[XeTeX] weird behaviour with LetterSpace

[Jonathan Kew]

On 31 May 2010, at 22:13, Pablo Rodríguez wrote:

> Hi there,
> 
> I have just accidentally discovered that LetterSpace behaves differently if the whole paragraph is set with this feature or not.
> 
> The minimal example:
> 
> \documentclass[12pt]{article}
> \usepackage{fontspec}
> \setmainfont{Theano Didot}
> \begin{document}
> χαλεπὰ \addfontfeature{LetterSpace=12}τὰ καλά
> 
> χαλεπὰ τὰ καλά
> 
> χαλεπὰ \addfontfeature{LetterSpace=0}τὰ καλά
> 
> Beauty \addfontfeature{LetterSpace=12}is difficult
> 
> Beauty is difficult
> 
> Beauty \addfontfeature{LetterSpace=0}is difficult
> \end{document}
> 
> If you copy the resulting text (from http://www.ousia.tk/wrong-letterspace.pdf), you will see that only the second line is properly typeset, or at least, there are no blank spaces between letters.
> 
> I guess this might be a probable cause for wrong hyphenation when using LetterSpace. (BTW, loading polyglossia makes no difference.)
> 
> Have I hit a bug in LetterSpace? Do you know any way to avoid this?

The PDF looks correct to me; where LetterSpace=12 is in effect, the letters are more widely spaced, and where LetterSpace=0, they're not. I don't see a bug here. Or am I missing something?

If you're specifically concerned about what happens when you use a viewer to select and copy the text from this PDF into an editor... well... that's a chancy operation. It worked fine for me with Acrobat (no extra spaces), but other viewers may give different results. Basically, this is a poorly-defined operation. As TeX does not use "space characters" between words, there is no clear indication in the PDF data of where the word boundaries should be, and so the viewer has to guess based on the glyph positions. That works most of the time for simple running text, but modifying the letter spacing carries a pretty high risk of confusing it.

As I see it, PDF was not really designed to be an interchange medium for text; it's designed to convey the graphical appearance of the page. "Extracting" the underlying text from the glyphs on the page is an afterthought that has never been 100% reliable. Added features such as /ActualText can help, but xetex does not currently support the automatic generation of /ActualText in the PDF output -- and I'd be reluctant to add it, considering how much it would bloat the output.

Basically, if you want to get the text reliably, you shouldn't be starting from the PDF! :)

JK