Arabic letters support.

[boulabiar]

Thanks for this impressive work !
I'm a bit ashamed to write the first issue, and tried to contact you by other means then got back here.

When opening a PDF or SVG with arabic letters, nothing is shown.
Do you have insights on what may be the cause?
You can try with this PDF
https://tug.ctan.org/macros/latex/exptl/mem/arabic.pdf
Or any SVG having arabic letters.

[mindbrix]

Thanks for raising the issue. SVG decoding has no text support, as nanosvg is quite basic. PDF text decoding is incomplete also. You will notice that no text is displayed at all from arabic.pdf.

RasterizerFont can load Arabic fonts, and layoutGlyphs() has basic right-to-left support.

[mindbrix]

And thanks for the tip. Hugely appreciated.

[mindbrix]

Having investigated this more deeply, the core issue is that the PDFium API FPDFFont_GetGlyphPath() is unreliable.

The current WIP version of Rasterizer has improved PDF text positioning, and shows missing text paths as red boxes. As you can see below, FPDFFont_GetGlyphPath() fails for every Arabic character.

I tried updating PDFium to the latest version, but it is no better.

[boulabiar]

I've seen on the internet that FPDFFont_GetGlyphPath() can't be used and that FPDFFont_GetFontData() should be used instead.
Then create an FT_Face from those bytes (FreeType), and an hb_font_t for HarfBuzz (for correct ligature etc.).

I will try to do some experiments whenever I have some time.
(do you have btw your last code of the last message?)

[mindbrix]

Many thanks for your help. FPDFFont_GetFontData() is an API that is not in the version of pdfium Rasterizer uses. The newer versions from https://github.com/bblanchon/pdfium-binaries trigger macOS security warnings, which are unacceptable for distribution.

The new code is in the ObjC branch where I am adding an Objective-C & Swift API and adding Swift Package Manager support, which is almost complete.

[mindbrix]

The security issue may be fixable: bblanchon/pdfium-binaries#183

[mindbrix]

The ObjC branch now uses FreeType (via FPDFFont_GetFontData()) to get font glyph outlines. If that fails it falls back to FPDFFont_GetGlyphPath(). Ligatures are still not working, and I'm currently still researching this.

[mindbrix]

The pdfium text APIs don't provide enough support for easy handling of complex typography. The best work around would probably to render complex type text object as images, which are now supported by Rasterizer.

[boulabiar]

I tried to use hyaro pdf parser to get the glyphs but haven't enough time to work on it more.
https://github.com/alaincontrol/hayro

Will do some tests myself whenever I get some time.

[mindbrix]

Good luck! PDF parsing is a huge rabbit due to the size of the specification and the age of the format.