Skip to content
Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

Literate Markdown Tangle

lmt is a tool for extracting text from the code blocks in markdown files. It allows programmers to write in a literate programming style using markdown as the source language.

Installing lmt

lmt is a self-contained Go program written using the LP paradigm. The source is committed alongside the markdown source of this repository for bootstrapping purposes.

You require the Go language if you don't already have it.

To build the tool:

git clone
cd lmt
go build

This will build the binary named lmt for your platform in the current directory. You can use the -o $path argument to go build to build the binary in a different location. (i.e. go build -o ~/bin/ to put the binary in ~/bin/.)

A note for Nix(OS) users

This repo also comes with a shell.nix file. While an existing version is included, like lmt itself, this is mainly for bootstrapping purposes. To compile it, use lmt


To observe lmt at work, put this file in an empty directory, cd to that directory, and run lmt Now look in the directory and you'll see files extracted from the code blocks alongside this markdown file. In literate programming lingo, this extraction is (somewhat counterintuitively) called "tangling." Generating documentation from the source is called "weaving", and lmt leaves that to existing markdown renderers (such as the GitHub frontend.)

lmt is language agnostic. The below demonstration of features is written in (very trivial) C++ to demonstrate using other languages.

Tangling into a file.

The markup for the code block below starts with ​```cpp hello.cpp +=:


int main() {
    <<<body of main>>>

The header says 3 things:

  1. cpp: the code block is written in C++. In the rendered markdown output, that affects syntax highlighting, to lmt it means that language-appropriate pragma directives will be added so that when debugging the extracted code, your debugger will show you the line in the original markdown source file. (If you don't want this effect, you can just use an unrecognized language name like cxx).
  2. hello.cpp: The code block will be written to the file hello.cpp.
  3. +=: The code block will be appended to the most recent code block defining that file, rather than overwriting its content. Since we haven't written anything to hello.cpp yet, the effect is the same, but this demonstrates the ability to use it.

Macro References

The <<<string>>> sequences in the body of the code block are called "macro references." An LMT "macro" is just a variable whose value can be extracted from one or more code blocks, and will be substituted wherever its name appears in triple angle brackets on a line. There are no arguments to lmt macros.

If we were to run lmt on this file at this point, we would get the warnings:

Warning: Block named copyright referenced but not defined.
Warning: Block named includes referenced but not defined.
Warning: Block named body of main referenced but not defined.

This allows us to stub in a macro reference whenever we want in our code, and only later define them in whatever order best fits our prose. When there are no more warnings, the hello.cpp file should build (assuming we didn't include any syntax or other compiler errors.)

Macro Content

The markup for the code block below starts with ​```cpp "body of main"

std::cout << "Hello, werld!" << std::endl;

The double quotes around body of main mean that the code block will be extracted into a macro of that name. You can see where its value will be injected into hello.cpp via <<<body of main>>>, above. Since there's no += at the end of the block's first line of markup, this code block overwrites any existing value the macro might already have (but since it has no existing value, it's a wash).

lmt uses quotation marks to differentiate between macros and file destinations. If a name is encased in quotes, it's a macro, if not, it's a file.

We can later re-define a macro to overwrite it (​```cpp "body of main", again)

std::cout << "Hello, world!" << std::endl;

lmt parses each file passed on the command line in order. The last definition of a macro will be used for all references to that macro in other code blocks (including blocks which preceeded it in the source.)

Appending To A Macro

We can use #includes to demonstrate += on macros. There are two includes in this program. The markup for the following block starts with ​```cpp "includes", which causes the (empty) value of the includes macro to be overwritten.

#include <iostream>

The markup for the next code block, however, starts with ​```cpp "includes" +=, which causes the block to be appended to the includes macro.

#include <numeric>

Its value is now:

#include <iostream>
#include <numeric>

(the code block above is not being tangled).

Hidden content.

The raw markdown in this file contains a comment containing a code block with a copyright notice. It looks a bit like this one:

```cpp "copyright"
// Copyright 42 BCE not the actual copyright

If you're reading the rendered markdown in your browser, you can't see the actual comment, but it still gets tangled into the copyright macro, which is substituted into hello.cpp by the <<<copyright>>> macro reference. This technique lets you tangle content that you don't want showing up in the documentation.

What Tangles and What Doesn't.

We can tangle into a random data file (​```csv data.csv)

foo, bar, baz,
qix, qux, quux,

You need to specify both a language and a destination (macro or file) if you want the code block tangled:

No language (​``` bar.txt—note the space):

This doesn't get tangled anywhere

No destination, but includes syntax highlighting (​```cpp)

auto x = "nor does this";

But any language string and filename (​```arbitrary foo.txt) will do

This gets tangled
into foo.txt.

Running lmt on this file at this point should generate the files data.csv, foo.txt, and hello.cpp with the expected contents and produce no warnings.

Building lmt from source

While the tangled source of lmt is included for bootstrapping purposes, the markdown is considered the canonical version. The Go source can be re-extracted with:


If you'd like to read the source, the order of the files and patches were written is the same as passed on the command line.

  1. Basic Implementation
  2. Whitespace Preservation
  3. Subdirectory Files
  4. Line Numbers
  5. Indented Blocks

Small bug fixes can be contributed by modifying the prose and code in the existing files. Larger features can be included as a patch in a new file.


lmt is primarily authored by Dave MacFarlane (@driusan). Bryan Allred (@bmallred) improved the parsing code to include the metadata in the code block header rather than a rendered markdown header. @mek-apelsin wrote the patch to include pragmas for line numbers, and Dave Abrahams (@dabrahams) wrote the demo of features in this README, making it more user-focused (it previously dove straight into implementation.)


literate markdown tangle







No releases published


No packages published