Bug 148433 - Extreme keyboard input delays on linux in paragraphs which contain both rtl and ltr characters
Summary: Extreme keyboard input delays on linux in paragraphs which contain both rtl a...
Status: NEW
Alias: None
Product: LibreOffice
Classification: Unclassified
Component: Writer (show other bugs)
Version:
(earliest affected)
5.4 all versions
Hardware: All Linux (All)
: low normal
Assignee: Not Assigned
URL:
Whiteboard: target:7.4.0
Keywords: bibisected, regression, text:rtl
Depends on:
Blocks: RTL-UI
  Show dependency treegraph
 
Reported: 2022-04-07 01:04 UTC by Yotam Benshalom
Modified: 2024-08-17 10:13 UTC (History)
5 users (show)

See Also:
Crash report or crash signature:


Attachments
Input delay testcase (11.56 KB, application/vnd.openxmlformats-officedocument.wordprocessingml.document)
2022-04-07 01:05 UTC, Yotam Benshalom
Details
Input selay testcase - corrcet version (11.65 KB, application/vnd.openxmlformats-officedocument.wordprocessingml.document)
2022-04-07 01:38 UTC, Yotam Benshalom
Details
Keyboard input delay in the 2nd line only (10.14 KB, application/vnd.oasis.opendocument.text)
2022-04-07 18:04 UTC, Yotam Benshalom
Details
Keyboard input delay in the 2nd line only (10.14 KB, application/vnd.oasis.opendocument.text)
2022-04-07 18:04 UTC, Yotam Benshalom
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Yotam Benshalom 2022-04-07 01:04:13 UTC
Description:
There are extreme keyboard input delays (when typing characters and/or navigating with the arrow keys) in paragraphs which contain both rtl and numeric characters in DOCX files.

Since the arrow keys and the characters are handled in independent queues, for some reason, this bug makes it very difficult to edit such files in libreoffice. The characters typed are often inserted before the arrows even if the user typed the arrow first.

Steps to Reproduce:
1.Open the attached file in writer.
2.Try to move the cursor along each of the four paragraphs using the arrow keys.

Actual Results:
In the second paragraph (the only one containing both rtl and numeric characters) the cursor moves extremely slowly and makes uncontrollable jumps. In the other three paragraphs the cursor moves as expected.

Expected Results:
No input delay should occur in the second paragraph.


Reproducible: Always


User Profile Reset: No



Additional Info:
This is tested on ubuntu 22.04 with gnome 42, but I suspect that this problem is platform-independent.
Comment 1 Yotam Benshalom 2022-04-07 01:05:09 UTC
Created attachment 179363 [details]
Input delay testcase
Comment 2 Yotam Benshalom 2022-04-07 01:37:45 UTC
Sorry, incomplete attachment. I add the complete one here.
Comment 3 Yotam Benshalom 2022-04-07 01:38:46 UTC
Created attachment 179364 [details]
Input selay testcase - corrcet version
Comment 4 Yotam Benshalom 2022-04-07 18:02:53 UTC
Upon further examination, this problem exists in original .odt files as well.
Testcase is updated.
Comment 5 Yotam Benshalom 2022-04-07 18:04:07 UTC
Created attachment 179384 [details]
Keyboard input delay in the 2nd line only
Comment 6 Yotam Benshalom 2022-04-07 18:04:58 UTC
Created attachment 179385 [details]
Keyboard input delay in the 2nd line only
Comment 7 Telesto 2022-04-07 23:22:38 UTC
I'm not seeing any delay with
Version: 7.4.0.0.alpha0+ / LibreOffice Community
Build ID: fbf739198aa7f02975d531521c6525073783c7f1
CPU threads: 8; OS: Mac OS X 12.2.1; UI render: Skia/Metal; VCL: osx
Locale: nl-NL (nl_NL.UTF-8); UI: en-US
Calc: threaded
Comment 8 Yotam Benshalom 2022-04-07 23:39:17 UTC
The delay is unmistakable when pressing the left arrow key continuously when the cursor is at the right end of the second rtl paragraph in the testcase.

I experience the problem on ubuntu with xorg and the nvidia proprietary driver (hardware acceleration disabled in LO options). Maybe this is related to one of these frameworks.

I'll try the 7.4 version here and see if it solves the problem.
Comment 9 Yotam Benshalom 2022-04-08 00:21:06 UTC
I checked out master from git. The problem definitely exists also in: 

Version: 7.4.0.0.alpha0+ / LibreOffice Community
Build ID: 3e94991d7fd3a085549c3a5d4c991688042d2cb9
CPU threads: 16; OS: Linux 5.15; UI render: default; VCL: gtk3
Locale: he-IL (he_IL.UTF-8); UI: en-US
Misc: CAlc: CL
Comment 10 Yotam Benshalom 2022-04-10 12:50:05 UTC
This bug does not occur on windows, and according to Telesto it does not occur on mac too. So, this seem to be a linux-only problem.
Comment 11 Yotam Benshalom 2022-04-10 19:23:18 UTC
The issue is noted in /var/log/Xorg.0.log:

[   865.856] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 32ms, your system is too slow
[  3226.007] (EE) client bug: timer event3 debounce: scheduled expiry is in the past (-90ms), your system is too slow
[  3274.957] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 21ms, your system is too slow
[  3286.328] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 21ms, your system is too slow
[  3392.073] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 33ms, your system is too slow
[  3618.939] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 24ms, your system is too slow
[  3618.939] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: WARNING: log rate limit exceeded (5 msgs per 60min). Discarding future messages.
[  6531.198] (EE) client bug: timer event3 debounce: scheduled expiry is in the past (-58ms), your system is too slow
[  6531.198] (EE) client bug: timer event3 debounce short: scheduled expiry is in the past (-71ms), your system is too slow
[  6924.439] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 36ms, your system is too slow
[  7090.677] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 33ms, your system is too slow
[  7292.991] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 23ms, your system is too slow
[  7654.182] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 22ms, your system is too slow
[  7767.279] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: client bug: event processing lagging behind by 24ms, your system is too slow
[  7767.279] (EE) event6  - Microsoft Microsoft® 2.4GHz Transceiver v8.0: WARNING: log rate limit exceeded (5 msgs per 60min). Discarding future messages.
Comment 12 Scott Clewell 2022-04-19 21:14:31 UTC
Hello Yotam and thank you for reporting.

I can reproduce this bug by holding down the arrow keys (single presses work normal) on the second paragraph using:

Version: 7.3.2.2 / LibreOffice Community
Build ID: 30(Build:2)
CPU threads: 16; OS: Linux 5.13; UI render: default; VCL: gtk3
Locale: en-US (en_US.UTF-8); UI: en-US
Ubuntu package version: 1:7.3.2~rc2-0ubuntu0.20.04.1~lo1
Calc: threaded

and using:
Version: 7.4.0.0.alpha0+ / LibreOffice Community
Build ID: cb56edb177f4db5b9cc4d140543c4b11d41ef1b0
CPU threads: 16; OS: Linux 5.13; UI render: default; VCL: gtk3
Locale: en-US (en_US.UTF-8); UI: en-US
Calc: CL

Not reproducible using:
LO 4.2.8.2, Ubuntu 14.04 LTS
Comment 13 Yotam Benshalom 2022-04-20 14:32:18 UTC
This regression was introduced in the 5.4 series.

I bibisected it:

Author: Jenkins Build User <tdf@pollux.tdf>
Date:   Thu Feb 16 10:27:58 2017 +0100

    source 7863798f2ab420fdb91ba3dcda6cea2ab2aded9d
    
    source 7863798f2ab420fdb91ba3dcda6cea2ab2aded9d

 instdir/program/libvcllo.so          | Bin 10394213 -> 10394438 bytes
 instdir/program/libvclplug_gtk3lo.so | Bin 1031387 -> 1031227 bytes
 instdir/program/versionrc            |   2 +-
 3 files changed, 1 insertion(+), 1 deletion(-)
Comment 14 Yotam Benshalom 2022-04-20 14:45:30 UTC
The first commit to initiate the problem can be seen here:
https://cgit.freedesktop.org/libreoffice/core/commit/?id=7863798f2ab420fdb91ba3dcda6cea2ab2aded9d
Comment 15 Yotam Benshalom 2022-04-20 20:24:13 UTC
On further investigation, the delay occurs whenever rtl and ltr characters exist in the same paragraph, not only in the specific case of numeric characters.
Comment 16 Caolán McNamara 2022-04-25 16:09:10 UTC
I can't reproduce, but I only have integrated Intel graphics. Does it make any difference if shift is held down to create a selection instead of just changing the input position? And is there any input methods enabled? (though using a Hebrew IM didn't make a difference when I tried it locally) And does changing the font make any difference?

The only thing that comes immediately to mind with specifically digits having anything special about them is "GetLocalizedChar" in vcl which shouldn't matter for this language. If someone can reproduce it and is able to build libreoffice they could see if deleting the block of code in vcl/source/outdev/text.cxx at approx line 1190, the "if( meTextLanguage )" block makes any difference.

There was trouble in the past with CAIRO_OPERATOR_DIFFERENCE and nvidia cards, again if someone can reproduce and can build libreoffice then they could see if disabling the line "cairo_set_operator(cr, CAIRO_OPERATOR_DIFFERENCE)" in vcl/headless/svpgdi.cxx makes a difference.

wrt "hardware acceleration disabled in LO options" that should only affect full screen impress presentation mode FWIW
Comment 17 Yotam Benshalom 2022-04-25 17:46:39 UTC
Holding shift does not make any difference.

No input method is defined in settings->area and language->manage installed languages. I think they are not relevant to rtl languages. Changing the keyboard input language has no effect.

Changing font (using fonts from various sources) makes no difference.

Please note that on further investigation I found that this happens whenever rtl and ltr (e.g., latin) characters are mixed together. This is not exclusive to the case of rtl and numeric characters in the same paragraph.
Comment 18 Caolán McNamara 2022-04-25 19:10:28 UTC
the remaining leading contender for investigation is then that CAIRO_OPERATOR_DIFFERENCE call, there was difficulties with it in the past on nvidia. I will push a commit which will allow experimenting with the next daily that contains it to see if that has a bearing on this problem or not.
Comment 19 Yotam Benshalom 2022-04-25 19:16:49 UTC
Thank you. Will you let me know how to install libreoffice from daily?
Comment 20 Commit Notification 2022-04-26 09:37:35 UTC
Caolán McNamara committed a patch related to this issue.
It has been pushed to "master":

https://git.libreoffice.org/core/commit/90057e372cd07b41c8dbba4e7b7f107568a5f451

Related: tdf#148433 test if CAIRO_OPERATOR_DIFFERENCE is the problem

It will be available in 7.4.0.

The patch should be included in the daily builds available at
https://dev-builds.libreoffice.org/daily/ in the next 24-48 hours. More
information about daily builds can be found at:
https://wiki.documentfoundation.org/Testing_Daily_Builds

Affected users are encouraged to test the fix and report feedback.
Comment 21 Caolán McNamara 2022-04-26 09:41:17 UTC
The link above gives the locations of dailies, i.e. .https://dev-builds.libreoffice.org/daily/master/current.html

They are dailies, so it won't be available until at least tomorrow. To test, in a terminal

export SAL_DISABLE_CAIRO_DIFFERENCE=1
and run libreoffice from the command line from that terminal, the cursor should then be invisible, but selection will still work,
Comment 22 Yotam Benshalom 2022-04-28 10:41:22 UTC
I confirm that running the current master version with export SAL_DISABLE_CAIRO_DIFFERENCE=1 makes the issue go away (although, as you said, the cursor is invisible).
Comment 23 Caolán McNamara 2022-04-28 11:25:33 UTC
aha!, well we have some useful data now
Comment 24 Commit Notification 2022-05-02 16:38:35 UTC
Caolán McNamara committed a patch related to this issue.
It has been pushed to "master":

https://git.libreoffice.org/core/commit/00e43c14558fa4d0369820fa3d2cd7b987e61377

Related: tdf#148433 experiment with CAIRO_OPERATOR_EXCLUSION

It will be available in 7.4.0.

The patch should be included in the daily builds available at
https://dev-builds.libreoffice.org/daily/ in the next 24-48 hours. More
information about daily builds can be found at:
https://wiki.documentfoundation.org/Testing_Daily_Builds

Affected users are encouraged to test the fix and report feedback.
Comment 25 Caolán McNamara 2022-05-02 16:49:21 UTC
With this experiment then now with
export SAL_DISABLE_CAIRO_DIFFERENCE=1
there should be a visible cursor (if its not visible then the change isn't in the build yet). With that visible cursor is there any performance difference than without SAL_DISABLE_CAIRO_DIFFERENCE set ?
Comment 26 Yotam Benshalom 2022-05-03 21:34:05 UTC
I installed the current version from master (LibreOfficeDev 7.4.0.0.alpha0 69f412459257398a7d779541dad95aa7227a5e74). After running: 

export SAL_DISABLE_CAIRO_DIFFERENCE=1 && /opt/libreofficedev7.4/program/soffice

I can see the cursor, but the issue, unfortunately, is not resolved.
Comment 27 Caolán McNamara 2022-05-05 20:07:44 UTC
Seeing as the problem is associated with mixed RTL and LTR characters and that it appears to be the cursor causing is I suspect the little directional arrow at the top of the cursor is the trigger for the problem.
Comment 28 Commit Notification 2022-05-06 11:11:29 UTC
Caolán McNamara committed a patch related to this issue.
It has been pushed to "master":

https://git.libreoffice.org/core/commit/c7a32dab567cce8ee435c65299ec60b638700a35

Related: tdf#148433 use a slightly different cursor directional indicator

It will be available in 7.4.0.

The patch should be included in the daily builds available at
https://dev-builds.libreoffice.org/daily/ in the next 24-48 hours. More
information about daily builds can be found at:
https://wiki.documentfoundation.org/Testing_Daily_Builds

Affected users are encouraged to test the fix and report feedback.
Comment 29 Caolán McNamara 2022-05-06 11:21:28 UTC
I think it's worth improving the look of the direction indicator on the cursor in any case. It's a total longshot if changing the cursor arrow to be slightly larger and an equilateral triangle has any knock on improvement on this performance problem. I suspect it's unlikely, but worth checking when this minor tweak becomes available for testing.
Comment 30 Yotam Benshalom 2022-05-07 07:44:12 UTC
With the newest build (LibreOfficeDev 7.4.0.0.alpha0 f15e6293cf78d67963a6e512f60a11ae58da72c5), the issue is not resolved and behaviour is the same.

However, something is weird: The cursor directionality triangle now appears only on the mixed paragraph (ltr and rtl). On the paragraphs that are rtl-only or ltr-only there is no triangle, only vertical line.
Comment 31 Caolán McNamara 2022-05-09 07:47:02 UTC
(In reply to Yotam Benshalom from comment #30)
> However, something is weird: The cursor directionality triangle now appears
> only on the mixed paragraph (ltr and rtl). On the paragraphs that are
> rtl-only or ltr-only there is no triangle, only vertical line.

I believe that is the way it always was, at least it was like that before my change to make the arrow more prominent.
Comment 32 Yotam Benshalom 2022-05-09 07:56:45 UTC
Oh! If the direction triangle appears only in mixed paragraphs, it probably means it is the culprit. Shouldn't we see what happens if we drop it completely?
Also, is there a way for me to help in further debugging the nature of the delay? If it represents a bug in cairo or in nvidia proprietary driver, the results might be useful.
Comment 33 Caolán McNamara 2022-05-09 17:03:06 UTC
(In reply to Yotam Benshalom from comment #32)
> Oh! If the direction triangle appears only in mixed paragraphs, it probably
> means it is the culprit.

That's my guess.

> Shouldn't we see what happens if we drop it completely?

I will push another environmental variable hack to clarify if not drawing the triangle part of the BiDi cursor would make a difference.

> Also, is there a way for me to help in further debugging the nature of the
> delay? If it represents a bug in cairo or in nvidia proprietary driver, the
> results might be useful.

The only hits I ever get on searching for this problem is similar cases to this in the past in LibreOffice. It always seems to be nvidia related as far as I can tell. I've never encountered it myself.
Comment 34 Commit Notification 2022-05-09 20:30:48 UTC
Caolán McNamara committed a patch related to this issue.
It has been pushed to "master":

https://git.libreoffice.org/core/commit/6dc41e33e087032a52ee11832ff1299632b4f5dd

Related: tdf#148433 experiment with SAL_DISABLE_CURSOR_INDICATOR

It will be available in 7.4.0.

The patch should be included in the daily builds available at
https://dev-builds.libreoffice.org/daily/ in the next 24-48 hours. More
information about daily builds can be found at:
https://wiki.documentfoundation.org/Testing_Daily_Builds

Affected users are encouraged to test the fix and report feedback.
Comment 35 Yotam Benshalom 2022-05-10 19:24:30 UTC
The problem indeed disappears, along with the cursor indicator, when running:
export SAL_DISABLE_CURSOR_INDICATOR=1 && /opt/libreofficedev7.4/program/soffice

In case the issue is not resolved, can this behaviour stay in place? One of my jobs is adding diacritics to entire books, and this bug makes it almost impossible.

FWIW, I don't really understand the reason behind the current implementation of the cursor direction indicator. I don't need an indicator telling me whether an existing character is rtl or not, because I can see that for myself. What I do need (or at least would appreciate) is an indicator that would tell me whether the character I am about to type is ltr or rtl, based on my current keyboard input language - exactly the way it is implemented in the comment textbox I am using right now via firefox. If I am not mistaken, this is the way ms word does it as well.
Comment 36 Caolán McNamara 2022-05-11 15:50:55 UTC
(In reply to Yotam Benshalom from comment #35)
> The problem indeed disappears, along with the cursor indicator, when running:
> export SAL_DISABLE_CURSOR_INDICATOR=1 &&
> /opt/libreofficedev7.4/program/soffice
> 
> In case the issue is not resolved, can this behaviour stay in place?

the variable for workaround can certainly stay in place until some proper solution is arrived at.

> FWIW, I don't really understand the reason behind the current implementation
> of the cursor direction indicator.

RTL is not my specialty, my understanding is that if the arrow is pointing left then if a RTL char is entered it will appear at the point indicated to the left, if the arrow is pointing right then if a LTR char is entered it will appear at the point to the right. Something like gedit instead uses a split-cursor and shows two cursors with arrows to show both locations a char might appear if there is ambiguity where a char would appear depending on its classification once entered.
Comment 37 Eyal Rozenberg 2024-08-17 10:13:28 UTC
Keyboard input delay - RTL-UI