The parseLatex package
In parseLatex: Parse 'LaTeX' Code

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Introduction

Parsing LaTeX is tricky, because LaTeX macros (in LaTeX packages, or in user code) can change the parsing rules as they go.

parseLatex is not a LaTeX interpreter (at least mostly it isn't, but see the detailed comparison below), so it can't do that: it uses the same parsing rules for all code that it looks at. If you're using a LaTeX package that uses non-standard rules, you can use those, but they have to apply to the whole section of code passed to parseLatex().

Subject to the limitation that the code only uses one set of rules, parseLatex should be able to parse any LaTeX code. It extends the base tools::parseLatex() function in a few ways:

It classifies every character in the source file according to TeX "catcodes". The base function only handles some of them.
The parseLatex::parseLatex() function marks its output with class "LaTeX2" instead of "LaTeX", and marks each item in the output with class "LaTeX2item". This allows it to print things in a more readable way.
The parseLatex package includes a large selection of functions for extracting and modifying parts of the parsed LaTeX.

More differences are listed below.

Demo

A simple demonstration is in order.

First, we use knitr to create a LaTeX table.

library(knitr)
latex <- kable(mtcars[1:2, 1:2], format = "latex")
cat(latex)

Next, we parse it in parseLatex.

library(parseLatex)
parsed <- parseLatex(latex)

Printing the result would appear to duplicate the input, but in fact it is quite different. parsed is a list of class "LaTeX2". Items in the list are of class "LaTeX2item". In this example, there are only two items: the blank that knitr puts at the beginning of each table, and a second entry which is the whole table environment:

parsed[[1]]
parsed[[2]]

"SPECIAL" and "ENVIRONMENT" label the types of items. The table environment contains the environment name, and a "LaTeX2" list containing all the content.

If we hadn't known where we put it, we could find the table location using find_env():

find_env(parsed, "tabular")

We can extract the table, and use other functions to work with it:

table <- parsed[[find_env(parsed, "tabular")]]
# Get the alignment options from the content
columnOptions(table)
tableCell(table, 2,2) # The title counts as a row
tableCell(table, 1,1) <- "Model"
table

Differences from `tools::parseLatex`

The parser in this package is based on the one used by the base R tools::parseLatex function (which I also wrote, based on other parsers in R). The output format is similar, but not compatible. These are the main differences.

In both this package and tools::parseLatex, the result of calling the parser is a list of items.
The list has class "LaTeX2" in this package, and class "LaTeX" in tools::parseLatex.
All items have an attribute returned by the latexTag() function identifying the type of item. In this package the possible tags are

tags <- data.frame(Tag = c("BLOCK", "COMMENT", "DISPLAYMATH", "ENVIRONMENT", "MACRO", "MATH", "SPECIAL", "TEXT", "VERB", "DEFINITION", "ERROR"),
        Description = c(
          "A block enclosed in curly braces",
          "A LaTeX comment",
          "A display math block",
          "A LaTeX environment",
          "A LaTeX macro",
          "An inline math block",
          "A non-alphabetic character",
          "Text (consisting of letters only)",
          "A verbatim environment",
          "A command or environment definition",
          "A block of items referenced in an error message"),
        Type = c("list", "character", "list", "list", "character",
                 "list", "character", "character", "character", "list", "list"))
knitr::kable(tags, booktabs = TRUE)

The tools::parseLatex() parser does not have the SPECIAL; such characters are included in TEXT. It also doesn't have the DEFINITION or ERROR tags. Definitions are treated as regular macros, which sometimes leads to parsing errors. Errors are always fatal.
This parser stops when it reaches \end{document}, just as LaTeX does. The tools::parseLatex() parser continues parsing beyond that, often leading to parsing errors as it tries to parse things that LaTeX would ignore.
In both implementations, some items (COMMENT, MACRO, SPECIAL, TEXT, and VERB) are stored as length 1 character vectors; the others are stored as lists of items corresponding to their content.
The list storage is different between the two. The tools::parseLatex() function stores some lists in two levels (e.g. the content of an environment named item would be in item[[2]]), while in this package, all lists contain the content directly (e.g. the content of that environment would be in item itself).
This package marks all items with class "LaTeX2item". tools::parseLatex() does not assign a class to items.
This package provides print methods for class "LaTeX2item" so that individual items print nicely.
Parsing errors are reported more informatively by this package.
This parser supports "magic comments". See the next section for details.
This parser is more flexible in handling verb macros like \Sexpr. The tools::parseLatex() parser assumed there would be no braces within the macro (which is the case for legal Sweave() source). This parser assumes any braces within the macro are balanced, e.g. this would be legal:\ \ \Sexpr{1 + {x <- 2; x + 1}}\ \ whereas unbalanced braces would not be.

As mentioned above, parseLatex() does a little bit more than parsing. Both versions recognize LaTeX environments and verbatim code.

The parser in this package also takes special action when it sees the document environment: it stops parsing at \end{document}. (You can use the get_leftovers() function to see what parts of the input were skipped.)

It also changes the rules a bit when it sees macros defining things: \newenvironment, \renewenvironment, \newcommand, \renewcommand and \providecommand. The arguments to these macros are parsed but not interpreted, allowing definitions to parse without triggering a syntax error. For example:

\newenvironment{newenv}{\begin{oldenv}}{\end{oldenv}}

The \begin{oldenv} part of the definition shouldn't be interpreted here as the start of an oldenv environment, because \end{oldenv} isn't in the same {} block.

One plain TeX version of these macros is \def. It is recognized and an attempt is made to handle it, but there's some really arcane syntax possible with \def. If you use that, it probably won't be parsed properly. Stick with simple syntax like

\def\bea{\begin{eqnarray*}}

and you should be okay.

Magic Comments

The parseLatex::parseLatex() parser can parse most LaTeX inputs, but not all. To allow it to be used on files that contain unsupported syntax, it allows "magic comments" to be inserted to control its actions.

Several LaTeX editors support magic comments of the form % !TEX ..., and those were the model for parseLatex magic comment support. There are 4 magic comments supported in this parser:

% !parser off This tells the parser to absorb all following text as part of the comment, so anything that would be classed as a parsing error is never seen.
% !parser on This tells it to resume normal parsing.
% !parser verb [name] This tells the parser to add the name to the list of macros holding verbatim text, i.e. the list given by the verb argument when parseLatex() was called. The name should include the backslash, e.g.\ \ % !parser verb \Sexpr\ \ would add the default verb macro.
% !parser defcmd [name] does the same for commands like \newcommand.
% !parser defenv [name] does it for commands like \newenvironment.
% !parser verbatim [name] This tells the parser to add the name to the list of environments holding verbatim text, i.e. the list given by the verbatim argument. For example\ \ % !parser verbatim Sinput\ \ would add one of the default verbatim environments.

The parser is quite strict about the format of the magic comments. The whitespace between parts of it must be spaces, not tabs, and nothing else can appear in the comment after the magic text other than more spaces.

Work in progress!

This is a work in progress, so if you have a use for something like this and need help, post an "issue" on the Github page: https://github.com/dmurdoch/parseLatex .