A user asked for right-to-left support in a Play Store review of 4.18.0, in Arabic, rating the app 5 ★ anyway. #468 is the same request from a support email, and #108 asked for it in 2019 and was closed without it. This is what is actually missing, because it is three separate things and only one of them is a feature.
The fix is in odrcore in all three cases; nothing on the Android side is involved. odrcore has #326 for the spreadsheet half of this.
The view has no base direction
HtmlWriter::write_begin writes a bare <html> (html_writer.cpp:134). No dir, no lang, anywhere in the output. So the WebView takes the default base direction, which is left-to-right.
That is not "Arabic does not render". The Unicode bidi algorithm still orders an Arabic run right-to-left inside a line, so a paragraph of pure Arabic reads correctly. What breaks is everything that depends on the paragraph's own direction: the sentence lands against the left edge, trailing punctuation goes to the wrong end, and a paragraph mixing Arabic with Latin words or digits is ordered by the wrong base.
style:writing-mode is never read
ODF states paragraph and page direction as style:writing-mode, rl-tb being right-to-left. Nothing parses it — there is no writing_mode anywhere in internal/odf/, and ParagraphStyle (style.hpp:166) has no field to put it in. The only writing-mode the translator ever emits is vertical-lr for a rotated table cell (document_style.cpp:415), which is unrelated.
So even with a base direction on <html>, a right-to-left paragraph inside an otherwise left-to-right document would still come out wrong, and vice versa.
start and end alignment resolve to the wrong edge
This one is a bug rather than a gap, and it is the visible one.
read_text_align (odf_style.cpp:108) maps fo:text-align values start → TextAlign::left and end → TextAlign::right, unconditionally. But start and end are relative to the writing direction: in a right-to-left paragraph start is the right edge. translate_paragraph_style (document_style.cpp:237) then writes that out as a literal text-align:left.
start is what writers emit for ordinary body text rather than left, so this hits the common case: a right-to-left document whose paragraphs are simply default-aligned renders flush left, and it stays flush left even if the base direction is fixed, because the wrong edge was baked in at parse time.
What we could do
TextAlign has the shape of the problem in it — left/right/center/justify, with no start/end. Two ways out:
- Keep
start/end all the way to the CSS. Add them to TextAlign and emit text-align:start / text-align:end, which browsers resolve against the element's direction. Then parse style:writing-mode into a direction on ParagraphStyle, emit it as direction:rtl, and the alignment follows for free. The public enum gains two values, so this is a breaking change for the bindings.
- Resolve at translation time, once direction is known, and leave the enum alone. Cheaper, and it puts the writing direction into the style model either way — so it is most of the same work with a worse result.
Either way <html dir> (or a direction on the document root element) still has to be set from the document's page style, or a document that states nothing per paragraph gets no direction at all.
Worth deciding whether the same pass carries lang, which is in the ODF too and which nothing currently emits — it is what a screen reader and hyphenation need, and it is the same attribute pair.
A user asked for right-to-left support in a Play Store review of 4.18.0, in Arabic, rating the app 5 ★ anyway. #468 is the same request from a support email, and #108 asked for it in 2019 and was closed without it. This is what is actually missing, because it is three separate things and only one of them is a feature.
The fix is in odrcore in all three cases; nothing on the Android side is involved. odrcore has #326 for the spreadsheet half of this.
The view has no base direction
HtmlWriter::write_beginwrites a bare<html>(html_writer.cpp:134). Nodir, nolang, anywhere in the output. So the WebView takes the default base direction, which is left-to-right.That is not "Arabic does not render". The Unicode bidi algorithm still orders an Arabic run right-to-left inside a line, so a paragraph of pure Arabic reads correctly. What breaks is everything that depends on the paragraph's own direction: the sentence lands against the left edge, trailing punctuation goes to the wrong end, and a paragraph mixing Arabic with Latin words or digits is ordered by the wrong base.
style:writing-modeis never readODF states paragraph and page direction as
style:writing-mode,rl-tbbeing right-to-left. Nothing parses it — there is nowriting_modeanywhere ininternal/odf/, andParagraphStyle(style.hpp:166) has no field to put it in. The onlywriting-modethe translator ever emits isvertical-lrfor a rotated table cell (document_style.cpp:415), which is unrelated.So even with a base direction on
<html>, a right-to-left paragraph inside an otherwise left-to-right document would still come out wrong, and vice versa.startandendalignment resolve to the wrong edgeThis one is a bug rather than a gap, and it is the visible one.
read_text_align(odf_style.cpp:108) mapsfo:text-alignvaluesstart→TextAlign::leftandend→TextAlign::right, unconditionally. Butstartandendare relative to the writing direction: in a right-to-left paragraphstartis the right edge.translate_paragraph_style(document_style.cpp:237) then writes that out as a literaltext-align:left.startis what writers emit for ordinary body text rather thanleft, so this hits the common case: a right-to-left document whose paragraphs are simply default-aligned renders flush left, and it stays flush left even if the base direction is fixed, because the wrong edge was baked in at parse time.What we could do
TextAlignhas the shape of the problem in it —left/right/center/justify, with nostart/end. Two ways out:start/endall the way to the CSS. Add them toTextAlignand emittext-align:start/text-align:end, which browsers resolve against the element's direction. Then parsestyle:writing-modeinto a direction onParagraphStyle, emit it asdirection:rtl, and the alignment follows for free. The public enum gains two values, so this is a breaking change for the bindings.Either way
<html dir>(or a direction on the document root element) still has to be set from the document's page style, or a document that states nothing per paragraph gets no direction at all.Worth deciding whether the same pass carries
lang, which is in the ODF too and which nothing currently emits — it is what a screen reader and hyphenation need, and it is the same attribute pair.