Pre-load time-date.el

Pre-load time-date.el

[Eli Zaretskii]

Since we turn on blink-cursor mode by default, and it requires timers,
we now always load time-date.el at startup.  Any reasons not to
pre-load it in loadup.el on the emacs-24 branch?

(Please respond before the moratorium on commits at 00:00 UTC
tomorrow.)

Re: Pre-load time-date.el

[Glenn Morris]

Eli Zaretskii wrote:

> Since we turn on blink-cursor mode by default, and it requires timers,
> we now always load time-date.el at startup.  Any reasons not to
> pre-load it in loadup.el on the emacs-24 branch?

This is not a new issue, right? It goes back to 24.1?
And it causes no measurable harm?
So why the rush to change it at the last minute?
(My fear would be that it somehow causes purespace overflow on some
platform. Maybe that is totally unfounded?)

Re: Pre-load time-date.el

[Glenn Morris]

PS since the cursor blinks in 23.4, why does it not happen there?
Maybe timer--time-less-p "now" just using time-less-p is the cause?
(Uninvestigated.) So maybe it can actually be avoided, though obviously
that's definitely something for trunk.

Re: Pre-load time-date.el

[Eli Zaretskii]

> From: Glenn Morris <rgm@gnu.org>
> Cc: emacs-devel@gnu.org
> Date: Thu, 16 Oct 2014 14:22:01 -0400
> 
> 
> PS since the cursor blinks in 23.4, why does it not happen there?

I didn't look.

> Maybe timer--time-less-p "now" just using time-less-p is the cause?

No, it's seconds-to-time, which is called by timer-set-idle-time,
which in turn is called by run-with-idle-timer.

Re: Pre-load time-date.el

[Paul Eggert]

On 10/16/2014 11:42 AM, Eli Zaretskii wrote:
> it's seconds-to-time, which is called by timer-set-idle-time,
> which in turn is called by run-with-idle-timer.
>

It strikes me that low-level time primitives should accept counts of 
seconds (either integers, or floats) in addition to the funky (HIGH LOW 
USEC PSEC) format that they currently accept.This would be cheap to add 
at the low level, and would remove the need for many calls to 
seconds-to-time, including the abovementioned one, and would mean Emacs 
wouldn't need to preload time-date.el.  Also, doing the conversion in C 
would lessen the amount of timing error introduced by the conversion itself.

Re: Pre-load time-date.el

[Paul Eggert]

Paul Eggert wrote:
> It strikes me that low-level time primitives should accept counts of seconds
> (either integers, or floats) in addition to the funky (HIGH LOW USEC PSEC)
> format that they currently accept.

I implemented something along these lines and installed it in the master as 
0921dbc.  This should fix the original problem, as time-date should no longer 
need to be loaded for timers to work.  (Of course it may have introduced new 
problems....)

Re: Pre-load time-date.el

[Eli Zaretskii]

> From: Glenn Morris <rgm@gnu.org>
> Cc: emacs-devel@gnu.org
> Date: Thu, 16 Oct 2014 14:03:01 -0400
> 
> Eli Zaretskii wrote:
> 
> > Since we turn on blink-cursor mode by default, and it requires timers,
> > we now always load time-date.el at startup.  Any reasons not to
> > pre-load it in loadup.el on the emacs-24 branch?
> 
> This is not a new issue, right? It goes back to 24.1?
> And it causes no measurable harm?
> So why the rush to change it at the last minute?

Because I just noticed it now.

> (My fear would be that it somehow causes purespace overflow on some
> platform. Maybe that is totally unfounded?)

We could bump PURESIZE by the same amount as pre-loading this file
requires.

Re: Pre-load time-date.el

[Glenn Morris]

Eli Zaretskii wrote:

>> So why the rush to change it at the last minute?
>
> Because I just noticed it now.

Obviously. I don't see _why_ you want to change it now though, since:

>> This is not a new issue, right? It goes back to 24.1?
>> And it causes no measurable harm?

Re: Pre-load time-date.el

[Eli Zaretskii]

> From: Glenn Morris <rgm@gnu.org>
> Cc: emacs-devel@gnu.org
> Date: Thu, 16 Oct 2014 14:29:47 -0400
> 
> Eli Zaretskii wrote:
> 
> >> So why the rush to change it at the last minute?
> >
> > Because I just noticed it now.
> 
> Obviously. I don't see _why_ you want to change it now though

Because I don't see a reason why not.

Re: Pre-load time-date.el

[Eli Zaretskii]

> Date: Thu, 16 Oct 2014 21:51:24 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: emacs-devel@gnu.org
> 
> > From: Glenn Morris <rgm@gnu.org>
> > Cc: emacs-devel@gnu.org
> > Date: Thu, 16 Oct 2014 14:29:47 -0400
> > 
> > Eli Zaretskii wrote:
> > 
> > >> So why the rush to change it at the last minute?
> > >
> > > Because I just noticed it now.
> > 
> > Obviously. I don't see _why_ you want to change it now though
> 
> Because I don't see a reason why not.

But if you are uneasy about this, let's forget it.

Re: Pre-load time-date.el

[Glenn Morris]

Eli Zaretskii wrote:

> But if you are uneasy about this, let's forget it.

I am, so thanks. :)

Re: Pre-load time-date.el

[Stefan Monnier]

>> Since we turn on blink-cursor mode by default, and it requires timers,
>> we now always load time-date.el at startup.  Any reasons not to
>> pre-load it in loadup.el on the emacs-24 branch?
> This is not a new issue, right? It goes back to 24.1?
> And it causes no measurable harm?
> So why the rush to change it at the last minute?
> (My fear would be that it somehow causes purespace overflow on some
> platform. Maybe that is totally unfounded?)

Let's keep it for 24.5, it's not an urgent issue.  Most of the times
I move something to loadup.el, I end up messing something up
(e.g. because the code uses defconst whose value gets purified, or
because I forget to update lisp.mk, or ...).


        Stefan