latex
A LaTeX ⇄ richdoc converter, written
in pure Go (CGO-free, including GOOS=js).
latex parses a practical subset of LaTeX into the neutral richdoc document
model, and emits a minimal, compilable LaTeX article from a richdoc.Document.
The two directions are designed as a faithful round-trip.
d, err := latex.Parse(src) // LaTeX subset -> *richdoc.Document
out, err := latex.Write(d) // *richdoc.Document -> compilable LaTeX article
API
func Parse(src []byte) (*richdoc.Document, error)
func Write(d *richdoc.Document) ([]byte, error)
Parse reads only the body of the document environment when one is present
(the preamble is mined for metadata); a bare fragment with no document
environment is parsed whole. Anything the model has no node for is preserved
verbatim through RawInline/RawBlock with Format: "latex", so nothing is
lost. Write loads only the packages the document needs and escapes LaTeX
specials in text.
Construct mapping
The supported subset maps to richdoc as follows (both directions):
| LaTeX |
richdoc |
\section / \subsection / \subsubsection / \paragraph / \subparagraph |
Heading (level 1–5) |
| blank-line-separated text |
Paragraph |
\textbf{} |
Strong |
\textit{}, \emph{} |
Emph |
\texttt{} |
Code (inline) |
\sout{} (ulem) |
Strikethrough |
\\, \newline |
LineBreak |
itemize / enumerate (\item) |
List (ordered for enumerate) |
verbatim / lstlisting |
CodeBlock (lstlisting[language=…] sets the language) |
quote / quotation |
BlockQuote |
tabular |
Table (& cells, \\ rows, l|c|r spec → alignment, \hline dropped) |
\href{url}{text}, \url{} |
Link |
\includegraphics[…]{path} |
Image |
\footnote{…} |
Footnote (inline arg wrapped in one Paragraph) |
\label{id} |
Anchor (point target); hoisted onto Heading.ID right after a \section… |
\ref{id}, \eqref{id} |
CrossRef (RefLabel) |
\cite{key} |
CrossRef (RefCite) |
$…$, \(…\) |
Math (inline) |
\[…\], equation, align, … |
MathBlock |
\hrulefill, \rule… |
ThematicBreak |
\documentclass, \title, \author, \date |
Document.Meta |
\maketitle |
dropped (title comes from Meta) |
| unknown command / environment |
RawInline / RawBlock (Format: "latex") |
Write emits \documentclass{article} plus only the packages in use
(fontenc, graphicx, hyperref, amsmath, ulem, listings), maps
Meta title/author/date onto \title/\author/\maketitle, and renders each
block and inline node back to LaTeX.
Two mappings normalise on the way back to LaTeX. A \section… (any level)
immediately followed by \label{id} (only whitespace or comments between) is
folded into a single Heading with that ID, and Write re-emits the
\label right after the section command — so the pair round-trips as one
Heading. Both \ref and \eqref parse to a CrossRef of kind RefLabel,
and Write always emits \ref; an \eqref source therefore round-trips as
\ref (a benign normalisation, since both denote the same labelled reference).
\footnote, \label, \ref and \cite are all core LaTeX, so the emitted
output typesets with no extra package. A \label that does not immediately
follow a heading stays a point Anchor in the inline stream. Only genuinely
unrecognised commands and environments still fall back to RawInline /
RawBlock.
Parsing
richdoc is a document model with no parser, and
go-tex/engine is a typesetting engine
whose tokenizer and macro machinery are internal — it exposes a compile API, not
a reusable structured parse tree. So latex ships its own focused,
well-tested LaTeX-subset parser (no full TeX macro expander, no non-Go
dependency). The go-tex engine is used to prove the round-trip: a test
compiles Write's output with the engine and asserts it typesets.
License
BSD-3-Clause. Copyright (c) the go-richdoc authors.
To a PDF
latex/pdf typesets a document rather than writing source for one:
data, err := pdf.Write(doc, pdf.Options{}) // *richdoc.Document -> PDF bytes
It typesets nothing itself. The document goes out as LaTeX through Write above
and is compiled by go-tex/engine, a pure-Go
TeX engine that runs the genuine LaTeX classes — so what comes back is a page
TeX laid out, with TeX's line breaking, rather than an approximation of one.
That composition worked before the package existed; what was missing was
somewhere to put it. Every converter in this organisation reaches richdoc, so
every one of them now reaches a PDF.
It is a package rather than a module, and not part of this one. The engine
is a six-megabyte TeX implementation. This module already names it, but only
from a test — checking that the LaTeX it emits actually typesets — so importing
latex does not link it. Putting Write beside the LaTeX writer would link it
into every consumer that only wanted LaTeX text. A separate module would avoid
that and bring another repository, another set of shared defaults and another
release to keep in step, for thirty-five lines composing two libraries. Go links
by package, so a package here costs neither.
What survives the whole way, read back out of the finished PDF with pdftotext
rather than trusted: headings, emphasis, bold, bulleted and numbered lists,
inline and block code, a quotation, a table, a link, a rule and accented text.