Package: emacs;
Reported by: Akib Azmain Turja <akib <at> disroot.org>
Date: Sun, 27 Nov 2022 20:02:01 UTC
Severity: wishlist
To reply to this bug, email your comments to 59641 AT debbugs.gnu.org.
Toggle the display of automated, internal messages from the tracker.
View this report as an mbox folder, status mbox, maintainer mbox
bug-gnu-emacs <at> gnu.org:bug#59641; Package emacs.
(Sun, 27 Nov 2022 20:02:02 GMT) Full text and rfc822 format available.Akib Azmain Turja <akib <at> disroot.org>:bug-gnu-emacs <at> gnu.org.
(Sun, 27 Nov 2022 20:02:02 GMT) Full text and rfc822 format available.Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):
From: Akib Azmain Turja <akib <at> disroot.org> To: bug-gnu-emacs <at> gnu.org Subject: term is very slow Date: Mon, 28 Nov 2022 02:00:29 +0600
[Message part 1 (text/plain, inline)]
Term mode is very slow. On my computer, Coterm is 2-3x faster than term, and my own terminal emulator, Eat, is 4-5x faster. I tried to profile Term. Steps: 1. emacs -Q (+ -nw, environment variable DISPLAY was unset) 2. M-x profiler-start RET cpu+mem RET 3. M-x term 4. Run the following stupid command: timeout 1 bash -c 'i=0 ; while true ; do echo $i ; i=$((i+1)) ; done' Note I also run emacs -nw in before this command. 5. Wait for finishing. 6. M-x profiler-stop 7. M-x profiler-report Here are the reports (attached), but don't seem to be human readable:
[term-cpu-profile-data (application/octet-stream, attachment)]
[term-memory-profile-data (application/octet-stream, attachment)]
[Message part 4 (text/plain, inline)]
-- Akib Azmain Turja, GPG key: 70018CE5819F17A3BBA666AFE74F0EFA922AE7F5 Fediverse: akib <at> hostux.social Codeberg: akib emailselfdefense.fsf.org | "Nothing can be secure without encryption."
[signature.asc (application/pgp-signature, inline)]
bug-gnu-emacs <at> gnu.org:bug#59641; Package emacs.
(Sun, 27 Nov 2022 20:31:02 GMT) Full text and rfc822 format available.Message #8 received at 59641 <at> debbugs.gnu.org (full text, mbox):
From: Eli Zaretskii <eliz <at> gnu.org> To: Akib Azmain Turja <akib <at> disroot.org> Cc: 59641 <at> debbugs.gnu.org Subject: Re: bug#59641: term is very slow Date: Sun, 27 Nov 2022 22:31:05 +0200
> Date: Mon, 28 Nov 2022 02:00:29 +0600 > From: Akib Azmain Turja via "Bug reports for GNU Emacs, > the Swiss army knife of text editors" <bug-gnu-emacs <at> gnu.org> > > Term mode is very slow. On my computer, Coterm is 2-3x faster than > term, and my own terminal emulator, Eat, is 4-5x faster. > > I tried to profile Term. Steps: > > 1. emacs -Q (+ -nw, environment variable DISPLAY was unset) > 2. M-x profiler-start RET cpu+mem RET > 3. M-x term > 4. Run the following stupid command: > timeout 1 bash -c 'i=0 ; while true ; do echo $i ; i=$((i+1)) ; done' > Note I also run emacs -nw in before this command. > 5. Wait for finishing. > 6. M-x profiler-stop > 7. M-x profiler-report > > Here are the reports (attached), but don't seem to be human readable: Copy them from the display shown by profiler-report, after you expand it completely. And the "cpu" profile is enough; the "mem" one doesn't add anything useful.
Stefan Kangas <stefankangas <at> gmail.com>
to control <at> debbugs.gnu.org.
(Sat, 03 Dec 2022 01:00:04 GMT) Full text and rfc822 format available.bug-gnu-emacs <at> gnu.org:bug#59641; Package emacs.
(Fri, 16 Dec 2022 18:30:02 GMT) Full text and rfc822 format available.Message #13 received at 59641 <at> debbugs.gnu.org (full text, mbox):
From: Akib Azmain Turja <akib <at> disroot.org> To: Eli Zaretskii <eliz <at> gnu.org> Cc: 59641 <at> debbugs.gnu.org Subject: Re: bug#59641: term is very slow Date: Sat, 17 Dec 2022 00:29:02 +0600
[Message part 1 (text/plain, inline)]
Eli Zaretskii <eliz <at> gnu.org> writes:
>> Date: Mon, 28 Nov 2022 02:00:29 +0600
>> From: Akib Azmain Turja via "Bug reports for GNU Emacs,
>> the Swiss army knife of text editors" <bug-gnu-emacs <at> gnu.org>
>>
>> Term mode is very slow. On my computer, Coterm is 2-3x faster than
>> term, and my own terminal emulator, Eat, is 4-5x faster.
>>
>> I tried to profile Term. Steps:
>>
>> 1. emacs -Q (+ -nw, environment variable DISPLAY was unset)
>> 2. M-x profiler-start RET cpu+mem RET
>> 3. M-x term
>> 4. Run the following stupid command:
>> timeout 1 bash -c 'i=0 ; while true ; do echo $i ; i=$((i+1)) ; done'
>> Note I also run emacs -nw in before this command.
>> 5. Wait for finishing.
>> 6. M-x profiler-stop
>> 7. M-x profiler-report
>>
>> Here are the reports (attached), but don't seem to be human readable:
>
> Copy them from the display shown by profiler-report, after you expand it
> completely.
>
> And the "cpu" profile is enough; the "mem" one doesn't add anything useful.
Sorry, I missed this reply. I found this in debbugs.gnu.org. :O
Here is the data: (I didn't redo the steps, I just did
'M-x profiler-find-profile RET path/to/profile'. ;) )
--8<---------------cut here---------------start------------->8---
9397 95% - term-emulate-terminal
2710 27% - term-handle-ansi-escape
1831 18% - term-down
8 0% term-move-columns
765 7% - term-goto
714 7% term-down
82 0% - term--handle-colors-list
66 0% - term--color-as-hex
34 0% - face-background
34 0% - face-attribute
20 0% - face-attribute-merged-with
16 0% face-attribute
6 0% face-attribute
32 0% - face-foreground
20 0% - face-attribute
8 0% face-attribute
1846 18% - term-down
68 0% - vertical-motion
68 0% - jit-lock-function
68 0% - jit-lock-fontify-now
28 0% - jit-lock--run-functions
8 0% - #<compiled -0x1552f4e0aae95e03>
8 0% - font-lock-fontify-region
8 0% - font-lock-default-fontify-region
8 0% font-lock-unfontify-region
8 0% - #<compiled -0x1554fa224e318f03>
8 0% - font-lock-fontify-region
8 0% font-lock-default-fontify-region
4 0% - #<compiled -0x1546c88d0dca1d03>
4 0% - font-lock-fontify-region
4 0% - font-lock-default-fontify-region
4 0% font-lock-extend-region-wholelines
4 0% - #<compiled -0x1546c2ed8adb7d03>
4 0% font-lock-fontify-region
68 0% - term-insert-char
28 0% jit-lock-after-change
28 0% term-move-columns
379 3% - vertical-motion
363 3% - jit-lock-function
317 3% - jit-lock-fontify-now
155 1% - run-with-timer
155 1% - run-at-time
100 1% - timer-activate
96 0% - timer--activate
8 0% timer--time-less-p
20 0% timer-set-time
98 0% - jit-lock--run-functions
22 0% - #<compiled -0x1552ff392e9c9603>
22 0% font-lock-fontify-region
20 0% - #<compiled -0x1552f71a36b2fc03>
20 0% - font-lock-fontify-region
20 0% - font-lock-default-fontify-region
20 0% - font-lock-extend-region-wholelines
20 0% syntax-propertize-wholelines
20 0% - #<compiled -0x1546fd5b53d9da03>
20 0% - font-lock-fontify-region
20 0% font-lock-default-fontify-region
12 0% - #<compiled -0x1552f745b3231203>
12 0% - font-lock-fontify-region
12 0% - font-lock-default-fontify-region
12 0% font-lock-extend-region-wholelines
12 0% - #<compiled -0x1540fac747b55603>
12 0% - font-lock-fontify-region
12 0% - font-lock-default-fontify-region
12 0% font-lock-unfontify-region
8 0% #<compiled -0x1540fac747b55603>
4 0% - #<compiled -0x1540f619efa85203>
4 0% - font-lock-fontify-region
4 0% font-lock-set-defaults
235 2% - term-move-columns
10 0% term-move-to-column
137 1% term-handle-deferred-scroll
111 1% jit-lock-after-change
24 0% - redisplay_internal (C function)
8 0% - term--update-term-menu
8 0% - easy-menu-change
8 0% - easy-menu-add-item
8 0% - easy-menu-get-map
8 0% - mapcar
8 0% - #<compiled 0x12afad0b062c4ebb>
8 0% easy-menu-lookup-name
8 0% - jit-lock-function
8 0% - jit-lock-fontify-now
8 0% - jit-lock--run-functions
4 0% - #<compiled -0x1540fe59cc42d603>
4 0% - font-lock-fontify-region
4 0% font-lock-set-defaults
4 0% - #<compiled -0x1540fe59cc42d603>
4 0% font-lock-fontify-region
8 0% - eval
4 0% if
12 0% syntax-ppss-flush-cache
12 0% - timer-event-handler
12 0% - apply
4 0% - show-paren-function
4 0% - show-paren--default
4 0% syntax-ppss
8 0% - eval
8 0% if
4 0% - jit-lock-function
4 0% - jit-lock-fontify-now
4 0% - jit-lock--run-functions
4 0% #<compiled -0x1546deb559806303>
347 3% - ...
347 3% Automatic GC
127 1% - command-execute
82 0% - funcall-interactively
77 0% - execute-extended-command
65 0% - command-execute
53 0% - byte-code
26 0% - require
26 0% byte-code
4 0% - read-shell-command
4 0% read-from-minibuffer
8 0% funcall-interactively
4 0% - defvar
4 0% byte-code
4 0% term-send-raw
45 0% - byte-code
45 0% - read-extended-command
45 0% - read-extended-command-1
40 0% - completing-read-default
11 0% - timer-event-handler
11 0% - apply
7 0% #<subr F616e6f6e796d6f75732d6c616d626461_anonymous_lambda_9>
4 0% - show-paren-function
2 0% show-paren--default
3 0% #<compiled -0xa873503662230e4>
6 0% - timer-event-handler
6 0% - apply
4 0% - #<compiled -0xb30f43a056c4866>
4 0% - execute-extended-command--shorter
4 0% #<compiled 0x1a044bbfab7de9f2>
2 0% - show-paren-function
2 0% buffer-match-p
2 0% - redisplay_internal (C function)
2 0% - redisplay--pre-redisplay-functions
2 0% - run-hook-with-args
2 0% redisplay--update-region-highlight
--8<---------------cut here---------------end--------------->8---
--
Akib Azmain Turja, GPG key: 70018CE5819F17A3BBA666AFE74F0EFA922AE7F5
Fediverse: akib <at> hostux.social
Codeberg: akib
emailselfdefense.fsf.org | "Nothing can be secure without encryption."
[signature.asc (application/pgp-signature, inline)]
bug-gnu-emacs <at> gnu.org:bug#59641; Package emacs.
(Fri, 16 Dec 2022 18:58:01 GMT) Full text and rfc822 format available.Message #16 received at 59641 <at> debbugs.gnu.org (full text, mbox):
From: Eli Zaretskii <eliz <at> gnu.org> To: Akib Azmain Turja <akib <at> disroot.org> Cc: 59641 <at> debbugs.gnu.org Subject: Re: bug#59641: term is very slow Date: Fri, 16 Dec 2022 20:57:41 +0200
> From: Akib Azmain Turja <akib <at> disroot.org> > Cc: 59641 <at> debbugs.gnu.org > Date: Sat, 17 Dec 2022 00:29:02 +0600 > > >> Here are the reports (attached), but don't seem to be human readable: > > > > Copy them from the display shown by profiler-report, after you expand it > > completely. > > > > And the "cpu" profile is enough; the "mem" one doesn't add anything useful. > > Sorry, I missed this reply. I found this in debbugs.gnu.org. :O > > Here is the data: (I didn't redo the steps, I just did > 'M-x profiler-find-profile RET path/to/profile'. ;) ) > > --8<---------------cut here---------------start------------->8--- > 9397 95% - term-emulate-terminal > 2710 27% - term-handle-ansi-escape > 1831 18% - term-down > 8 0% term-move-columns > 765 7% - term-goto Doesn't surprise me: most of the time is spent in term.el's business logic, not in Emacs display primitives. If you want to work on speeding up term.el, I suggest to start by loading term.el (not term.elc) manually, and repeating the profiling session. That could tell you in more detail where are the hot sports in those functions at the top of the profile. Then some clever optimization ideas are needed to make it significantly faster. Thanks.
bug-gnu-emacs <at> gnu.org:bug#59641; Package emacs.
(Sat, 17 Dec 2022 14:46:02 GMT) Full text and rfc822 format available.Message #19 received at 59641 <at> debbugs.gnu.org (full text, mbox):
From: Akib Azmain Turja <akib <at> disroot.org> To: Eli Zaretskii <eliz <at> gnu.org> Cc: 59641 <at> debbugs.gnu.org Subject: Re: bug#59641: term is very slow Date: Sat, 17 Dec 2022 18:32:09 +0600
[Message part 1 (text/plain, inline)]
I missed this again! There is definitely something wrong. Anyway...
Eli Zaretskii <eliz <at> gnu.org> writes:
>> From: Akib Azmain Turja <akib <at> disroot.org>
>> Cc: 59641 <at> debbugs.gnu.org
>> Date: Sat, 17 Dec 2022 00:29:02 +0600
>>
>> >> Here are the reports (attached), but don't seem to be human readable:
>> >
>> > Copy them from the display shown by profiler-report, after you expand it
>> > completely.
>> >
>> > And the "cpu" profile is enough; the "mem" one doesn't add anything useful.
>>
>> Sorry, I missed this reply. I found this in debbugs.gnu.org. :O
>>
>> Here is the data: (I didn't redo the steps, I just did
>> 'M-x profiler-find-profile RET path/to/profile'. ;) )
>>
>> --8<---------------cut here---------------start------------->8---
>> 9397 95% - term-emulate-terminal
>> 2710 27% - term-handle-ansi-escape
>> 1831 18% - term-down
>> 8 0% term-move-columns
>> 765 7% - term-goto
>
> Doesn't surprise me: most of the time is spent in term.el's business
> logic, not in Emacs display primitives.
I know that very well.
>
> If you want to work on speeding up term.el, I suggest to start by
> loading term.el (not term.elc) manually, and repeating the profiling
> session. That could tell you in more detail where are the hot sports
> in those functions at the top of the profile. Then some clever
> optimization ideas are needed to make it significantly faster.
>
> Thanks.
I guess the hot spot here is 'vertical-motion'. 'vertical-motion' takes
various things into account, which are unnecessary for a terminal
emulator. Eat implements its own motion functions, which make their own
assumptions (e.g. each character takes exactly one column, each line
finishes with a newline, etc). I guess this might be a reason why Eat
is faster.
Note that I'm the author of Eat, and Term is a direct competitor of Eat.
So although I want Term to be faster, I will be of course biased if I
try to improve Term. But don't get me wrong, I really want Term to do
better; competition is fun when you have tough competitors!
Here is the CPU profile data, with 'term.el' loaded:
--8<---------------cut here---------------start------------->8---
1551 87% - term-emulate-terminal
1551 87% - if
1551 87% - progn
1551 87% - save-current-buffer
1515 85% - let*
1515 85% - let
1515 85% - save-current-buffer
1515 85% - unwind-protect
1515 85% - progn
1361 76% - save-restriction
1361 76% - while
1341 75% - let
1130 63% - let*
1126 63% - cond
1126 63% - let
586 33% - funcall
108 6% - vertical-motion
100 5% - jit-lock-function
84 4% - jit-lock-fontify-now
64 3% - jit-lock--run-functions
24 1% - #<compiled -0x1546ff15e132d103>
24 1% - font-lock-fontify-region
24 1% font-lock-default-fontify-region
20 1% - #<compiled -0x1540f7d5aaa5a503>
20 1% font-lock-fontify-region
4 0% - #<compiled -0x1546e4779f110a03>
4 0% - font-lock-fontify-region
4 0% - font-lock-default-fontify-region
4 0% - font-lock-extend-region-wholelines
4 0% syntax-propertize-wholelines
4 0% #<compiled -0x15468ebfcfc1d703>
4 0% - #<compiled -0x15464f08f89a1f03>
4 0% - font-lock-fontify-region
4 0% - font-lock-default-fontify-region
4 0% #<compiled -0x7e8feab343044be>
532 30% - if
528 29% - term-down
528 29% - let
344 19% - if
304 17% - if
304 17% - progn
304 17% - setq
300 16% - -
300 16% - funcall
56 3% - vertical-motion
52 2% - jit-lock-function
48 2% - jit-lock-fontify-now
32 1% - jit-lock--run-functions
12 0% - #<compiled -0x1546ecc8334f3d03>
12 0% font-lock-fontify-region
8 0% - #<compiled -0x1546d143afc77e03>
8 0% - font-lock-fontify-region
8 0% font-lock-default-fontify-region
8 0% - #<compiled -0x1546c17e6a2d8c03>
8 0% - font-lock-fontify-region
8 0% - font-lock-default-fontify-region
8 0% font-lock-extend-region-wholelines
28 1% - progn
28 1% - term-move-columns
28 1% - term-move-to-column
28 1% let
4 0% - and
4 0% =
4 0% term-adjust-current-row-cache
184 10% - cond
176 9% - term-insert-char
172 9% - let
56 3% - put-text-property
20 1% jit-lock-after-change
4 0% syntax-ppss-flush-cache
48 2% - insert-char
20 1% jit-lock-after-change
4 0% and
203 11% - if
187 10% - progn
147 8% - let
24 1% - progn
24 1% setq
24 1% - put-text-property
8 0% jit-lock-after-change
20 1% - insert
8 0% jit-lock-after-change
4 0% - term-horizontal-column
4 0% - -
4 0% - term-current-column
4 0% - cond
4 0% setq
8 0% setq
8 0% if
16 0% setq
136 7% - if
136 7% - progn
136 7% - term-handle-deferred-scroll
136 7% - let
136 7% - if
136 7% - progn
136 7% - save-excursion
136 7% funcall
18 1% - while
18 1% - progn
18 1% - if
18 1% - progn
18 1% - let
18 1% - if
18 1% - progn
18 1% - if
18 1% not
36 2% - if
36 2% - progn
36 2% - redisplay
4 0% - redisplay_internal (C function)
4 0% - eval
4 0% if
171 9% - ...
171 9% Automatic GC
40 2% - command-execute
22 1% - byte-code
22 1% - read-extended-command
22 1% - read-extended-command-1
8 0% completing-read-default
18 1% - funcall-interactively
10 0% - minibuffer-complete
10 0% - completion-in-region
10 0% - completion--in-region
6 0% - #<compiled -0x737dfeebbd42962>
6 0% - apply
6 0% - #<compiled 0x147bf53b9756e246>
6 0% - completion--in-region-1
6 0% - completion--do-completion
6 0% - completion-try-completion
6 0% - completion--nth-completion
6 0% - completion--some
6 0% - #<compiled -0xb70f507d86b95cc>
6 0% - completion-basic-try-completion
6 0% - #<subr F616e6f6e796d6f75732d6c616d626461_anonymous_lambda_54>
6 0% complete-with-action
4 0% - #<compiled -0x737dfeebbd42962>
4 0% - apply
4 0% - #<compiled 0x147bf53b9756e246>
4 0% - completion--in-region-1
4 0% - completion--do-completion
4 0% - completion-try-completion
4 0% - completion--nth-completion
4 0% - completion--some
4 0% - #<compiled -0xb70f507d86b95cc>
4 0% - completion-pcm-try-completion
4 0% - completion-pcm--find-all-completions
4 0% - completion-pcm--all-completions
4 0% - #<subr F616e6f6e796d6f75732d6c616d626461_anonymous_lambda_54>
4 0% complete-with-action
8 0% - execute-extended-command
8 0% - command-execute
8 0% funcall-interactively
7 0% - timer-event-handler
7 0% - apply
4 0% #<compiled -0xabecf6a07658066>
3 0% - #<compiled -0xabecf6a07658066>
3 0% execute-extended-command--shorter
2 0% - redisplay_internal (C function)
2 0% - term--update-term-menu
2 0% - if
2 0% - progn
2 0% - let
2 0% - easy-menu-change
2 0% - mapcar
2 0% - #<lambda 0x1984cdb809973d69>
2 0% - vector
2 0% format
--8<---------------cut here---------------end--------------->8---
--
Akib Azmain Turja, GPG key: 70018CE5819F17A3BBA666AFE74F0EFA922AE7F5
Fediverse: akib <at> hostux.social
Codeberg: akib
emailselfdefense.fsf.org | "Nothing can be secure without encryption."
[signature.asc (application/pgp-signature, inline)]
bug-gnu-emacs <at> gnu.org:bug#59641; Package emacs.
(Sat, 17 Dec 2022 15:48:02 GMT) Full text and rfc822 format available.Message #22 received at 59641 <at> debbugs.gnu.org (full text, mbox):
From: Eli Zaretskii <eliz <at> gnu.org> To: Akib Azmain Turja <akib <at> disroot.org> Cc: 59641 <at> debbugs.gnu.org Subject: Re: bug#59641: term is very slow Date: Sat, 17 Dec 2022 17:47:31 +0200
> From: Akib Azmain Turja <akib <at> disroot.org> > Cc: 59641 <at> debbugs.gnu.org > Date: Sat, 17 Dec 2022 18:32:09 +0600 > > I guess the hot spot here is 'vertical-motion'. 'vertical-motion' takes > various things into account, which are unnecessary for a terminal > emulator. Eat implements its own motion functions, which make their own > assumptions (e.g. each character takes exactly one column, each line > finishes with a newline, etc). I guess this might be a reason why Eat > is faster. If there's no need for the specific features that vertical-motion provides, calling it might indeed be unnecessary.
Stefan Kangas <stefankangas <at> gmail.com>
to control <at> debbugs.gnu.org.
(Wed, 10 Jan 2024 10:49:01 GMT) Full text and rfc822 format available.Stefan Kangas <stefankangas <at> gmail.com>
to control <at> debbugs.gnu.org.
(Tue, 11 Feb 2025 07:29:02 GMT) Full text and rfc822 format available.
GNU bug tracking system
Copyright (C) 1999 Darren O. Benham,
1997,2003 nCipher Corporation Ltd,
1994-97 Ian Jackson.