farsi character in knitr

In R 3.3.1, I'm using knitr 1.13, and I'm having trouble recognizing Farsi characters. I save the following code in a file called test.Rnw and then run knit("test.Rnw"), but the Farsi characters change to ????? in output.

\documentclass{article}
\usepackage{xepersian}%package for type Farsi in LaTex
\begin{document}
این یک متن آزمایشی است
<<model>>=
fit <- lm(dist ~ speed, data = cars)
@
برای آزمایش داریم:
\Sexpr{coef(fit)[2]}
\end{document}

Solution

Ensure that the file is saved in UTF-8 encoding
On most operating systems, that’s enough, since UTF-8 is the system’s default encoding anyway. Especially on Windows, you additionally need to specify the encoding when knitting, e.g.:
```
knit('filename.rnw', encoding = 'UTF-8')
```

Futhermore, you need explicitly handle reading direction of your text: Farsi is right-to-left, but the (Western) source code requires left-to-right text direction. This needs to be set explicitly in the code using e.g. \(un)setLTR:

\documentclass{article}
\usepackage{xepersian}
% Set a font for the text
\settextfont{XB Niloofar}

\begin{document}
این یک متن آزمایشی است
\setLTR
<<model>>=
fit <- lm(dist ~ speed, data = cars)
@
\unsetLTR
برای آزمایش داریم:
\Sexpr{coef(fit)[2]}
\end{document}

You could also use a Knitr hook to set this automatically for each code chunk. You could probably adapt example 74 for this:

<<setup, include=FALSE>>=
library(knitr)
knit_hooks$set(ltr = function(before, options, envir) {
    if (before) '\\setLTR' else '\\unsetLTR'
})
@

Now you can write a code chunk as follows:

<<model, ltr=TRUE>>
fit <- lm(dist ~ speed, data = cars)
@

No need for manual \setLTR…\unsetLTR any more.