verso|recto - A Different Approach to Literate Programming

Literate programming is the art of preparing programs for human readers.

Norman Ramsey, on the noweb homepage.

Literate Programming (LP) tries to address a common problem with software systems: by reading the code one can discover how a thing is done, but not why it is done that way. Every program has a "theory of operation" embedded within its logic, but this is often hidden. Comments within the code and documentation of APIs are helpful but insufficient for this task. Often the documentation generation systems provided with a programming language do not provide a way to contextualize the code they describe within the larger system. For programs which rely on a more advanced mathematical background, it is also often difficult to embed equations or other markup in the documentation.

Existing LP tools such as WEB, noweb, and Org Babel attempt to address this by taking the following approach:

1. Embed the program source code within a prose document which explains the author's thinking.
2. Provide mechanisms for organizing code into abstract chunks, which can be recombined in different orders than they are defined and referenced throughout the document.
3. Process the combined document in one of two ways: either tangle the source code out of the document into a computer-friendly version, or weave the document into a form ready for humans to read.

Overall this is an improvement on inline documentation and can provide much more context to the reader than mainstream approaches. Unfortunately it does suffer some serious drawbacks as well:

• The source code is embedded within a markup file, making it inaccessible to language-specific tooling such as editors, compilers, and static analysis tools. In order to use these the code must first be tangled.
• Similarly, to build your literate program the end user must have the appropriate tools installed in addition to the language tooling needed to compile the source.
• Most programmers have spent years working directly in the machine-friendly source representation rather than a literate style, introducing a barrier to entry to writing literate programs. Porting an existing project to a literate representation is nontrivial.

verso|recto takes a different approach. It considers the source file as a first-class citizen to be referenced by the documentation. Rather than embedding source code in documents, or documents in source code, lightweight annotations are used to mark sections of interest within the code which can be easily referenced by prose documents. There is no tangle step, and source files remain fully valid inputs for compilers, editors, and other tools. There is no need to translate line numbers or formatting between literate sources and what the compiler sees.

What traditional LP systems get right

A lot of things! Here are a few ideas that the traditional systems pioneered and verso|recto borrows:

• Support a many-to-many relationship between LP documents and source files.
  • In noweb, a document may contain several root chunks which can each be sent to different source files, and multiple LP documents can be fed into the tangle tool at once (they are concatenated).
• The more modern tools support multiple programming and markup languages.
• noweb offers a pipelined architecture which makes it easy to insert processing stages to meet a user's needs.
• They generate indices and cross references within the human-readable output.

Using verso|recto

verso|recto is driven by annotations within your source files, defining regions which can be referenced by other documents. These regions are called "fragments".

Annotating a file

Annotations are quite simple. To mark a region of code and make a fragment, simply add a pair of comments around the region with the symbols @< and >@ followed by a unique ID. The ID can be any string of alphanumeric characters and the characters /, _, or -, though it should be both unique within your project and valid in the source file you're annotating. (The period character (.) is reserved, as it is used for inserting metadata about fragments.) Other characters may be added to the "safe" list in the future. If a character you want to use is not listed here, please file an issue on GitHub (or better yet, send a PR)!

Fragments can also be nested. This is particularly useful when you want to annotate a region of a source file that is already contained within a larger "outer" fragment. The annotations for the "inner" fragment will not appear in the output. This is best illustrated with an example; see the following source, prose, and output files.

Referencing annotations

In order to insert a fragment in another file, add a line containing the symbol @@ followed by the ID of the annotation (e.g. @@12345). When the file is woven using the recto command (see the next section), the line will be replaced with the contents of th fragment. You can add any markup you like around the line to provide formatting.

To insert a group of fragments, a regular expression can be used after the @* symbol. All of the fragments whose ID matches the expression will be inserted in place of the symbol, in lexicographic order by their IDs.

Sometimes it is also desirable to refer to metadata about a fragment. Currently, verso|recto supports the following metadata insertion operators:

1. Filename. @?id.file inserts the name of the file the fragment was drawn from.
2. Line number. @?id.line inserts the line number on which the fragment began.
3. Column number. @?id.col inserts the column number at which the fragment began. (Currently this value is always 0, as fragments always begin at the start of a line.)
4. Quick location. @?id.loc inserts the file name, starting line number, and column number for the fragment in the format file (line:col). This is useful if you just want to quickly refer to the metadata without futzing with the formatting.

Weaving a document for human consumption

The verso command will read all of the files specified on the command line, extract their fragments, and output the result to stdout. In turn the recto command will read fragments from stdin. This makes the two programs easy to use together via pipes:

verso main.rs lib.rs | recto build chap1.tex chap2.tex blog/home.md
      ^       ^              ^     ^         ^         ^
      +-------+              |     +---------+---------+
      |                      |                         |
      |                      |                         |
      +--- Source files      +--- Output directory     +--- Prose files

Each of the woven files is written to the output directory, provided as the first argument, in the same relative location as given on the command line. So, for example, the file blog/home.md above will be written to build/blog/home.md when it is woven.

Note that, although the two programs appear to run in parallel, verso won't send input to recto until it has successfully extracted fragments from all of the source files it was given and that recto will not start weaving files together until it receives those fragments. Because of this if verso fails, recto will also fail.

Full symbology

For reference, here is a table with the full symbology. Note that in the (hopefully rare) case that your language has symbols which collide with the defaults used by verso|recto, you can override them by using the listed environment variables.

The Name

Recto and verso are respectively, the text written or printed on the "right" or "front" side and on the "back" side of a leaf of paper in a bound item such as a codex, book, broadsheet, or pamphlet.

Wikipedia - Recto and Verso

(Note that, conveniently, the verso program goes on the left of the pipe while recto goes on the right.)

Contributors

• @nickpascucci
• @karlicoss
• @elidhu

Future Work

• Add support for custom formatting of annotation properties within the woven output.
• Paralellize file processing in Verso, and both reading from stdin and file reading in Recto.
• Add --fragments-from option to specify a source other than stdin for fragments.