% Copyright 2011 Barbara Beeton <tex-live@tug.org>
% You may freely use, modify and/or distribute this file.
% This is the original annotated test report, with various extra info.
% See below \end{document}.
\documentclass{amsbook}
\usepackage{verbatim}

\makeindex

\begin{document}

\chapter{Demonstration of a Makeindex bug}

\index{random text|(}%
Here we open an index span for ``random text''.

\index{periodic Hilbert transform|(}%
How could we define the \emph{periodic Hilbert transform}?
Here is an open index span for ``periodic Hilbert transform''.

(lots more text omitted)
\newpage

We defined in Section~3 the
\index{periodic Hilbert transform|(}%
\emph{periodic Hilbert transform} or \emph{conjugate function}.
A second open index span for ``periodic Hilbert transform''.

\index{random text|(}%
(more omitted text and a second open index span for ``random text'')

Closing index spans for the two opened entries.
\index{periodic Hilbert transform|)}%
\index{random text|)}%

\vspace{1\baselineskip}
\verbatiminput{\jobname.tex}
\newpage
\noindent This is the .idx file from the previous run:
\verbatiminput{\jobname.idx-prev}
\vspace{2\baselineskip}
This is the resulting .ilg file:
\verbatiminput{\jobname.ilg}
\vspace{2\baselineskip}
And the resulting .ind file; note spurious \verb+\(+ inclusions.
\verbatiminput{\jobname.ind}
\end{document}


Date: Thu, 22 Sep 2011 12:23:48 -0400
From: Barbara Beeton
Subject: a makeindex bug

[...]

to get the desired output, run latex, then makeindex.
copy or rename <jobname>.idx to <jobname>.idx-prev,
rerun latex, and take a look at the output.  (the
reason for the rename is that i've included in the
output listings of the .idx(-prev), .ilg, and .ind
files, all small, but very interesting.)

the problem is that if a span is begun for a string
and then a second span is begun for the same string
without the first one being closed, the |( in the
input is put into the output as \( -- with the
result that an attempt to print the index will get
an unmatched $ error.  the unmatched span is noted
in the .ilg file with a warning, but since most
people ignore warnings the first time around (i
certainly do), the unrecoverable (la)tex error in
the index can really throw them into a tizzy.