Don't fill code blocks

[vermiculus]

Code is usually line-sensitive – it's inappropriate to fill them with fill-paragraph.

[syohex]

Could you show us example code which reproduces your issue ?

[vermiculus]

(insert "```
hello
world
```")
(fill-region 0 (point-max))

[xiongtx]

This actually seems like a hard problem.

It's easy to modify paragraph-start to stop filling code blocks formed by lines that begin with 4 or more spaces with the regexp " .*".

The trickier part is recognizing code blocks between two sets of triple tick marks. The paragraph-start and paragraph-separate regexps only apply to a single line, so in the following situation (where | indicates point):

```
|foo
bar
```

you have no way of knowing that you're in a code block and not a regular paragraph.

GitHub also stupidly chose start and end delimiters to be the same, so regexp-based solutions are in for a world of pain. The syntax was clearly designed for parsing, not text matching.

[vermiculus]

Yep – I encountered the same complications when I tried to implement. I think the solution is going to have to re-implement commands like fill-paragraph entirely.

[jrblevin]

I was able to fix this without reimplementing fill-paragraph entirely. Instead, a custom fill-paragraph-function blocks filling when in a code block. There is a test case, but bet me know if you see any problems.

[jrblevin]

Actually, I just realized that the original report was about fill-region not fill-paragraph. This only fixes the latter. Oddly, fill-region seems to work on its own, via fill-region-as-paragraph, without calling out to fill-paragraph.

Even in emacs-lisp-mode, if one uses fill-region it fills code as paragraphs. For example, start with this in the buffer:

(defun markdown-show-version ()
  "Show the version number in the minibuffer."
  (interactive)
  (message "markdown-mode, version %s" markdown-mode-version))

Then issue (fill-region (point-min) (point-max)) and you're left with a jumbled mess:

(defun markdown-show-version () "Show the version number in the
minibuffer."  (interactive) (message "markdown-mode, version %s"
markdown-mode-version))

[vermiculus]

I would've thought fill-region would be based on fill-paragraph (or perhaps vice-versa). Should this be reported as an Emacs bug?

[jrblevin]

I agree--I'm surprised too. It does seem to be a common issue, so maybe a bug report is warranted. For example, I checked latex-mode. It rightly won't fill a verbatim environment when fill-paragraph (M-q) is called directly, but fill-region doesn't respect verbatim.