Help with an old regression in text rendering needed

Milos Sramek sramek.milos at
Thu Aug 11 08:02:09 UTC 2016


thank you for your response and for pointing me towards kerning. It
looks like the problem resides in kerning and not in harfbuzz.

LO ignores setting of kerning in docx documents: disregarding its
setting in the input document the setting (Character->Position->Pair
Kerning) is always ON and when saved it is in the output file always OFF
(verified in MSOffice). This has been so in LO since version 3.3
(verified right now) and is still so in the most recent one.

Therefore,  roundtrip tests (open docx in LO, save, open saved on MSO
and compare with the original docx) with documents with kerning set to
OFF always pass and documents with kerning set to ON always fail.
The situation is different in print test (open original and print by LO
and compare with printed by MSO). Bibisection shows that up to revision
59a69f7 in the bibisect-43all repository documents with Kerning ON fail
and those with kerning OFF pass.

>From the bibisection log:
* 59a69f7 tag: source-hash-48ad2f61fe71edc1a8967b322d3e0f368f4be06f,
* 96f14f1 -tag: source-hash-f0393d7ff69011a16b100541ef18e5090544e4a1,

I wanted to verify this by compiling the sources (I've tried
f0393d7ff69011a16b100541ef18e5090544e4a1), but compilation failed in
unit tests. I have used gcc-4.8  and  switches from the file autogen.log:

CC=/usr/bin/gcc-4.8 CXX=/usr/bin/g++-4.8 ./ 
--disable-dependency-tracking --disable-linkoo --disable-option-checking
--without-junit --without-help --without-myspell-dicts --without-doxygen
--disable-gnome-vfs --disable-odk --without-system-jpeg

Can I switch the unit test off somehow?

Maybe I do something wrong, since the first bad commit:
does not seem to be related to docx at all :(

My results and also files with kerning switched ON and OFF are attached.

Thank in advance for your help


On 2016-07-29 14:09, Caolán McNamara wrote:
> On Fri, 2016-07-29 at 10:34 +0200, Milos Sramek wrote:
>> Thank you, Caolán, for your suggestions.
>> The first thing I've tried was to compile the current version with
>> and without harfbuzz using
>> ./ --with-harfbuzz=no
>> The result was the same (bad).
>> I am, however, not sure if harfbuzz was really switched off
> Yeah, you can't turn off harfbuzz under Linux, the option is only to
> enable turning it *on* for other platforms since it became the only
> implementation under Linux.
>> I've tried to investigate it bit deeper:
>> revision mentioned:
>> bff8fa9 - (2013-05-06 11:08:29 +0200) Enable HarfBuzz by default -
>> Khaled Hosny
>> The last good revision according to my bibisection
>> f0393d7 - (2013-05-06 16:54:53 +0200) [harfbuzz] Fix text width
>> calculation, 3rd try - Khaled Hosny
> Note that the very next commit after f0393d7 is actually bff8fa9, so
> you can probably short circuit bisecting and jump straight
> between bff8fa9 and f0393d7
>> The first bad revision according to my bibisection
>> 48ad2f6 - (2013-05-06 12:48:23 +0200) fdo#46808, Convert some code to
>> getProcessComponentContext
>> So, they really are from the same day.
>> I unfortunately cannot compile these revisions, maybe because my
>> compiter is too new (gcc version 5.4.0 20160609 (Ubuntu
>> 5.4.0-6ubuntu1~16.04.1)). I happens because of std::auto_ptr:
>> /workdir/
>> r.hpp:66:31:
>> warning: ‘template<class> class std::auto_ptr’ is deprecated
>> [-Wdeprecated-declarations]
> That's only a warning. So unless you have used --enable-werror to makes
> warnings fatal then that warning shouldn't be a problem. But seeing as
> f0393d7 built, presumably bff8fa9 (because its the next commit) builds
> as well for comparison.
> C.
> _______________________________________________
> LibreOffice mailing list
> LibreOffice at

email & jabber: sramek.milos at

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: File999-results.ods
Type: application/vnd.oasis.opendocument.spreadsheet
Size: 15570 bytes
Desc: not available
URL: <>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: File_999l-kernon.docx
Type: application/vnd.openxmlformats-officedocument.wordprocessingml.document
Size: 13764 bytes
Desc: not available
URL: <>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: File_999l-kernoff.docx
Type: application/vnd.openxmlformats-officedocument.wordprocessingml.document
Size: 13681 bytes
Desc: not available
URL: <>

More information about the LibreOffice mailing list