

Re: setting an optimal value of gc-cons-threshold

[Stefan Monnier]

>> Perhaps this recommendation should be mentioned in the manual and
>> docstrings.

Not worth the trouble.  It's not like such a setting is really harmful.


        Stefan

Re: setting an optimal value of gc-cons-threshold

[Leo]

----- Stefan Monnier (2007-06-12) wrote:-----

>>> Perhaps this recommendation should be mentioned in the manual and
>>> docstrings.
>
> Not worth the trouble.  It's not like such a setting is really harmful.
>
>
>         Stefan

I think it needs clarification.

There is a 'customization' link in the doc string which I guess suggests
user to customize the variable.

Regards,
-- 
Leo <sdl.web AT gmail.com>                         (GPG Key: 9283AA3F)

Re: setting an optimal value of gc-cons-threshold

[Stefan Monnier]

> There is a 'customization' link in the doc string which I guess suggests
> user to customize the variable.

It just indicates that it can be customized, which it can indeed.


        Stefan