Bug Tracker

I see. Would putting indexing afterwards for PARTICULAR commands (e.g., the set of Booleans 𝔹 or the power set 𝒫) in NEW documents have an adverse effect? I might imagine suboptimal kerning if punctuation follows, though I was unable to construct an example. In any case, we might wish to continue to have a choice of where to put the indexing command: to the front or to the back for other reasons (say, hyphenation points or calling from a macro).

As for the issue with \index, I now found some other workarounds (thanks to David, cf. [TeX.SX Link]):

\documentclass{article}
\pagestyle{empty}
\usepackage{amssymb}
\usepackage{makeidx}
\makeindex
\begin{document}%
\newlength{\myLen}\setlength{\myLen}{33em}%
\noindent 1st line. Plain \(B\), no index:\\
\hspace*{\myLen}\(\Sigma\to B\)\\
\noindent 4th line. Index before plain B:\\
\hspace*{\myLen}\(\Sigma\to\index{B}B\)\\
\noindent 7th line. Index after plain B:\\
\hspace*{\myLen}\(\Sigma\to B\index{B}\)\\
\noindent 10th line. Double-struck \(\mathbb{B}\), no index:\\
\hspace*{\myLen}\(\Sigma\to \mathbb{B}\)\\
\noindent 13th line. Double-struck \(\mathbb{B}\), index before \texttt{\textbackslash mathbb}:\\
\hspace*{\myLen}\(\Sigma\to \index{B}\mathbb{B}\)\\
\noindent 16th line. Double-struck \(\mathbb{B}\), index after \texttt{\textbackslash mathbb}:\\
\hspace*{\myLen}\(\Sigma\to \mathbb{B}\index{B}\)\\
\noindent 19th line. Double-struck \(\mathbb{B}\), index inside \texttt{\textbackslash mathbb} before B:\\
\hspace*{\myLen}\(\Sigma\to \mathbb{\index{B}B}\)\\
\noindent 22nd line. Double-struck \(\mathbb{B}\), index inside \texttt{\textbackslash mathbb} after B:\\
\hspace*{\myLen}\(\Sigma\to \mathbb{B\index{B}}\)\\
\noindent 25th line. Index before plain B, all braced:\\
\hspace*{\myLen}\(\Sigma\to {\index{B}B}\)\\
\noindent 28th line. Index before plain B, index braced:\\
\hspace*{\myLen}\(\Sigma\to {\index{B}}B\)\\
\noindent 31st line.
\printindex
\end{document}

I have a similar situation in my slightly less minimal example. Namely, I have enclosing \mathbb (for pdflatex), \mathfrak (for pdflatex), and even hyperref links (for both pdflatex and lualatex). Feeding

\documentclass{article}
\pagestyle{empty}
\usepackage{amssymb}
\usepackage[unicode,hidelinks]{hyperref}
\usepackage[makeindex,nomain]{glossaries-extra}
\newglossary[nlg]{notation}{not}{ntn}{Symbols}
\makeglossaries
\newcommand{\booleanSetSymbol}{\mathbb{B}}%%% for pdflatex
%\newcommand{\booleanSetSymbol}{𝔹}%%% for lualatex
\newcommand{\powerSetSymbol}{\mathfrak{P}}%%% for pdflatex
%\newcommand{\powerSetSymbol}{𝒫}%%% for lualatex
%\usepackage{unicode-math}%%% for lualatex
\newglossaryentry{not:booleanValues}{type=notation, name=\(\booleanSetSymbol\), text=\booleanSetSymbol, description={The set of Boolean values.}}
\newglossaryentry{not:powerset}{type=notation, name=\(\powerSetSymbol\), text=\powerSetSymbol, description={The power set of the set \(Z\) is denoted by \(\powerSetSymbol(Z)\).}}
\begin{document}%
\newlength{\myLen}\setlength{\myLen}{33em}%
\noindent Line 1. Good:\\
\hspace*{\myLen}\(\Sigma\to\booleanSetSymbol\)\\
\noindent Line 4. Good:\\
\hspace*{\myLen}\(\Sigma\to{\gls{not:booleanValues}}\)\\
\noindent Line 7. Bad:\\
\hspace*{\myLen}\(\Sigma\to\gls{not:booleanValues}\)\\
\noindent Line 10. Good:\\
\hspace*{\myLen}\(\Sigma\to\powerSetSymbol(\Sigma)\)\\
\noindent Line 13. Good:\\
\hspace*{\myLen}\(\Sigma\to{\gls{not:powerset}}(\Sigma)\)\\
\noindent Line 16. Bad:\\
\hspace*{\myLen}\(\Sigma\to\gls{not:powerset}(\Sigma)\)\\
\noindent Line 19.
\printglossaries
\end{document}

Having said that, I wonder whether we might invoke the internal indexing commands from some internal set of braces (ideally, coming from hyperref, and if not, from \mathbb or \mathfrak)? I feel that it's not the user's main text that has to deal with hyphenation issues, but glossaries, or glossaries-extra, or the preamble of the user's LaTeX document at worst.

Setting wrgloss=after doesn't work locally: feeding

\documentclass{article}
\pagestyle{empty}
\usepackage{amssymb}
\usepackage[unicode,hidelinks]{hyperref}
\usepackage[makeindex,nomain]{glossaries-extra}
\newglossary[nlg]{notation}{not}{ntn}{Symbols}
\makeglossaries
\newcommand{\booleanSetSymbol}{\mathbb{B}}%%% for pdflatex
%\newcommand{\booleanSetSymbol}{𝔹}%%% for lualatex
\newcommand{\powerSetSymbol}{\mathfrak{P}}%%% for pdflatex
%\newcommand{\powerSetSymbol}{𝒫}%%% for lualatex
%\usepackage{unicode-math}%%% for lualatex
\newglossaryentry{not:booleanValues}{type=notation, name=\(\booleanSetSymbol\), text=\booleanSetSymbol, description={The set of Boolean values.}}
\newglossaryentry{not:powerset}{type=notation, name=\(\powerSetSymbol\), text=\powerSetSymbol, description={The power set of the set \(Z\) is denoted by \(\powerSetSymbol(Z)\).}}
\newglossaryentry{not:function}{type=notation, name={\(\_\mathord{\to}\_\)}, text={\_\to\_}, sort={function}, description={The notation \(X{\to}Y\) means the set of all functions with domain \(X\) and codomain \(Y\). Following Bourbaki, we see a function as a triple (function graph, domain, codomain); a function \(f\allowbreak\in\allowbreak X{\to}Y\), traditionally written as \(f\colon\allowbreak X\penalty1\to\penalty1 Y\), is thus \(\bigl(\mskip-1mu plus.5mu minus.5mu\{(x,f\mskip-.4mu plus.2mu minus.2mu (x\mskip-.3mu plus.15mu minus.15mu)\mskip-.5mu plus.25mu minus.25mu)\penalty2\mid\penalty1 x\mskip-3mu plus1mu minus1mu\in\mskip-3mu plus1mu minus1mu X \penalty3\land\penalty3 y\mskip-3mu plus1mu minus1mu\in\mskip-3mu plus1mu minus1mu Y\},\allowbreak X,\allowbreak Y\mskip-.3mu plus.15mu minus.15mu\bigr)\).}, user1={function}, user2={functions}, user3={map}, user4={maps}, user5={\colon}, user6={\to}}
\begin{document}%
\newlength{\myLen}\setlength{\myLen}{33em}%
\noindent Line 1. Without gls:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\booleanSetSymbol\)\\
\noindent Line 4. With gls and wrgloss=after:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\gls[wrgloss=after]{not:booleanValues}\)\\
\noindent Line 7. With gls:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\gls{not:booleanValues}\)\\
\noindent Line 10. Without gls:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\powerSetSymbol(\Sigma)\)\\
\noindent Line 13. With gls and wrgloss=after:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\gls[wrgloss=after]{not:powerset}(\Sigma)\)\\
\noindent Line 16. With gls:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\gls{not:powerset}(\Sigma)\)\\
\noindent Line 19.
\printglossaries
\end{document}

I'll look into it further. If the page numbers aren't important (for example, if you are only indexing to ensure the symbols appear in the list but are using nonumberlist to suppress the number list) then the savewrites package option might solve the problem. This defers the file writes to the end of the document. The main drawback with this option is that the page numbers can be off by one due to TeX's asynchronous output routine.

Thank you in advance!

Page numbers are important to me: it's a long book.

In the preamble, I have

\PassOptionsToPackage{mainaux,starttoc,index,glossary}{rerunfilecheck}
\ifxetexorluatex
  \usepackage[unicode,pdftitle={\bookTitleInOneLine},hidelinks,pdfauthor={\authorList}]{hyperref}
\else
  \ifpdf
    \usepackage[unicode,pdftitle={\bookTitleInOneLine},hidelinks,pdfauthor={\authorList}]{hyperref}
  \else
    \usepackage[unicode,pdftitle={\bookTitleInOneLine},hidelinks,breaklinks=true,pdfauthor={\authorList}]{hyperref}%%& In DVI, we want to wrap links properly.
    \usepackage[hyphenbreaks]{breakurl}
  \fi
\fi

\usepackage[makeindex,toc,nogroupskip,nomain]{glossaries-extra}
\setglossarystyle{long3col}
\renewcommand{\glspagelistwidth}{7.35em}%%% Adjust this value manually such that it fits.
\GlsSetQuote{+} %%% Hack to work with "makeindex -g", cf. http://tug.ctan.org/tex-archive/macros/latex/contrib/glossaries/glossaries-user.html#sec:languages

{\makeatletter\setlength\tabcolsep{5\p@}\makeatother%
\let\oldarraystretch\arraystretch%
\renewcommand{\arraystretch}{2}%
\renewenvironment{theglossary}%
{\begin{longtable}%
    {@{}lp{\glsdescwidth}>{\smaller}p{\glspagelistwidth}@{}}%
}{%
\end{longtable}%
}%
\LTpre=0.0pt%%%% Cf. http://tex.stackexchange.com/a/482259. Original \LTpre value is \bigskipamount .
\printglossaries}

Typo.

I feel that it's not the user's main text that has to deal with hyphenation issues

I feel that it's not the user's main text that has to deal with line-breaking issues

Following on from David's suggestion on TeX.SX to group \index you could try the following (untested):

\renewcommand*{\glswriteentry}[2]{%
  \ifglsindexonlyfirst
    \ifglsused{#1}{}{{#2}}%
  \else
    {#2}%
  \fi
}

The option you've suggested cures the standard setting at least in the original minimal example. The setting wrgloss=after is as ineffective (or as broken) as it was before. Feeding

\documentclass{article}
\pagestyle{empty}
\usepackage{amssymb}
\usepackage[unicode,hidelinks]{hyperref}
\usepackage[makeindex,nomain]{glossaries-extra}
\renewcommand*{\glswriteentry}[2]{%
  \ifglsindexonlyfirst
    \ifglsused{#1}{}{{#2}}%
  \else
    {#2}%
  \fi
}
\newglossary[nlg]{notation}{not}{ntn}{Symbols}
\makeglossaries
\newcommand{\booleanSetSymbol}{\mathbb{B}}%%% for pdflatex
%\newcommand{\booleanSetSymbol}{𝔹}%%% for lualatex
\newcommand{\powerSetSymbol}{\mathfrak{P}}%%% for pdflatex
%\newcommand{\powerSetSymbol}{𝒫}%%% for lualatex
%\usepackage{unicode-math}%%% for lualatex
\newglossaryentry{not:booleanValues}{type=notation, name=\(\booleanSetSymbol\), text=\booleanSetSymbol, description={The set of Boolean values.}}
\newglossaryentry{not:powerset}{type=notation, name=\(\powerSetSymbol\), text=\powerSetSymbol, description={The power set of the set \(Z\) is denoted by \(\powerSetSymbol(Z)\).}}
\newglossaryentry{not:function}{type=notation, name={\(\_\mathord{\to}\_\)}, text={\_\to\_}, sort={function}, description={The notation \(X{\to}Y\) means the set of all functions with domain \(X\) and codomain \(Y\). Following Bourbaki, we see a function as a triple (function graph, domain, codomain); a function \(f\allowbreak\in\allowbreak X{\to}Y\), traditionally written as \(f\colon\allowbreak X\penalty1\to\penalty1 Y\), is thus \(\bigl(\mskip-1mu plus.5mu minus.5mu\{(x,f\mskip-.4mu plus.2mu minus.2mu (x\mskip-.3mu plus.15mu minus.15mu)\mskip-.5mu plus.25mu minus.25mu)\penalty2\mid\penalty1 x\mskip-3mu plus1mu minus1mu\in\mskip-3mu plus1mu minus1mu X \penalty3\land\penalty3 y\mskip-3mu plus1mu minus1mu\in\mskip-3mu plus1mu minus1mu Y\},\allowbreak X,\allowbreak Y\mskip-.3mu plus.15mu minus.15mu\bigr)\).}, user1={function}, user2={functions}, user3={map}, user4={maps}, user5={\colon}, user6={\to}}
\begin{document}%
\newlength{\myLen}\setlength{\myLen}{33em}%
\noindent Line 1. Without gls:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\booleanSetSymbol\)\\
\noindent Line 4. With gls and wrgloss=after:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\gls[wrgloss=after]{not:booleanValues}\)\\
\noindent Line 7. With gls:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\gls{not:booleanValues}\)\\
\noindent Line 10. Without gls:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\powerSetSymbol(\Sigma)\)\\
\noindent Line 13. With gls and wrgloss=after:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\gls[wrgloss=after]{not:powerset}(\Sigma)\)\\
\noindent Line 16. With gls:\\
\hspace*{\myLen}\(\Sigma\mathrel{\glsuservi{not:function}}\gls{not:powerset}(\Sigma)\)\\
\noindent Line 19.
\printglossaries
\end{document}

I will add a patch to glossaries-extra first. If that doesn’t have any unexpected side effects reported I’ll also add it to the base package.

Thanks in advance!

In fact,

The option you've suggested cures the standard setting at least in the original minimal example.

The solution (or the hack, depending on the viewpoint) you've suggested cures the problem in the original minimal example.

Cheers,
Peter

… or, perhaps, there might be a way to somehow sidestep the whatsits in math mode completely favour of some other feature of (Lua)(La)TeX. When I think about it, I use pdflatex for a draft version and lualatex for a clean version of my book (and xelatex is dead), so, ultimately, I care more about lualatex now.

My suggestion was just a stopgap measure until the issue can be properly fixed. While wrgloss=after may address the problem for documents such as yours where \gls only produces short content (such as a mathematical symbol), longer content (such as phrases) may need to have the indexing performed at the start in the event that the content crosses a page-break. So the issue needs fixing for both cases.

The modification that I've applied (which will be in the next release of glossaries-extra) fixes the problem in your original MWE. This follows the suggestions on TeX.SX to group both the indexing and the text (regardless of which side of the text the indexing occurs).

There are some other unrelated changes being made to glossaries-extra. Once they are finished I'll upload the new version.