Skip to content

TomasHubelbauer/markright

Repository files navigation

MarkRight

MarkRight is a tool for didactic literate programming.

It works by scanning a MarkDown files for fenced code blocks adorned with extra instructions and acting on those instructions to execute code, create files etc.

The artifact of the MarkRight document exection might be a file, a program or a process that has happened and left desirable side effects.

MarkRight documents are primarily append-only capturing new changes to apply to the existing instructions and their artifacts.

Examples

By default, MarkRight will ignore code blocks which lack a language tag or whose language tag it doesn't have support for.

A code block like this will have no effect:

```
Hello, world!
```

MarkRight will exit with empty output upon executing such a document.

The most basic example of a MarkRight document that would give output would be this document where the code block has the special stdout language tag:

```stdout
Hello, world!
```

Run markright demo/stdout/readme.md or markright demo/stdout to execute this example.

The stdout language tag instructs MarkRight to dump the code block's text to the standard output stream of the MarkRight process.

stderr is supported with analoguous behavior for the standard error stream and there is a respective example of this in demo/stderr. The standard error lines appear in red and you can verify they are in stderr by splitting the I/O streams and writing them into respective files: markright demo/stderr 2>> error 1>> output.

See Handlers for the list of supported language tag handlers.

Another similar example would be the JavaScript Hello World expression one in demo/javascript-expression which demonstrates that JavaScript code blocks that produce output redirect their output to MarkRight's output stream.

See also demo/javascript-exception to see the special-cased behavior of exceptions getting written to the standard error stream.

Note that by default MarkRight looks for a readme.md file in the current directory.

MarkRight code blocks can also create and update files. To create a file, prepend a code span with the file name suffixed by a colon atop the code block it relates to with nothing but whitespace in between:

`file-name.ext`:

```
Hello, world!
```

This code block's text will get written to file-name.ext even though it has no language tag, because it have an associated path. If there was a language tag, its handler's standard streams would get redirected to the file.

Handlers

txt

Alias for the stdout language tag. Useful for forcing the code block to be included in the output which it wouldn't without a language tag. Shorter than stdout and without the stdout/stderr parity baggage if only the former is used in the document. Implied when the code block has an associated path even if it has no language tag.

stdout

Writes the code block's text to the MarkRight's standard output I/O stream.

stderr

Writes the code block's text to the MarkRight's standard error I/O stream.

js / javascript

Evaluates the code block as JavaScript using Bun's eval function. Writes the output of the last expression, if any, to standard output, unless an exception was thrown, in which case it gets written to standard error.

Note that this handler uses eval and as such does not support async/await. See oven-sh/bun#14873 for more information.

Note that JSX is not supported.

ts / typescript

Transpiles the provided TypeScript code to JavaScript using Bun's transpiler and executes it using the JavaScript handler.

Note that TLS and JSX is not supported, see the JavaScript handler section for more information.

Note that Bun doesn't support evaluating TypeScript directly yet, see here: oven-sh/bun#11976

Development

Run tests using bun test. Use .only to work on individual tests.

Run using bun . to run on readme.md or bun . /directory/file.md to run on a non-default document.

To-Do

Return GitHub Actions workflow and its corresponding readme badge

  • Use Bun GitHub Actions support
  • Run the unit tests
  • Run Bun's bundler to product executables for all platforms
  • Bump package.json version using bun version patch if supported
  • Stage the package.json Git change
  • Use jq to query the new version
  • Use actions/create-release to create a new release with the version
  • Use actions/upload-release-asset to upload the platform executables
  • Commit and push the package.json change

Distribute to JSR or maybe even NPM as an executable package

Do this in a Bun-friendly way, something easy, not a PITA.

Add a --watch command to keep watching the file for changes

Clear the terminal in between runs.

Improve the parser to not require duplicating mode sigils with inline path

Currently tag ..file-name.ext is needed to specify a tag and the inline path while also distinguishing it from the meta.

Doing it this way simplifies the parser implementation but makes for a worse user experience. It is a worthwhile trade-off for now, but should be improved down the line.

One option to solve this would be to require the paths to start with a period like in ESM path module specifiers, but I don't like that option too much.

Add unit tests for the demo files

Create a wrapper to catch console.log and console.error calls like I already do in processBlocks and compare the expected output with the real output.

In file management demos, also check the files on disk and clean up after.

Find a way to allow diff/patch code blocks to not have hunk context line

The @@ line. The patch utility seems to require these, it will not try to match and patch loosely. The git apply command has --unidiff-zero which seems to allow not providing the context lines but it was giving me trouble probably because the file I am trying to apply to is not necessarily tracked by Git?

If there is no way to get rid of these, maybe I could generate the hunk context by assuming the changes are for the whole file? This would massively lessen the utility of this code block though. Maybe there is a path where I derive the start and end lines and only require that the diff/patch code is a single hunk maybe?

Consider allowing to specify what shell to use in the shell handler

Support the bash, zsh etc. language tags and take the choice of the shell into an account in the execution.

Add examples that showcase the various shell's features to prove they shell selection works.

Return support for the Windows Sandbox feature

Run PowerShell scripts in the Windows Sandbox and get back their outputs. Also maybe file manage in the Windows Sandbox?

Recover the CLI calculator example and reimplement it in Bun

Once I have the file management sigils and operations figured out, I should be able to reimplement that example in MarkRight again with the new version running on Bun to showcase the features it uses.

https://raw.githubusercontent.com/TomasHubelbauer/markright/4fbead6a0b5a769d3ab04f1e3cb0c91b82df0e00/example/node-cli-calculator/readme.md

Add support for leaving out diff/patch meta and the _ sigil

When multiple code blocks in a row operate on the same file, it would be a pain to keep repeating its name, so let's make these two changes:

  1. Add the _ sigil to code blocks without external path which makes the block inherit the path of the last block that had a path
  2. Allow the diff and patch blocks to not specify meta in which case the path of the last block that had a path is used
  3. Make code blocks without external path assume the path of the last block that had a path when they use the ! or ? sigil

Add a subcommand to take in the difference between the file and the document

When the user goes and runs the document, then changes the files, make this new subcommand determine the differences between what is there and what MarkRight had generated and add the diffs at the end of the document.

Make a VS Code extension for automatically colleting MarkRight changes

This is related to the above task about bringing in diffs after user made edits to the artifacts MarkRight dropped. Make it possible to just work on files and have the MarkRight document be built for you with the possibility to jump in and add your comments as needed.

Consider adding a new file management sigil to tee to file and terminal both

This could be useful?