bug#51094: Check if run-with{-idle,}-timer needs to create a timer

bug#51094: Check if run-with{-idle,}-timer needs to create a timer

[Philip Kaludercic]

[-- Attachment #1: Type: text/plain, Size: 1524 bytes --]

Tags: patch


I have seen a few packages use run-with-timer or run-with-idle-timer,
where the SECS parameter is configurable with a user option. When this
timer doesn't repeat itself and it makes sense to set SECS to 0 when you
want something to run immediately, I don't think it makes sense to
create a timer object.

The following patch would remove this overhead, and calls the function
immediately if it makes sense.

Comparing these two test cases

--8<---------------cut here---------------start------------->8---
(benchmark-run-compiled 10000
  (funcall #'message "Test %s" "case"))

(benchmark-run-compiled 10000
  (run-with-timer 0 nil #'message "Test %s" "case"))
--8<---------------cut here---------------end--------------->8---

it turns out that funcall takes longer, but run-with-timer (or
run-with-idle-timer for that matter) block Emacs for significantly
longer.  Now this is a very forced example, because run-with-timer is
usually not called in a loop, but it do think it demonstrates a general
advantage.

In GNU Emacs 28.0.60 (build 6, x86_64-pc-linux-gnu, X toolkit, cairo version 1.16.0, Xaw scroll bars)
 of 2021-10-05 built on icterid
Repository revision: 1cd1b2835b5e35562c677c48dcf185bb73af4275
Repository branch: emacs-28
Windowing system distributor 'The X.Org Foundation', version 11.0.12011000
System Description: Debian GNU/Linux 11 (bullseye)

Configured using:
 'configure --with-native-compilation --with-x-toolkit=lucid
 --with-imagemagick 'CFLAGS=-O2 -march=native -pipe' LDFLAGS=-flto'


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-Avoid-creating-timer-if-possible.patch --]
[-- Type: text/patch, Size: 2149 bytes --]

From 1e7c19e1aba6152dbc49173faa04d614fe2961c2 Mon Sep 17 00:00:00 2001
From: Philip Kaludercic <philipk@posteo.net>
Date: Fri, 8 Oct 2021 11:20:53 +0200
Subject: [PATCH] Avoid creating timer if possible

* timer.el (run-with-timer): Check if function can be called
  immediately
(run-with-idle-timer): Check if function can be called
  immediately
---
 lisp/emacs-lisp/timer.el | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)

diff --git a/lisp/emacs-lisp/timer.el b/lisp/emacs-lisp/timer.el
index 1ef4931b7b..22029798fd 100644
--- a/lisp/emacs-lisp/timer.el
+++ b/lisp/emacs-lisp/timer.el
@@ -422,7 +422,9 @@ run-with-timer
 
 This function returns a timer object which you can use in `cancel-timer'."
   (interactive "sRun after delay (seconds): \nNRepeat interval: \naFunction: ")
-  (apply 'run-at-time secs repeat function args))
+  (if (and (numberp secs) (= secs 0) (null repeat))
+      (apply function args)
+    (apply 'run-at-time secs repeat function args)))
 
 (defun add-timeout (secs function object &optional repeat)
   "Add a timer to run SECS seconds from now, to call FUNCTION on OBJECT.
@@ -448,13 +450,15 @@ run-with-idle-timer
 This function returns a timer object which you can use in `cancel-timer'."
   (interactive
    (list (read-from-minibuffer "Run after idle (seconds): " nil nil t)
-	 (y-or-n-p "Repeat each time Emacs is idle? ")
-	 (intern (completing-read "Function: " obarray 'fboundp t))))
-  (let ((timer (timer-create)))
-    (timer-set-function timer function args)
-    (timer-set-idle-time timer secs repeat)
-    (timer-activate-when-idle timer t)
-    timer))
+         (y-or-n-p "Repeat each time Emacs is idle? ")
+         (intern (completing-read "Function: " obarray 'fboundp t))))
+  (if (and (numberp secs) (= secs 0) (null repeat))
+      (apply function args)
+    (let ((timer (timer-create)))
+      (timer-set-function timer function args)
+      (timer-set-idle-time timer secs repeat)
+      (timer-activate-when-idle timer t)
+      timer)))
 \f
 (defvar with-timeout-timers nil
   "List of all timers used by currently pending `with-timeout' calls.")
-- 
2.30.2


[-- Attachment #3: Type: text/plain, Size: 24 bytes --]


-- 
	Philip Kaludercic

bug#51094: Check if run-with{-idle, }-timer needs to create a timer

[Basil L. Contovounesios via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Philip Kaludercic [2021-10-08 09:36 +0000] wrote:

> I have seen a few packages use run-with-timer or run-with-idle-timer,
> where the SECS parameter is configurable with a user option. When this
> timer doesn't repeat itself and it makes sense to set SECS to 0 when you
> want something to run immediately, I don't think it makes sense to
> create a timer object.

IIUC, the semantics of SECS=0 (alias nil) is not the same as eager
funcall, because timer functions are intended to be run asynchronously
in a separate command loop.  So often what is meant by "now" is e.g. "as
soon as I quit the current active minibuffer".

I realise this patch does not touch run-at-time, but it's documented as
being interchangeable with run-with-timer, so the eager funcall sounds
like a breaking change.

If packages indeed want to run something immediately, why create a timer
at all?  Or am I misunderstanding something?

Thanks,

-- 
Basil

bug#51094: Check if run-with{-idle, }-timer needs to create a timer

[Lars Ingebrigtsen]

"Basil L. Contovounesios" via "Bug reports for GNU Emacs, the Swiss army
knife of text editors" <bug-gnu-emacs@gnu.org> writes:

> IIUC, the semantics of SECS=0 (alias nil) is not the same as eager
> funcall, because timer functions are intended to be run asynchronously
> in a separate command loop.  So often what is meant by "now" is e.g. "as
> soon as I quit the current active minibuffer".

Yes, the semantics of running things from a timer are different from
running them in the normal flow, so we can't transform 0-delay timers
into funcalls.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

bug#51094: Check if run-with{-idle, }-timer needs to create a timer

[Lars Ingebrigtsen]

Lars Ingebrigtsen <larsi@gnus.org> writes:

> Yes, the semantics of running things from a timer are different from
> running them in the normal flow, so we can't transform 0-delay timers
> into funcalls.

See for instance:

  (when (> (minibuffer-depth) 0)
    ;; We're inside a minibuffer already, so if the emacs-client is trying
    ;; to open a frame on a new display, we might end up with an unusable
    ;; frame because input from that display will be blocked (until exiting
    ;; the minibuffer).  Better exit this minibuffer right away.
    (run-with-timer 0 nil (lambda () (server-execute-continuation proc)))

So this change is a no go.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

bug#51094: Check if run-with{-idle, }-timer needs to create a timer

[Philip Kaludercic]

Lars Ingebrigtsen <larsi@gnus.org> writes:

> Lars Ingebrigtsen <larsi@gnus.org> writes:
>
>> Yes, the semantics of running things from a timer are different from
>> running them in the normal flow, so we can't transform 0-delay timers
>> into funcalls.
>
> See for instance:
>
>   (when (> (minibuffer-depth) 0)
>     ;; We're inside a minibuffer already, so if the emacs-client is trying
>     ;; to open a frame on a new display, we might end up with an unusable
>     ;; frame because input from that display will be blocked (until exiting
>     ;; the minibuffer).  Better exit this minibuffer right away.
>     (run-with-timer 0 nil (lambda () (server-execute-continuation proc)))
>
> So this change is a no go.

Ok, that makes sense.

One last question, would it make sense to provide a function like
run-with-timer that would work like my patch?

-- 
	Philip Kaludercic

bug#51094: Check if run-with{-idle, }-timer needs to create a timer

[Lars Ingebrigtsen]

Philip Kaludercic <philipk@posteo.net> writes:

> One last question, would it make sense to provide a function like
> run-with-timer that would work like my patch?

It sounds really specialised -- I don't think there'd be much point.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

bug#51094: Check if run-with{-idle, }-timer needs to create a timer

[Philip Kaludercic]

"Basil L. Contovounesios" <contovob@tcd.ie> writes:


> Philip Kaludercic [2021-10-08 09:36 +0000] wrote:
>
>> I have seen a few packages use run-with-timer or run-with-idle-timer,
>> where the SECS parameter is configurable with a user option. When this
>> timer doesn't repeat itself and it makes sense to set SECS to 0 when you
>> want something to run immediately, I don't think it makes sense to
>> create a timer object.
>
> IIUC, the semantics of SECS=0 (alias nil) is not the same as eager
> funcall, because timer functions are intended to be run asynchronously
> in a separate command loop.  So often what is meant by "now" is e.g. "as
> soon as I quit the current active minibuffer".

I see, it might be necessary to consider examples where this might go
wrong.

> I realise this patch does not touch run-at-time, but it's documented as
> being interchangeable with run-with-timer, so the eager funcall sounds
> like a breaking change.
>
> If packages indeed want to run something immediately, why create a timer
> at all?  Or am I misunderstanding something?

Because you might want to delay certain things, such as highlighting or
completion. It makes sense on older devices, but modern systems can run
a lot of things immediately. Consider lazy-highlight-initial-delay or
the icomplete options. On my desktop I have no issue with setting
lazy-highlight-initial-delay to 0 but on my laptop I keep the default
value.

> Thanks,

-- 
	Philip Kaludercic

bug#51094: Check if run-with{-idle, }-timer needs to create a timer

[Basil L. Contovounesios via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Philip Kaludercic [2021-10-08 11:03 +0000] wrote:

> "Basil L. Contovounesios" <contovob@tcd.ie> writes:
>> Philip Kaludercic [2021-10-08 09:36 +0000] wrote:
>>
>>> I have seen a few packages use run-with-timer or run-with-idle-timer,
>>> where the SECS parameter is configurable with a user option. When this
>>> timer doesn't repeat itself and it makes sense to set SECS to 0 when you
>>> want something to run immediately, I don't think it makes sense to
>>> create a timer object.
>>
>> IIUC, the semantics of SECS=0 (alias nil) is not the same as eager
>> funcall, because timer functions are intended to be run asynchronously
>> in a separate command loop.  So often what is meant by "now" is e.g. "as
>> soon as I quit the current active minibuffer".
>
> I see, it might be necessary to consider examples where this might go
> wrong.

Off the top of my head there's also ivy-quit-and-run in the ivy package,
but Lars pointed to effectively the same idiom.

>> I realise this patch does not touch run-at-time, but it's documented as
>> being interchangeable with run-with-timer, so the eager funcall sounds
>> like a breaking change.
>>
>> If packages indeed want to run something immediately, why create a timer
>> at all?  Or am I misunderstanding something?
>
> Because you might want to delay certain things, such as highlighting or
> completion.

What I meant is, the choice whether to run-with-timer or funcall can be
left to the calling package, since it knows better what it wants to do.

Thanks,

-- 
Basil

bug#51094: Check if run-with{-idle, }-timer needs to create a timer

[Eli Zaretskii]

> Cc: 51094@debbugs.gnu.org
> Date: Fri, 08 Oct 2021 11:46:00 +0100
> From:  "Basil L. Contovounesios" via "Bug reports for GNU Emacs,
>  the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>
> 
> Philip Kaludercic [2021-10-08 09:36 +0000] wrote:
> 
> > I have seen a few packages use run-with-timer or run-with-idle-timer,
> > where the SECS parameter is configurable with a user option. When this
> > timer doesn't repeat itself and it makes sense to set SECS to 0 when you
> > want something to run immediately, I don't think it makes sense to
> > create a timer object.
> 
> IIUC, the semantics of SECS=0 (alias nil) is not the same as eager
> funcall, because timer functions are intended to be run asynchronously
> in a separate command loop.  So often what is meant by "now" is e.g. "as
> soon as I quit the current active minibuffer".

Right, and that's one more aspect of this change to consider.  It
could very well change the behavior in incompatible ways, so I wonder
whether we really should make this change.

bug#51094: Check if run-with{-idle, }-timer needs to create a timer

[Eli Zaretskii]

> From: Philip Kaludercic <philipk@posteo.net>
> Date: Fri, 08 Oct 2021 09:36:22 +0000
> 
> I have seen a few packages use run-with-timer or run-with-idle-timer,
> where the SECS parameter is configurable with a user option. When this
> timer doesn't repeat itself and it makes sense to set SECS to 0 when you
> want something to run immediately, I don't think it makes sense to
> create a timer object.
> 
> The following patch would remove this overhead, and calls the function
> immediately if it makes sense.

I'm not sure we should care, but in any case, what you do there
requires a comment explaining the rationale.

Thanks.