* [PATCH] Fix python-info-current-defun performance for large defuns
@ 2023-09-04 1:37 Tom Gillespie
2023-09-04 11:49 ` Eli Zaretskii
0 siblings, 1 reply; 6+ messages in thread
From: Tom Gillespie @ 2023-09-04 1:37 UTC (permalink / raw)
To: Emacs developers
[-- Attachment #1: Type: text/plain, Size: 445 bytes --]
Hi,
Here is a patch to fix a performance issue in
python-info-current-defun that appears when the function
is run on large defuns. The issue can appear to the user as a
noticeable (~500ms) freeze when the point is in a large defun
in a python buffer with which-function-mode enabled. More
details in the commit message and comments in the code.
I made the patch against the emacs-29 branch, but will also
on master without issue.
Best,
Tom
[-- Attachment #2: 0001-Fix-python-info-current-defun-performance-for-large-.patch --]
[-- Type: text/x-patch, Size: 8271 bytes --]
From 523ea2e041d1efdcc6ab6bac637ce4253b856fe2 Mon Sep 17 00:00:00 2001
From: Tom Gillespie <tgbugs@gmail.com>
Date: Sun, 3 Sep 2023 15:09:17 -0700
Subject: [PATCH] Fix python-info-current-defun performance for large defuns
* lisp/progmodes/python.el (python-info-current-defun): Significant
performance improvement for large functions.
* test/lisp/progmodes/python-tests.el (python-info-current-defun-5):
Added test for multiply nested non-defun lines.
The previous pervious implementation had two issues. First, it checked
every single intervening defun in the file, even those at the same
nesting level. Second, when it found a putative parent defun it would
call python-nav-end-of-defun which is incredibly slow for large defuns
(e.g. python classes that are many thousands of lines long).
The new implementation avoids these issues by using re-search-backward
to find the containing deindented line. It also handles cases where
there are multiple nested deindented lines that are not defuns.
---
lisp/progmodes/python.el | 85 ++++++++++++++++++-----------
test/lisp/progmodes/python-tests.el | 20 +++++++
2 files changed, 74 insertions(+), 31 deletions(-)
diff --git a/lisp/progmodes/python.el b/lisp/progmodes/python.el
index d9ca37145e1..1fbd220ddb8 100644
--- a/lisp/progmodes/python.el
+++ b/lisp/progmodes/python.el
@@ -5607,42 +5607,31 @@ python-info-current-defun
(starting-pos (point))
(first-run t)
(last-indent)
+ (non-defun-indent (python-info-looking-at-beginning-of-defun))
(type))
(catch 'exit
- (while (python-nav-beginning-of-defun 1)
+ (while (python-nav-beginning-of-defun 1) ; `python-nav-beginning-of-defun' doesn't respect indentation
(when (save-match-data
(and
(or (not last-indent)
(< (current-indentation) last-indent))
- (or
- (and first-run
- (save-excursion
- ;; If this is the first run, we may add
- ;; the current defun at point.
- (setq first-run nil)
- (goto-char starting-pos)
- (python-nav-beginning-of-statement)
- (beginning-of-line 1)
- (looking-at-p
- python-nav-beginning-of-defun-regexp)))
- (< starting-pos
- (save-excursion
- (let ((min-indent
- (+ (current-indentation)
- python-indent-offset)))
- (if (< starting-indentation min-indent)
- ;; If the starting indentation is not
- ;; within the min defun indent make the
- ;; check fail.
- starting-pos
- ;; Else go to the end of defun and add
- ;; up the current indentation to the
- ;; ending position.
- (python-nav-end-of-defun)
- (+ (point)
- (if (>= (current-indentation) min-indent)
- (1+ (current-indentation))
- 0)))))))))
+ (if first-run
+ (or (save-excursion
+ ;; If this is the first run, we may add
+ ;; the current defun at point.
+ (goto-char starting-pos)
+ (python-nav-beginning-of-statement)
+ (beginning-of-line 1)
+ (looking-at-p
+ python-nav-beginning-of-defun-regexp))
+ (< (current-indentation) starting-indentation))
+ (or
+ ;; we are at the next enclosing defun
+ (not non-defun-indent)
+ ;; we are searching from a not-defun and do not match cases
+ ;; where a defun is at the same level as the not-defun we
+ ;; started from
+ (< (current-indentation) non-defun-indent)))))
(save-match-data (setq last-indent (current-indentation)))
(if (or (not include-type) type)
(setq names (cons (match-string-no-properties 1) names))
@@ -5650,7 +5639,41 @@ python-info-current-defun
(setq type (car match))
(setq names (cons (cadr match) names)))))
;; Stop searching ASAP.
- (and (= (current-indentation) 0) (throw 'exit t))))
+ (and (= (current-indentation) 0) (throw 'exit t))
+ (when first-run
+ ;; `python-nav-beginning-of-defun' will go to the previous defun regardless of indentation
+ ;; so on `first-run' we have to reset to `starting-pos' to ensure that `next-ind' will be
+ ;; calculated from the correct starting point
+ (setq first-run nil)
+ (goto-char starting-pos))
+ (let (found-defun next-ind)
+ (while (and (not found-defun) (>= (setq next-ind (- (current-indentation) python-indent-offset)) 0))
+ (progn
+ ;; skip intervening defuns at the same or greater indentation by
+ ;; searching backward to find the next line with less indentation
+ (re-search-backward (format "^[ ]\\{%s\\}[^ \n]" next-ind))
+ (setq non-defun-indent
+ ;; we have to record `non-defun-indent' so that
+ ;; we don't incorrectly match cases where a function
+ ;; is e.g. defined inside an if statement that is at
+ ;; the same level as another defun, e.g.
+ ;;
+ ;; def a():
+ ;; def b(): return
+ ;; if True:
+ ;; def c(): return
+ ;; if True:
+ ;; if True:
+ ;; def d(): return
+ ;;
+ ;; the while loop is needed to handle the multiply nested case
+ ;; otherwise it incorrectly matchs a.d when the point is in d
+ (if (python-info-looking-at-beginning-of-defun)
+ (progn
+ (forward-line)
+ (setq found-defun t)
+ nil)
+ (current-indentation))))))))
(and names
(concat (and type (format "%s " type))
(mapconcat #'identity names ".")))))))
diff --git a/test/lisp/progmodes/python-tests.el b/test/lisp/progmodes/python-tests.el
index e1b4c0a74c0..84ae784bff0 100644
--- a/test/lisp/progmodes/python-tests.el
+++ b/test/lisp/progmodes/python-tests.el
@@ -5223,6 +5223,26 @@ python-info-current-defun-4
(should (string= (python-info-current-defun t)
"def func"))))
+(ert-deftest python-info-current-defun-5 ()
+ "Ensure multiple nested non-defun lines are handle correctly."
+ (python-tests-with-temp-buffer
+ "
+def a():
+ def b(): return
+ if True:
+ def c(): return
+
+if True:
+ if True:
+ def d(): return
+"
+ (python-tests-look-at "def c")
+ (should (string= (python-info-current-defun) "a.c"))
+ (should (string= (python-info-current-defun t) "def a.c"))
+ (python-tests-look-at "def d")
+ (should (string= (python-info-current-defun) "d"))
+ (should (string= (python-info-current-defun t) "def d"))))
+
(ert-deftest python-info-current-symbol-1 ()
(python-tests-with-temp-buffer
"
--
2.41.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix python-info-current-defun performance for large defuns
2023-09-04 1:37 [PATCH] Fix python-info-current-defun performance for large defuns Tom Gillespie
@ 2023-09-04 11:49 ` Eli Zaretskii
2023-09-05 14:17 ` kobarity
0 siblings, 1 reply; 6+ messages in thread
From: Eli Zaretskii @ 2023-09-04 11:49 UTC (permalink / raw)
To: Tom Gillespie, kobarity; +Cc: emacs-devel
> From: Tom Gillespie <tgbugs@gmail.com>
> Date: Sun, 3 Sep 2023 18:37:34 -0700
>
> Hi,
> Here is a patch to fix a performance issue in
> python-info-current-defun that appears when the function
> is run on large defuns. The issue can appear to the user as a
> noticeable (~500ms) freeze when the point is in a large defun
> in a python buffer with which-function-mode enabled. More
> details in the commit message and comments in the code.
>
> I made the patch against the emacs-29 branch, but will also
> on master without issue.
Thanks.
kobarity, any comments?
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix python-info-current-defun performance for large defuns
2023-09-04 11:49 ` Eli Zaretskii
@ 2023-09-05 14:17 ` kobarity
2023-09-15 6:56 ` Eli Zaretskii
0 siblings, 1 reply; 6+ messages in thread
From: kobarity @ 2023-09-05 14:17 UTC (permalink / raw)
To: Tom Gillespie, Eli Zaretskii; +Cc: emacs-devel
Eli Zaretskii wrote:
> > From: Tom Gillespie <tgbugs@gmail.com>
> > Date: Sun, 3 Sep 2023 18:37:34 -0700
> >
> > Hi,
> > Here is a patch to fix a performance issue in
> > python-info-current-defun that appears when the function
> > is run on large defuns. The issue can appear to the user as a
> > noticeable (~500ms) freeze when the point is in a large defun
> > in a python buffer with which-function-mode enabled. More
> > details in the commit message and comments in the code.
> >
> > I made the patch against the emacs-29 branch, but will also
> > on master without issue.
>
> Thanks.
>
> kobarity, any comments?
Hi,
Thank you for the patch. I have tried it and have confirmed that it
improves performance. I think that not using the time-consuming
`python-nav-end-of-defun' has helped.
However, I found one unexpected behavior. The following code is an
example added as an ERT:
def a():
def b(): return
if True:
def c(): return
if True:
if True:
def d(): return
In this example, `python-info-current-defun' seems to return "a" if
the point is on the second line from the bottom. Could you please
take a look?
My personal impression is that 114 characters per line is a bit long.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix python-info-current-defun performance for large defuns
2023-09-05 14:17 ` kobarity
@ 2023-09-15 6:56 ` Eli Zaretskii
2023-09-15 18:20 ` Tom Gillespie
0 siblings, 1 reply; 6+ messages in thread
From: Eli Zaretskii @ 2023-09-15 6:56 UTC (permalink / raw)
To: tgbugs, kobarity; +Cc: emacs-devel
Ping! Tom, could you please respond to the comments, and perhaps post
an updated patch? I'd like to install this when it's ready.
Thanks.
> Date: Tue, 05 Sep 2023 23:17:48 +0900
> From: kobarity <kobarity@gmail.com>
> Cc: emacs-devel@gnu.org
>
>
> Eli Zaretskii wrote:
> > > From: Tom Gillespie <tgbugs@gmail.com>
> > > Date: Sun, 3 Sep 2023 18:37:34 -0700
> > >
> > > Hi,
> > > Here is a patch to fix a performance issue in
> > > python-info-current-defun that appears when the function
> > > is run on large defuns. The issue can appear to the user as a
> > > noticeable (~500ms) freeze when the point is in a large defun
> > > in a python buffer with which-function-mode enabled. More
> > > details in the commit message and comments in the code.
> > >
> > > I made the patch against the emacs-29 branch, but will also
> > > on master without issue.
> >
> > Thanks.
> >
> > kobarity, any comments?
>
> Hi,
> Thank you for the patch. I have tried it and have confirmed that it
> improves performance. I think that not using the time-consuming
> `python-nav-end-of-defun' has helped.
>
> However, I found one unexpected behavior. The following code is an
> example added as an ERT:
>
> def a():
> def b(): return
> if True:
> def c(): return
>
> if True:
> if True:
> def d(): return
>
> In this example, `python-info-current-defun' seems to return "a" if
> the point is on the second line from the bottom. Could you please
> take a look?
>
> My personal impression is that 114 characters per line is a bit long.
>
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix python-info-current-defun performance for large defuns
2023-09-15 6:56 ` Eli Zaretskii
@ 2023-09-15 18:20 ` Tom Gillespie
2023-09-16 14:51 ` kobarity
0 siblings, 1 reply; 6+ messages in thread
From: Tom Gillespie @ 2023-09-15 18:20 UTC (permalink / raw)
To: Eli Zaretskii; +Cc: kobarity, emacs-devel
[-- Attachment #1: Type: text/plain, Size: 1923 bytes --]
Here's an updated patch with the fix and additional test for the
case kobarity identified. I also shortened the long lines. Best!
Tom
On Thu, Sep 14, 2023 at 11:56 PM Eli Zaretskii <eliz@gnu.org> wrote:
>
> Ping! Tom, could you please respond to the comments, and perhaps post
> an updated patch? I'd like to install this when it's ready.
>
> Thanks.
>
> > Date: Tue, 05 Sep 2023 23:17:48 +0900
> > From: kobarity <kobarity@gmail.com>
> > Cc: emacs-devel@gnu.org
> >
> >
> > Eli Zaretskii wrote:
> > > > From: Tom Gillespie <tgbugs@gmail.com>
> > > > Date: Sun, 3 Sep 2023 18:37:34 -0700
> > > >
> > > > Hi,
> > > > Here is a patch to fix a performance issue in
> > > > python-info-current-defun that appears when the function
> > > > is run on large defuns. The issue can appear to the user as a
> > > > noticeable (~500ms) freeze when the point is in a large defun
> > > > in a python buffer with which-function-mode enabled. More
> > > > details in the commit message and comments in the code.
> > > >
> > > > I made the patch against the emacs-29 branch, but will also
> > > > on master without issue.
> > >
> > > Thanks.
> > >
> > > kobarity, any comments?
> >
> > Hi,
> > Thank you for the patch. I have tried it and have confirmed that it
> > improves performance. I think that not using the time-consuming
> > `python-nav-end-of-defun' has helped.
> >
> > However, I found one unexpected behavior. The following code is an
> > example added as an ERT:
> >
> > def a():
> > def b(): return
> > if True:
> > def c(): return
> >
> > if True:
> > if True:
> > def d(): return
> >
> > In this example, `python-info-current-defun' seems to return "a" if
> > the point is on the second line from the bottom. Could you please
> > take a look?
> >
> > My personal impression is that 114 characters per line is a bit long.
> >
[-- Attachment #2: 0001-Fix-python-info-current-defun-performance-for-large-.patch --]
[-- Type: text/x-patch, Size: 9213 bytes --]
From b906381079c4cea5b20986902e82e759756ae17d Mon Sep 17 00:00:00 2001
From: Tom Gillespie <tgbugs@gmail.com>
Date: Sun, 3 Sep 2023 15:09:17 -0700
Subject: [PATCH] Fix python-info-current-defun performance for large defuns
* lisp/progmodes/python.el (python-info-current-defun): Significant
performance improvement for large functions.
* test/lisp/progmodes/python-tests.el (python-info-current-defun-5):
Added test for multiply nested non-defun lines.
The previous pervious implementation had two issues. First, it checked
every single intervening defun in the file, even those at the same
nesting level. Second, when it found a putative parent defun it would
call python-nav-end-of-defun which is incredibly slow for large defuns
(e.g. python classes that are many thousands of lines long).
The new implementation avoids these issues by using re-search-backward
to find the containing deindented line. It also handles cases where
there are multiple nested deindented lines that are not defuns.
---
lisp/progmodes/python.el | 101 +++++++++++++++++++---------
test/lisp/progmodes/python-tests.el | 23 +++++++
2 files changed, 94 insertions(+), 30 deletions(-)
diff --git a/lisp/progmodes/python.el b/lisp/progmodes/python.el
index 4b940b3f13b..1ef9dcc90eb 100644
--- a/lisp/progmodes/python.el
+++ b/lisp/progmodes/python.el
@@ -5699,42 +5699,44 @@ python-info-current-defun
(starting-pos (point))
(first-run t)
(last-indent)
+ (non-defun-indent (python-info-looking-at-beginning-of-defun))
(type))
(catch 'exit
(while (python-nav-beginning-of-defun 1)
+ ;; `python-nav-beginning-of-defun' doesn't respect indentation
(when (save-match-data
(and
(or (not last-indent)
(< (current-indentation) last-indent))
- (or
- (and first-run
- (save-excursion
- ;; If this is the first run, we may add
- ;; the current defun at point.
- (setq first-run nil)
- (goto-char starting-pos)
- (python-nav-beginning-of-statement)
- (beginning-of-line 1)
- (looking-at-p
- python-nav-beginning-of-defun-regexp)))
- (< starting-pos
- (save-excursion
- (let ((min-indent
- (+ (current-indentation)
- python-indent-offset)))
- (if (< starting-indentation min-indent)
- ;; If the starting indentation is not
- ;; within the min defun indent make the
- ;; check fail.
- starting-pos
- ;; Else go to the end of defun and add
- ;; up the current indentation to the
- ;; ending position.
- (python-nav-end-of-defun)
- (+ (point)
- (if (>= (current-indentation) min-indent)
- (1+ (current-indentation))
- 0)))))))))
+ (if first-run
+ (or (save-excursion
+ ;; If this is the first run, we may add
+ ;; the current defun at point.
+ (goto-char starting-pos)
+ (python-nav-beginning-of-statement)
+ (beginning-of-line 1)
+ (looking-at-p
+ python-nav-beginning-of-defun-regexp))
+ (and
+ (> starting-indentation 0)
+ (save-excursion ; ensure no non-defun
+ ;; deindented lines in between start
+ ;; and `python-nav-beginning-of-defun'
+ (goto-char starting-pos)
+ (re-search-backward
+ (format "^[ ]\\{%s\\}[^ \n]"
+ (- starting-indentation
+ python-indent-offset)))
+ (looking-at-p
+ python-nav-beginning-of-defun-regexp))
+ (< (current-indentation) starting-indentation)))
+ (or
+ ;; we are at the next enclosing defun
+ (not non-defun-indent)
+ ;; we are searching from a not-defun and do not match cases
+ ;; where a defun is at the same level as the not-defun we
+ ;; started from
+ (< (current-indentation) non-defun-indent)))))
(save-match-data (setq last-indent (current-indentation)))
(if (or (not include-type) type)
(setq names (cons (match-string-no-properties 1) names))
@@ -5742,7 +5744,46 @@ python-info-current-defun
(setq type (car match))
(setq names (cons (cadr match) names)))))
;; Stop searching ASAP.
- (and (= (current-indentation) 0) (throw 'exit t))))
+ (and (= (current-indentation) 0) (throw 'exit t))
+ (when first-run
+ ;; `python-nav-beginning-of-defun' will go to the previous
+ ;; defun regardless of indentation so on `first-run' we
+ ;; have to reset to `starting-pos' to ensure that `next-ind'
+ ;; will be calculated from the correct starting point
+ (setq first-run nil)
+ (goto-char starting-pos))
+ (let (found-defun next-ind)
+ (while (and (not found-defun)
+ (>= (setq next-ind
+ (- (current-indentation)
+ python-indent-offset))
+ 0))
+ (progn
+ ;; search backward to find the next line with less
+ ;; indentation to skip defuns at same or greater indentation
+ (re-search-backward (format "^[ ]\\{%s\\}[^ \n]" next-ind))
+ (setq non-defun-indent
+ ;; we have to record `non-defun-indent' so that
+ ;; we don't incorrectly match cases where a function
+ ;; is e.g. defined inside an if statement that is at
+ ;; the same level as another defun, e.g.
+ ;;
+ ;; def a():
+ ;; def b(): return
+ ;; if 1:
+ ;; def c(): return
+ ;; if 2:
+ ;; if 3:
+ ;; def d(): return
+ ;;
+ ;; while loop needed to handle the multiply nested case
+ ;; otherwise it incorrectly matchs a.d when the point is in d
+ (if (python-info-looking-at-beginning-of-defun)
+ (progn
+ (forward-line)
+ (setq found-defun t)
+ nil)
+ (current-indentation))))))))
(and names
(concat (and type (format "%s " type))
(mapconcat #'identity names ".")))))))
diff --git a/test/lisp/progmodes/python-tests.el b/test/lisp/progmodes/python-tests.el
index 9f935f2748c..47391336193 100644
--- a/test/lisp/progmodes/python-tests.el
+++ b/test/lisp/progmodes/python-tests.el
@@ -5400,6 +5400,29 @@ python-info-current-defun-4
(should (string= (python-info-current-defun t)
"def func"))))
+(ert-deftest python-info-current-defun-5 ()
+ "Ensure multiple nested non-defun lines are handle correctly."
+ (python-tests-with-temp-buffer
+ "
+def a():
+ def b(): return
+ if 1:
+ def c(): return
+
+if 2:
+ if 3:
+ def d(): return
+"
+ (python-tests-look-at "def c")
+ (should (string= (python-info-current-defun) "a.c"))
+ (should (string= (python-info-current-defun t) "def a.c"))
+ (python-tests-look-at " if 3:")
+ (should (not (python-info-current-defun)))
+ (should (not (python-info-current-defun t)))
+ (python-tests-look-at "def d")
+ (should (string= (python-info-current-defun) "d"))
+ (should (string= (python-info-current-defun t) "def d"))))
+
(ert-deftest python-info-current-symbol-1 ()
(python-tests-with-temp-buffer
"
--
2.41.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix python-info-current-defun performance for large defuns
2023-09-15 18:20 ` Tom Gillespie
@ 2023-09-16 14:51 ` kobarity
0 siblings, 0 replies; 6+ messages in thread
From: kobarity @ 2023-09-16 14:51 UTC (permalink / raw)
To: Tom Gillespie; +Cc: Eli Zaretskii, emacs-devel
Tom Gillespie wrote:
> Here's an updated patch with the fix and additional test for the
> case kobarity identified. I also shortened the long lines. Best!
Thank you for the updated patch. However, there still seems to be a
bit of a problem.
def a():
def b(): return
if 1:
def c(): return
z = 1
When the point is at the "z = 1" line, python-info-current-defun
should return "a", but the new function returns nil.
I also noticed another subtle difference in behavior with the example
added as an ERT:
def a():
def b(): return
if 1:
def c(): return
if 2:
if 3:
def d(): return
If the point is on the empty line above "if 2:", the new function
returns "a", while the current function returns nil.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2023-09-16 14:51 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-09-04 1:37 [PATCH] Fix python-info-current-defun performance for large defuns Tom Gillespie
2023-09-04 11:49 ` Eli Zaretskii
2023-09-05 14:17 ` kobarity
2023-09-15 6:56 ` Eli Zaretskii
2023-09-15 18:20 ` Tom Gillespie
2023-09-16 14:51 ` kobarity
Code repositories for project(s) associated with this public inbox
https://git.savannah.gnu.org/cgit/emacs.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).