single column Markdown pipe table rendered by GFM and PHP fails under Pandoc

[vassudanagunta]

Explain the problem.

Pandoc fails to render the following markdown as a table:

single column table | 
------------------- | 
item 1              |
item 2              |

As can be seen directly below as well as in Babelmark 3, this is rendered as a table by GFM, PHP Markdown Extra, and many others.

single column table
item 1
item 2

Pandoc does render the following form as a table:

| single column table 
| ------------------- 
| item 1             
| item 2              

Pandoc version?
Pandoc 2.17.1, the current version on pandoc.org/try

[jgm]

I see this in the code of the Markdown reader:

  -- surrounding pipes needed for a one-column table:
  guard $ not (length cells == 1 && not openPipe)

My guess is that this was put here for a reason! But what?

[mb21]

@vassudanagunta this should be fixed when using pandoc -f gfm in the next pandoc release...

[vassudanagunta]

It wasn't affecting me directly but I thought you'd want to fix this. I only discovered it as I run the same input cases through Pandoc, GFM and my project to test compatibility.

@jgm @mb21 would a report of significant behavioral differences between Pandoc and GFM be useful to you?

[mb21]

The long-term plan is for what is currently commonmark_x to replace the current implementation of pandoc's markdown...

perhaps we should mention that in https://pandoc.org/MANUAL.html#markdown-variants ?

[jgm]

@jgm @mb21 would a report of significant behavioral differences between Pandoc and GFM be useful to you?

Sure, but be sure you're comparing pandoc -f gfm behavior, not pandoc -f markdown which uses a different parser.

[vassudanagunta]

OK, sounds like you're not interested in differences between -f markdown and GFM, but only those between -f gfm and GFM. Makes sense.

But what about commonmark_x? Is it essentially CommonMark + the same extensions used for -f markdown, making it a strict superset of CommonMark but not of GFM? Some of the GFM-overlapping extensions (e.g. pipe tables) behave differently from GFM, and in my opinion better than GFM, so I'm not suggesting being a superset of GFM is the best thing.

If there's no quick answer and this should be a discussion, I can move this to the Pandoc forum.

[jgm]

commonmark_x is a superset of GFM in our implementation.
But it is NOT "the same extensions used for -f markdown"; it's a somewhat different set of extensions which includes many of those included by default in markdown, but not all of them, and also includes some different extensions like attributes.

[vassudanagunta]

commonmark_x is a superset of GFM in our implementation.

I couldn't find that documented anywhere. I think that's important to make clear, both for user expectations as well as to gate bug reports, whether it is "contractually" a strict superset. I don't know if I should be filling bug reports for differences in pipe_table behavior, for example.

[jgm]

Well, I spoke too soon. It's not quite a superset, because in gfm autolink_bare_uris is enabled, but commonmark_x doesn't turn this on by default. Both gfm and commonmark_x are defined as a set of extensions over commonmark; you can see which extensions are enabled by doing:

pandoc --list-extensions=gfm
pandoc --list-extensions=commonmark_x

(We prefer to make the information accessible in this form rather than updating the manual and potentially getting it out of sync with the code.)