Merge Consecutive Blockquotes in Markdown

Yihui Xie 2023-09-14

Older versions of Hugo use a library named Blackfriday to convert Markdown to HTML, which is not compliant with CommonMark (unlike Goldmark in newer versions of Hugo). One problem with Blackfriday is that blank lines inside blockquotes are treated as part of the blockquotes. For example,

> One

> Two

will be converted to:

<blockquote>
  <p>One</p>
  <p>Two</p>
</blockquote>

With the CommonMark specs, it should be converted to two blockquotes because of the blank line:

<blockquote>
  <p>One</p>
</blockquote>
<blockquote>
  <p>Two</p>
</blockquote>

To generate a single blockquote, the blank line must also start with >.

A lot of my old posts have this problem. I guess it might be solved by a Pandoc (Lua) filter, but I do not need rigor, so I wrote an R function to process my Markdown files instead (like last time):

# add > to empty lines to merge consecutive blockquotes
merge_bq = function(x) {
  if ((n <- length(x)) < 3) return(x)
  
  r = '^(>( >)* ?)(.*)$'
  i = grep(r, x)
  if (length(i) < 2) return(x)

  # exclude consecutive lines that start with >
  i = i[c(diff(i) > 1, TRUE)]
  if ((n <- length(i)) < 2) return(x)
  
  for (k in 1:(n - 1)) {
    i1 = i[k]; i2 = i[k + 1]
    # from line #i1 on, a line should not be blank unless
    # the rest of lines are all blank till line #i2
    i3 = (i1 + 1):(i2 - 1)
    b = xfun::is_blank(x[i3])
    if (any(b) && !all(tail(b, -which(b)[1]))) next

    p = sub(r, '\\1', x[i1])  # '> ' (or nested '> > ...')
    p = sub(' *$', ' ', p)    # ensure trailing space
    x[i3] = paste0(p, x[i3])
  }
  
  x
}

The code is nowhere near elegant or easy to understand. I’m leaving a record here nonetheless because I have written a lot of dumb code like this in my career. I do not want to pretend that I’m always writing elegant code. On the contrary, most of code I write is perhaps quite dumb and boring.

Recently I talked with Peng Liu at JMP about our software development career since he was hoping to organize a JSM session on this topic (which did not work out, unfortunately). Although I did not plan to attend JSM, I thought for a while what I could share if I were to give a talk. One thing that came to my mind was how much formal training in computer science we need to be a successful software engineer. I was thinking about this because I did not receive any formal training, and I do not really know much about algorithms. If I were to look for a software development job, I guess I’d be immediately rejected due to my poor knowledge in computer science.

But I can write for loops (with care and patience) and use regular expressions. I feel I have been making a living on them (thankfully the impostor syndrome cannot beat me). Usually I write code very slowly because my brain is slow. The above for loop took me nearly two hours. On my mind, I was just wrestling with a few lines of text—some start with >, and some are blank.

Some quick tests:

merge_bq(c('> a', '', '> b'))

> a
> 
> b

merge_bq(c('> a', 'c', '', '> b'))

> a
> c
> 
> b

merge_bq(c('> a', '', 'c', '> b'))

> a

c
> b

merge_bq(c('> a', '', '> > c', '', 'd', '', '> b'))

> a
> 
> > c

d

> b

merge_bq(c('> a', '', '> > c', 'd', '', '', '> b'))

> a
> 
> > c
> > d
> > 
> > 
> b

All seem good, so I batch processed all my posts:

# process all writable [R]md files
files = blogdown:::list_rmds(pattern = blogdown:::md_pattern)
files = files[file.access(files, 2) == 0]  # writable
for (f in files) xfun::process_file(f, merge_bq)

Thanks to Git, I could do a manual check on the changes before I commit them (I did revert a few changes because blank lines actually meant to be blockquote separators).