At first, I’ve tried to use the English dash and en-dash but result weren’t satisfactory. Both of them were too light and positioned too low to be used in an Hebrew text written in the Culmus fonts (see screenshot).

So finally, I’ve decided to write a simple macro to emulate a Kav-Mafrid for the David font.

\newcommand{\dd}{\lower0.17em\hbox{-\kern-0.13em-}}

The command works by kerning two Hebrew, Maqaf’s, together (with a slight overlap to adjust width), and lowering their position to the middle of the line.

In the following screensot the first line features the Frank Ruehl font and it\’s Maqaf and Kav-Mafrid. The second line featured the david font with a Maqaf and the missing Kav-Mafrid ligature. The third line features the David font with English dash and en-dash. The last line uses David with the regular Maqaf and the \dd{} macro.

Different dash examples

As it can be seen, the \dd{} macro does provide a good alternative for the Kav-Mafrid. However this solution isn’t perfect. One can improve the macro by detecting if the Kav-Mafrid ligature is present in the current font and if so, use it instead of the kerned one. This would enable one to use the macro and safely switch fonts, knowing he will always get the best Kav-Mafrid possible.

The problem is that because TeX doesn’t store information regarding the direction of text inside the dvi, RTL text is actually stored in reverse. When a link is inserted in such text, the command to start the link ends up after the command to end the link, which is problematic. When trying to compile such document with pdfTex it throws errors like:

! pdfTeX error (ext4): pdf_link_stack empty, \pdfendlink used without \pdfstart

On the otherhand XeTeX behaves a little nicer and compiles the document, but the wrong parts of the text are marked as links.

My workaround basically changes the order of the start and end link commands. Thus, in RTL text the commands get reversed and actually end up in the correct order. The workaround doesn’t support links spanning multiple lines and color links (although link border is supported). Further work needs to be done in order to support links in the table of contents. I could get it to work only in LTR mode (which isn’t good). You should insert the following code to your preamble in order to enable the workaround.

\makeatletter
\def\hyper@link#1#2#3{%
  \if@rl
    \setLR
      \begingroup
      \hyper@linkend
        #3
      \hyper@linkstart{#1}{#2}
    \setRL
  \else
    \hyper@linkstart{#1}{#2}
        #3
    \hyper@linkend
  \fi
}
\makeatother

I’ll also note that in order to generate correct pdf bookmarks (not in gibberish), you should set unicode=false in the hyperref‘s options.

You can find a working example bellow (and the corresponding PDF output here).

\documentclass{article}
\usepackage{fontspec}
\usepackage{xunicode}
\usepackage{bidi}
 
\usepackage[unicode=false,
 bookmarks=true,bookmarksnumbered=false,bookmarksopen=false,
 breaklinks=false,pdfborder={0 0 1},backref=false,colorlinks=false]
 {hyperref}
 
\makeatletter
\def\hyper@link#1#2#3{%
  \if@rl
    \setLR
      \begingroup
      \hyper@linkend
        #3
      \hyper@linkstart{#1}{#2}
    \setRL
  \else
    \hyper@linkstart{#1}{#2}
        #3
    \hyper@linkend
  \fi
}
\makeatother
\setromanfont{Times New Roman}
 
\begin{document}
\setRL
\title{בדיקה}
\author{גיא רוטנברג}
\date{}
\maketitle
\pagebreak{}
 
\section{סעיף ראשון}\label{testlabel}
\subsection{תת-סעיף ראשון}\label{testlabel2}
 
לה לה
 
לי לי
 
לו לו
 
\pagebreak{}
\section{סעיף שני}
\subsection{תת-סעיף ראשון}
 
לה לה
 
\subsection{תת-סעיף שני}
 
לי לי
חלק \ref{testlabel}
אבגד
\hyperref[testlabel2]{בדיקה ארוכה}
 
\setLR
This is a test for LTR links. See section \ref{testlabel} and \hyperref[testlabel2]{long label}.
\end{document}

For sometime before letting go of the issue, I’ve tried to correct the direction of the label. It could have been fixed by marking the float as English text in LyX (I use it for editing Hebrew papers, as it makes editing multi-lingual texts easy), effectively adding \selectlanguage{english} before the \begin{algorithm} line. However, it resulted in display mess in LyX and some LaTeX errors when compiling (also it seemed LaTeX could complete the first pass).

Today, I’ve decided to tackle and solve the problem for a new paper I’m writing. After searching a bit for a way to override the Algorithm environment label, I’ve found a general way to override a float label in LaTeX:

\floatname{floatname}{floatlabel}

So for our case it would look like this:

\floatname{algorithm}{אלגוריתם}

You should put it in your LaTeX preamble (Documents->Settings in LyX). This neatly fixes the problem (see screenshot below).

So I wrote a patch for the he.po file that fixes the problem.

Index: he.po
===================================================================
--- he.po	(revision 28095)
+++ he.po	(working copy)
@@ -15050,15 +15050,15 @@
 
 #: src/Buffer.cpp:2902
 msgid "\\arabic{enumi}."
-msgstr "\\arabic{enumi}."
+msgstr ".\\arabic{enumi}"
 
 #: src/Buffer.cpp:2908
 msgid "\\roman{enumiii}."
-msgstr "\\roman{enumiii}."
+msgstr ".\\roman{enumiii}"
 
 #: src/Buffer.cpp:2911
 msgid "\\Alph{enumiv}."
-msgstr "\\Alph{enumiv}."
+msgstr ".\\Alph{enumiv}"
 
 #: src/Buffer.cpp:2928 src/insets/InsetCaption.cpp:293
 msgid "Senseless!!! "

While this patch is against the SVN trunk (LyX 2.0), it could also be applied manually to older LyX versions.
After yo apply the patch you should do the following (I assume that you are in the root of the LyX source tree):

cd po/
make he.gmo
cd ../
make install

Now restart LyX and you will see the numbering appear as it should.

Package amsmath Error: \begin{split} won't work here.

I was using LyX, so I’ve assumed that it would generate the correct syntax for this environment. But the error clearly said that the \begin{split} was used wrong. I’ve went through the user’s guide for the amsmath package and rechecked the syntax. The document has a very handy section that contain common error messages and explanations. I was happy to find out that the error I got was documented there. Unfortunately, the explanation only confirmed that LyX generated the appropriated code for the split environment.

At this point, I’ve decided to export the document to a LaTeX file and manually tweak it till I find a solution. I’ve removed everything from the document apart from some Hebrew text and the split formula. I’ve suspected this error had to do with the Hebrew, as nobody else reported errors using split to the LyX mailing list. I’ve remembered that someone mentioned to me once that LaTex is very sensitive to the order the packages are added. So I’ve started to play with the order and moved the \usepackage statements around. When I’ve put the \usepackage{babel} above the \usepackage{amsmath} the document suddenly compiled fine.

Some further testing showed that in order for the split environment to work properly in an Hebrew document the babel package must be included before amsmath and any other package that depends on it (like mathtools).

Getting things to work back in LyX was a bit more tricky. LyX automatically adds the babel package after the LaTeX preamble, and by default it adds the amsmath when needed, but it does so before the LaTeX preamble. This, as you can see, results in the \usepackage{amsmath} appearing before the \usepackage{babel}. The solution had to override the placements of both statements.

The first step is to go to Document->Settings->Math Options and uncheck both “Use the AMS math package automatically” and “Use the AMS math package”. This won’t prevent you from using any to the amsmath package feature if you include it manually (like we will do). The next step is to go to Documents->Settings->LaTeX Preamble, and insert directly at the top of the preamble the following two lines:

\usepackage{babel}
\usepackage{amsmath}

Now the Hebrew document will compile fine even if you use the amsmath‘s split environment. Note that LyX will add \usepackage statement for for babel after the preamble despite the fact you already have one. But don’t worry, LaTeX will ignore that second statement and no harm will be done.

TEXMF will be the directory of you local TeX tree (usually /usr/share/texmf). Before Begining the installation process make sure you have the Culmus fonts installed. Apparently Culmus is not optional it’s a requirement. I’ll assume that your Culmus fonts are installed in /usr/share/fonts/culmus.

1. Download the ivritex-1.2.1 source-code from here .
2. Extract the archive into a temporary directory.
3. Save the diff file below a file named “Makefile_patch” and save it inside ivritex-1.2.1/fonts/culmus .
4. Apply the patch by going to the ivritex-1.2.1/fonts/culmus directory (under the directory where you extracted the source archive) and executing “patch Makefile_patch. The patch will alter the places where some file will be installed.
5. As root execute “updmap –enable Map culmus.map”.
6. Still as root execute “mktexlsr”.
7. Ivritex 1.2.1 should be installed now.

--- Makefile	2007-02-14 19:59:52.000000000 +0200
+++ Makefilenew	2007-02-16 10:11:07.000000000 +0200
@@ -20,8 +20,8 @@
 vf_target     = $(TEX_ROOT)/fonts/vf/culmus
 # this is where ivritex will eventually be:
 tex_target    = $(TEX_ROOT)/tex/generic/babel
-encode_dir    = $(TEX_ROOT)/dvips/base
-dvips_cfg_dir = $(TEX_ROOT)/dvips/config
+encode_dir    = $(TEX_ROOT)/fonts/enc/dvips/base
+map_dir       = $(TEX_ROOT)/fonts/map/
 sysconf       = $(DESTDIR)/etc
 updmap_dir    = $(sysconf)/texmf/updmap
 #culmus_target = $(PREFIX)/fonts/culmus
@@ -137,11 +137,11 @@
 	mkdir -p $(sysconf)/texmf/updmap.d
 	echo "Map culmus.map" >$(sysconf)/texmf/updmap.d/10culmus.cfg
 else
-	mkdir -p $(dvips_cfg_dir)
-	cp culmus.map $(dvips_cfg_dir)/
+	mkdir -p $(map_dir)
+	cp culmus.map $(map_dir)/
   ifeq ($(tetex_ver),2)
 	# this should run mktexlsr as well
-	$(updmap) --enable Map $(dvips_cfg_dir)/culmus.map
+	$(updmap) --enable Map $(map_dir)/culmus.map
   else # for tetex-1
     ifeq ($(tetex_ver),1)
 	# TODO: fill in sed line here

Hebrew TrueType fonts can be downloaded from here.

In the following, TEXMF is the place of the (local) TeX tree (usually /usr/share/texmf), and TMP is some temporary directory you need to create. Steps 1-3 can be done as a regular user, while the other steps need to be done as root.

1. Create the directory TMP, and chdir into it. Copy (or link) your TrueType fonts to TMP. Download the tar.gz file ttf.tar.gz and extract it into TMP.
2. Run the python script `ttf.py’:

   python ttf.py *.ttf

   If you are unable to run this script, you can manually perform the following steps for each font (you also need to create an empty file called `hebrew.map’). Steps (d) and (e) are optional. These steps are needed for nikud. Step (e) is also needed for ligatures. The instruction are written for the times.ttf font (for other font, you might need to replace winheb2.enc by another encoding file).

   1. Create a tfm file by

      ttf2tfm times.ttf -N -T winheb2.enc -v tmp.vpl htimes.tfm

   2. Get the font name by running

      ttf2afm times.ttf | grep FontName

      For the times font, the result is `TimesNewRomanPSMT’

   3. Add the following line to the file `hebrew.map’:

      htimes TimesNewRomanPSMT <winheb2.enc <times.ttf

   4. Edit the file `tmp.vpl’. Change the width of each nikud character to 0. For example, change

      (CHARACTER O 300(comment sheva)
      (CHARWD R 333)
      (CHARDP R 215) )

      to

      (CHARACTER O 300(comment sheva)
      (CHARWD R 0)
      (CHARDP R 215) )

   5. Run

   vptovf tmp.vpl tmp.vfm htimes.tfm

3. If you don’t have the times/arial/courier fonts (or you don’t want to use them as default fonts), you need to edit the file `ttfheb.sty’
4. Run `su’ to get root access.
5. Move the file ttfheb.sty to the directory of the Hebrew latex files (it should be `TEXMF/tex/generic/babel/hebrew/’. If it is not, you can find this directory with `locate rlbabel.def‘.)
6. Run

   sh install TEXMF

   If you are unable to use this script, perform the following steps manually:

   1. Create a directory called `TEXMF/fonts/truetype/hebrew/‘,
      and move the .ttf files to this directory.
   2. Create a directory called `TEXMF/fonts/tfm/truetype/hebrew/‘,
      and move the .tfm files into this directory.
   3. Move the .enc files and `hebrew.map’ to `TEXMF/fonts/map/pdftex/‘
   4. while in the TEXMF/fonts/map/pdftex/ directory run:

      updmap-sys –enable Map hebrew.map
      

      
      

   5. Run

      mktexlsr TEXMF

7. The final test: as a regular user run

   pdftex test.tex
   pdfelatex test2.tex

To make a LaTeX document use the TrueType fonts, add the line `\usepackage{ttfheb}’ to the preamble (see `test2.tex’).

UPDATES:

18/03/2007: added the new ttfheb.sty file by Artyom Tonkikh and fixed some typos.