Add an option to prevent colgroup column width ratios being added to pipe tables

[daviewales]

If you create a markdown table using the pipe tables extension, and any line of the markdown source is more than the column width, then the relative cell widths in the output are calculated based on the ratio of dashes under the headings in the markdown.

For HTML output, this results in the insertion of an HTML colgroup block which forces the proportions of the cells.

However, I'm using R Markdown to insert computed values into my table cells, which means that the cells could realisticly have any number of characters in them, and I'd prefer to rely on the default column widths that the tables would have if there were no colgroup block at all.

I understand that if I use any of the other table extensions I'll get this, but as I'm inserting computed column values it makes the other table exentions unviable, as they rely on getting the spacing lined up exactly right, which I can't do here.

The workaround I have found to prevent this is to use the pandoc argument --columns 1000 , but it would be nice to be able to explicitly disable this feature.

[tarleb]

Here's a Lua filter that does this:

function Table (tbl)
  tbl.colspecs = tbl.colspecs:map(function (colspec)
      local align = colspec[1]
      local width = nil  -- default width
      return {align, width}
  end)
  return tbl
end

See the Manipulate Markdown via Pandoc Lua filters in the R Markdown Cookbook for more info on this approach.

[jgm]

and any line of the markdown source is less than the column width

I think you meant "more than the column width."

[jgm]

I think the filter solution is probably adequate. It's more complicated if you need this to apply to some but not all tables, but perhaps then you could enclose the relevant tables in a Div?

[daviewales]

Thanks @tarleb and @jgm, the filter does exactly what I want it to do.
Personally I think this would be more discoverable if it were a pandoc option (or if there were another type of pipe table where this is the default).

However, if this is not preferred, perhaps the docs could be updated to point to this filter solution to make it easier to find?

[daviewales]

Also, I've just discovered that this filter doesn't work with older versions such as Pandoc 2.5 (pandoc-types 1.17.5.4) which is the default on Ubuntu 20.04. Is there a way to fix the filter to work on older versions?

I'm getting this error:

pandoc -s -o test.html test.md --lua-filter filter-html-auto-table-column-widths.lua
Error running filter filter-html-auto-table-column-widths.lua:
filter-html-auto-table-column-widths.lua:4: attempt to index a nil value (field 'colspecs')
stack traceback:
        filter-html-auto-table-column-widths.lua:4: in function 'Table'

I'm guessing that colspecs doesn't exist in older versions of pandoc?

[tarleb]

I believe that this version should work with both pandoc versions, but I didn't test it:

function Table (tbl)
  if PANDOC_VERSION >= '2.10' then
    tbl.colspecs = tbl.colspecs:map(function (colspec)
        local align = colspec[1]
        local width = nil  -- default width
        return {align, width}
    end)
  else
    for i, w in ipairs(tbl.widths) do
      tbl.widths[i] = 0
    end
  end
  return tbl
end

[tarleb]

Related issue: #2165

[daviewales]

Thanks @tarleb. I had to tweak it slightly as I was getting the error attempt to compare string with table. The following works:

-- Unset the width attribute of HTML colspecs in tables
-- See https://github.com/jgm/pandoc/issues/8139
function Table (tbl)
  if PANDOC_VERSION[1] >= 2 and PANDOC_VERSION[2] >= 10 then
    tbl.colspecs = tbl.colspecs:map(function (colspec)
        local align = colspec[1]
        local width = nil  -- default width
        return {align, width}
    end)
  else
    for i, w in ipairs(tbl.widths) do
      tbl.widths[i] = 0
    end
  end
  return tbl
end

[jgm]

I'm adding a FAQ entry which should make this more discoverable.

[gregdan3]

Just ran into this myself.

The issue stood out badly because I'm using ligatures with wildly different sizes from the original text, but of course pandoc can't know I'm providing a font with ligatures when it gets to the CSS.

This is Toki Pona. Third from the top is the joke word "kijetesantakalu", which demonstrates the issue nicely!

For comparison, normal-appearing ligatures (still affected, but the underlying words are more consistent in length).

[binyomen]

The version check in the code above doesn't seem to work once you get to Pandoc version 3, since the minor version is no longer 10 or greater. I think what we want is:

-- Unset the width attribute of HTML colspecs in tables
-- See https://github.com/jgm/pandoc/issues/8139
function Table (tbl)
  if PANDOC_VERSION[1] > 2 or (PANDOC_VERSION[1] == 2 and PANDOC_VERSION[2] >= 10) then
    tbl.colspecs = tbl.colspecs:map(function (colspec)
        local align = colspec[1]
        local width = nil  -- default width
        return {align, width}
    end)
  else
    for i, w in ipairs(tbl.widths) do
      tbl.widths[i] = 0
    end
  end
  return tbl
end

This should work until a future breaking change alters how column specs are accessed.

[Julynx]

I was able to work around this by adding:

/* style.css */
col {
  width: auto !important;
}

To the CSS stylesheet, then passing --css style.css