index.coffee.md no longer recognized as default main in node packages

[joyrexus]

For simple packages its nice to drop your source in an index.coffee file and forgo specifying "main:" in package.json.

When litcoffee hit the scene you could also use index.litcoffee or index.coffee.md as your default main file.

This worked with v1.6.1 but no longer with the new revision.

This is a minor issue, but I for one liked having simple packages in literate form.

I'll try to look through the changes since the last revision, but thought I'd point out the issue to those in the know.

[joyrexus]

This commit is the culprit.

When I add the extension back and rebuild, it fixes the issue. All tests pass except for test/cluster.coffee.

[jashkenas]

... but we can't blindly treat any .md file as CoffeeScript, right?

[michaelficarra]

Right. Which is why we should be finding out more about exactly why we are seeing this behaviour.

[joyrexus]

Curious, what would the harm be? Are you anticipating adoption of the literate style amongst javascript noders such that node needs to deal with *.js.md files? I guess that's conceivable.

[jashkenas]

Yes, and other Markdown documentation files living in the folder containing CoffeeScript etc. Any way you slice it, it would be incorrect to register the CoffeeScript compiler as the handler for all .md files ever loaded by Node.

Can you just use the .litcoffee extension? This sort of thing is one of the reasons why I prefer it as the standard, and not bastardized hyphenated-last-name extensions ;)

[joyrexus]

Yep. .litcoffee will work.

FWIW, I liked the fact that the .md files are automagically rendered by GH. However, for GH-readability it's true we're better off fencing the code blocks in a parallel file .md file to get the code highlighted ... and hopefully these Github folks will recognize and render .litcoffee files appropriately before long.

Final point: I do think that a .coffee.md, .js.md, .python.md, etc., convention enables tool support in that it announces to the tool (whether render or compiler) that the indented code blocks are all of a certain language (and can then be highlighted or safely compiled) in lieu of explicit code fencing. Simple example.

Not a big deal though. Feel free to close out the issue.

[jashkenas]

I'll close -- if @michaelficarra, you think that it's correct that this isn't something we can change without Node changing to recognize multi-dot extensions?

[bjmiller]

If I'm not mistaken, doesn't github have a board just like this, for just these sorts of requests?

[humanchimp]

-1 for multidot extensions. The idea is convoluted, and polluted. .tar.gz is a precedent, which was always ugly. no need to add or maintain unnecessary complexity and cruft. Much clearer to have one, correct extension.

[michaelficarra]

You both have valid points. Lets close this for now and start a discussion with the node guys. If they're willing to have require.extensions behave more like require.suffixes, we can revisit this later.

@joyrexus: do you mind opening the issue and linking to it from here?

[michaelficarra]

Also, a PR to remove .coffee.md would be great.

[jashkenas]

Nah, lat's keep it in there for completeness' sake, even if it's ineffectual at the moment.

[michaelficarra]

That's probably just going to end up confusing somebody that looks in the code later.