gdoc2latex

It’s time to move beyond writing Latex for most papers. Google Docs provides a much better editing experience than Overleaf or local latex editors for many projects: A good spell and grammar checker, a great collaborative editor with decent commenting features, citations with Paperpile, and a visually much less cluttered interface. Yet, Latex is a good backend for producing high-quality PDFs in a final layout.

This project converts Google Doc documents to Latex and onward to PDFs. The document is edited in Google Docs and a PDF version can be viewed in parallel in the required paper template to judge layout and length.

Features (Latex)

This project reads the formatting from Google Docs and converts it to Latex code. It supports the following functionality:

Bold and italics formatting is converted to \textbf and \emph
Links are converted to \url and \href
Itemization is supported and rendered as \begin{compactitem}
Citations are supported through Paperpile. See details below.
Strikethrough formatting is considered as comments and omitted in Latex
Headings in GDoc are converted to \section, \subsection, and \subsubsection in Latex (up to 6 levels deep) -- to use \subsubsubsubsection etc they need to be defined in most latex templates. To use chapters in books, redefine macros (eg. \section to \chapter, \subsection to \section etc.)
Paragraphs starting with bold text are converted to \paragraph parameters (to avoid that, just redefine \paragraph to \textbf)
Footnotes should work as expected
Subfix and superfix formatting renders as $_\text{...} and $^\text{...}
Images embedded in the document as the only element of a paragraph are downloaded as .png file and embedded in the Latex output with \begin{figure}[h!tp]\centering\includegraphics[width=...]{...png}\caption{...}\alt{...}\end{figure}. If the paragraph after the image is formatted in italics it is interpreted as the image caption. Alt text of the image in the Google doc is converted to \alt{...} if available. Images are scaled down relatively if they are scaled in the Google doc.
Text with grey background color is interpreted as an index term and produces an \index entry. Text formatted grey and strikethrough will produce an index entry without having the text in the document too.
Internal links to section headers are converted to \ref references (the link text is discarded).
A heading formatted as Title replaces the \TITLE command in the Latex template
A paragraph starting with “Abstract:” is used as the abstract in the Latex template
Text is otherwise not transformed or escaped, so Latex commands for math or images can be inserted directly in the Google Doc document.
The main template and images and other files to be included can be configured through a Google Drive folder.

Additional formatting conversion could be added as needed.

Paperpile citations/references

Paperpile can be used for citations. Bibtex is not used, references and formatting entirely relies on Paperpile.

Citations generate \gencite{keys}{text} where keys are paperpile-internal IDs and text is the original text in the document, e.g., "[1, 3]" or "(Foo et al., 2004)" depending on formatting used.

Bibliography entries are rendered as \genbibitem{key}{prefix}{text} in a thebibliography environment where key is the Paperpile-internal key, prefix is an optional element of a list such as "[3]", and text is the bibliography entry generated by Paperpile.

To use normal numeric citations in Latex use the following two definition for the generated macros, which generate \bibitem and \cite macros that Latex can then resolve to numbers:

\newcommand\gencite[2]{\cite{#1}}
\newcommand\genbibitem[3]{\bibitem{#1} #3}

For any other format like natbib or alpha let Paperpile do all the formatting and ignore the internal Ids, for example with the following macro definitions (the square-braket parameter for bibitem is not used but needed so that natbib does not complain):

\newcommand\gencite[2]{#2}
\newcommand\genbibitem[3]{\bibitem #3}

If for some reason the *natbib library needs to be included in the document, generate the bibitem with fake (not used) author and year strings:

\usepackage{natbib}
\newcommand\gencite[2]{#2}
\newcommand\genbibitem[3]{\bibitem[A(2020)] #3}

Only the first solution will create links within the document (if enabled). Creating links for other citation styles is not easily possible if multiple references may be used in the same citation, since Paperpile does not map references to individual text fragments of the citation and the text separators (brackets, commas or semicolons, etc) differ between citation styles. Some postprocessing of the keys and text in \gencite may be possible to pair them up for a given citation style.

Markdown

Conversion to markdown are also supported with mostly the same features as for Latex, but without section references, citations, images, and footnotes.

Try it

Until I exceed the free limits of fly.io, here is a version deployed online to try: https://gdoc2latex.fly.dev/update/1yUWsgyIDd7_C7s2SghSAaaPxmKNE0_hlcqfgbUjVci8

This link uses this document, but simply replace the document ID with a different one to try other documents.

Here is more complete example from a recent paper draft (document and template folder): https://gdoc2latex.fly.dev/update/1yZZqEWgR7C7DcrkJZeuSpXRDqCyndUP4o8pvXF-1jvc/1qY_DmTZhPDb0SnaL5S3dH-hvSKk8VV5A

Command-Line Use

If using the command-line interface rather than the more convenient server setup described below, follow these instructions.

Installation: This project is implemented in Scala. It can be compiled with sbt. After installing sbt, simply callsbt stage to download dependencies and build the project. This will create an executable CLI tool in /target/universal/stage/bin.

Google API permissions: Unfortunately, accessing Google’s APIs programmatically requires some setup. Go to the Google Developer Console to create a project, enable the Google Docs and Google Drive APIs, and create a service account under “Credentials” (service accounts seem easier than OAuth2, but OAuth2 is possible with a few lines of code changes in GDocConnection too). The basic “Viewer Role” is sufficient for the service account. The console will provide a json file with credentials for downloads that should be placed in credentials/api.json.

Use: Simply execute the project with gdoc2latex <documentid>, where the document id is the long character sequence in the URL of a Google Doc document. The project will access the document and print the converted Latex code to standard out.

Using parameter -o <file> the output latex file can be specified. Using parameter -t <file> a Latex template can be provided in which \TITLE, \ABSTRACT and \CONTENT will be replaced with the converted text.

Note that the program will need permission to access the document. To this end, the document needs to be public (“share with everybody with this link” as “commenter”) or shared with the email of the service account.

Typically one would automate all steps with something like

gdoc2latex <documentid> -o paper/main.tex -t paper/template.tex
cd paper
latexmk --pdf --interaction=nonstopmode main.tex
cd ..

Server

The more convenient way of using this project is hosting it as a service. This way, it is easy to open the Google Doc in one browser window and refresh the PDF in a different one by refreshing the page, possibly side by side. When using Firefox rather than Chrome the PDF reader even remembers the position within the PDF.

The project contains a Server class for this purpose that responds to GET requests as follows:

http://<host:port>/update/<docid>/<templateid>: Convert the document identified by docid and return a PDF
http://<host:port>/latex/<docid>/<templateid>: Show the generated latex document
http://<host:port>/pdf/<docid>/<templateid>: Show the last PDF without rebuilding or updating it
http://<host:port>/log/<docid>/<templateid>: Show the latex output from the last
http://<host:port>/clean/<docid>/<templateid>: Remove cached results to force a rebuild (should never be needed, but may be useful if updates to the template do not propagate)

In all these commands, templateid is optional. It can either point to a Google Docs document of which the plain text is used as a template or to a Google Drive folder. The folder can contain a main.tex file as the template and possibly other files such as images. Subfolders are not supported.

Installation and credentials: The process to compile the project and provide a credentials/api.json file are the same as above. The server can be started with the server binary in /target/universal/stage/bin and will listen on http://localhost:3000 by default.

The server expects that latex is installed locally. The server will store results in the out directory. It uses a temporary directory of the operating system to compile the generated latex. By default it calls pdflatex --pdf --interaction=nonstopmode main.tex twice with a 10 second timeout each.

Docker and Heroku/Fly.io: Optionally, the system can be easily deployed to Docker and a hosting service like Heroku or Fly.io. Due to needed credentials, no prebuild containers are provided, but the Dockerfile and the build-and-deploy.sh script should make this process hopefully fairly straightforward. The default Dockerfile uses the full texlive installation which is huge and may not be needed.