Bug 112849 - Kashida characters appears at particular zoom levels (310%)
Summary: Kashida characters appears at particular zoom levels (310%)
Alias: None
Product: LibreOffice
Classification: Unclassified
Component: Writer (show other bugs)
(earliest affected)
Hardware: All All
: medium normal
Assignee: ⁨خالد حسني⁩
Whiteboard: target:7.5.0 inReleaseNotes:7.5
Keywords: bibisected, bisected, implementationError
Depends on: 104921
Blocks: Font-Rendering RTL-Arabic-and-Farsi Kashida-Justification
  Show dependency treegraph
Reported: 2017-10-02 20:15 UTC by Yousuf Philips (jay) (retired)
Modified: 2022-12-07 15:45 UTC (History)
5 users (show)

See Also:
Crash report or crash signature:

kashidas appear in red and period in green disappears (182.00 KB, image/png)
2017-10-02 20:15 UTC, Yousuf Philips (jay) (retired)

Note You need to log in before you can comment on or make changes to this bug.
Description Yousuf Philips (jay) (retired) 2017-10-02 20:15:01 UTC
Created attachment 136712 [details]
kashidas appear in red and period in green disappears

1. open attachment 132839 [details]
2. zoom in to 310 percent and notice kashida's appearing in various places and notice periods disappearing

regression as this doesnt happen in 5.3.7.

Build ID: a2a3e06a29077d4274dc15eea28a01afe22e3658
CPU threads: 2; OS: Linux 4.4; UI render: default; VCL: gtk2; 
Locale: en-US (en_US.UTF-8); Calc: group
Comment 1 Xisco Faulí 2017-10-02 21:43:16 UTC
Regression introduced by:

author	Miklos Vajna <vmiklos@collabora.co.uk>	2017-03-30 07:27:45 (GMT)
committer	Miklos Vajna <vmiklos@collabora.co.uk>	2017-03-30 10:19:02 (GMT)
commit 38b0c24fa5cbb4246e03d77ac022dfdc9fdede03 (patch)
tree 6c500a9f343126a28508aaa2adc581d819a2eb96
parent a82627aa787e80c9473453acd5ec955b6cf265ff (diff)
Related: tdf#105454 DOCX import: fix unwanted enabled-by-default kerning
Similar to DOC, DOCX doesn't enable kerning by default.

Bisected with: bibisect-linux-64-5.4

Adding Cc: to Miklos Vajna
Comment 2 Miklos Vajna 2017-10-03 07:46:03 UTC
I would argue this worked only by accident before. See bug 105454, DOC/DOCX does not enable kerning by default, enabling it by default for those documents in Writer was a bug that caused real problems.

If this change causes problems for Kashida, then it's just a coincidence that it worked previously. Somebody who understands complex text layout should find out how Word manages to handle that without kerning and based on that what we could do to improve the situation.

What do you think? I'm OK with looking at this at some stage, but I think "real" regressions are a priority.
Comment 3 Yousuf Philips (jay) (retired) 2017-10-03 16:20:09 UTC
Lets ask Khaled as he knows best about CTL and font rendering.
Comment 4 ⁨خالد حسني⁩ 2017-10-03 18:12:39 UTC
Kashida in LibreOffice is fundamentally broken:

// A note on how Kashida justification is implemented (because it took me 5
// years to figure it out):
// The decision to insert Kashidas, where and how much is taken by Writer.
// This decision is communicated to us in a very indirect way; by increasing
// the width of the character after which Kashidas should be inserted by the
// desired amount.
// Writer eventually calls IsKashidaPosValid() to check whether it can insert a
// Kashida between two characters or not.
// Here we do:
// - In LayoutText() set KashidaJustification flag based on text script.
// - In ApplyDXArray():
//   * Check the above flag to decide whether to insert Kashidas or not.
//   * For any RTL glyph that has DX adjustment, insert enough Kashidas to
//     fill in the added space.

from https://gerrit.libreoffice.org/gitweb?p=core.git;a=blob;f=vcl/source/gdi/CommonSalLayout.cxx;h=8e8565717932088558ee7a4b31fc0bfe405e98e3;hb=HEAD#l698

The insanity described above together with the the whole DX array madness and the use of integers to hold glyph positions, makes is so that any rounding errors would cause VCL to think that writer wants to insert a kashida in this place.

Fixing kashida once and for all requires 1) making the communication between Writer and VCL explicit e.g. have a separate “kashida insertion points” array 2) Kill DX array for good and enter the 21 century (it is an API “inspired” by Windows 3.1 3) Fix rounding errors by preserving fractional glyph positions in a way or another.
Comment 5 Miklos Vajna 2017-10-04 07:58:18 UTC
Thanks for the input, Khaled. None of the stuff you mention is introduced by the above commit, so let me re-tag this from regression to implementationError.
Comment 6 QA Administrators 2019-02-19 03:43:20 UTC Comment hidden (noise)
Comment 7 QA Administrators 2021-02-19 04:22:34 UTC Comment hidden (noise)
Comment 9 Stéphane Guillou (stragu) 2022-12-07 15:45:58 UTC
Can see the artifacts at 310% and 490% in:

Version: / LibreOffice Community
Build ID: 8d71d29d553c0f7dcbfa38fbfda25ee34cce99a2
CPU threads: 8; OS: Linux 5.15; UI render: default; VCL: gtk3
Locale: en-AU (en_AU.UTF-8); UI: en-US
Calc: threaded

But they are already gone in:

Version: / LibreOffice Community
Build ID: e114eadc50a9ff8d8c8a0567d6da8f454beeb84f
CPU threads: 8; OS: Linux 5.15; UI render: default; VCL: gtk3
Locale: en-AU (en_AU.UTF-8); UI: en-US
Calc: threaded

Not sure if there was an earlier fix?
In any case, verified fixed in 7.5:

Version: (X86_64) / LibreOffice Community
Build ID: da3dd48eaf9086f8ab28d6a6655f9a638e51433a
CPU threads: 8; OS: Linux 5.15; UI render: default; VCL: gtk3
Locale: en-AU (en_AU.UTF-8); UI: en-US
Calc: threaded