Tomorrow, I'll see whether the suggested function that perplexity.ai gave me does what I want.

Next round of programming yields:

def panflute_to_bike_etree_nr(e, level=0) -> ET.Element:

    # questions about how header levels are handled as we put them into the etree
    etree = None
    stack = [(e, level)]
    # current ul_elem
    ul_elem = None
    heading_level = 0

    while stack:
        (e, level) = stack.pop()
        print("  " * level, e.tag)

        if is_inline_or_contentless(e):

            li_elem = ET.Element("li")

            # TO DO: handle rich text
            p_elem = ET.SubElement(li_elem, "p")
            p_elem.text = pf.stringify(e).strip()

            # 2 things to figure out: parent_ul (where to attach li_elem) and what the current ul_elem is

            if e.tag == "Header":
                print ("header", e.level, e.identifier, e.classes, e.attributes, pf.stringify(e))
                li_elem.attrib["data-type"] = "heading"
                li_elem.attrib["data-level"] = str(e.level)

                if e.level > heading_level:
                    # child
                    parent_ul = ul_elem
                    print ("e.level > heading_level", ul_elem)
                elif e.level == heading_level:
                    # sibling
                    # parent_ul has to be ul parent of ul_elem                    
                    parent_ul = ul_elem.getparent().getparent()
                else:
                    # e.level < heading_level
                    # uncle or higher
                    parent_ul = find_ul_ancestor(ul_elem, int(e.level)-1)

                heading_level = e.level
                ul_elem = ET.SubElement(li_elem, "ul")
                parent_ul.append(li_elem)
            else:
                ul_elem.append(li_elem)

        else:
            if e.tag == "Doc":
                etree = empty_bike_etree()
                ul_elem = etree.xpath("//ul")[0]
            else:
                # BulletList
                # OrderedList
                # ListItem
                print("block", e.tag)

            try:
                for c in reversed(e.content):
                    stack.append((c, level + 1))
            except AttributeError:
                pass

    # clean up etree by adding ids
    etree = add_ids_to_bike_etree(etree)
    return etree

My first pass -- taking a recursive approach:

def panflute_to_bike_etree(pfe) -> ET.Element:
    """
    pfe: panflute element    
    base_header_level: the base header level to use for the document
    TO DO: handle different header levels -- might need to abandon this approach in favor of non-recursive approach with a manual stack
    """

    if is_inline_or_contentless(pfe):
        p_elem = ET.Element("p")
        p_elem.text = pf.stringify(pfe).strip()
        return p_elem
    else:
        if pfe.tag == "Doc":
            etree = empty_bike_etree()
            ul_elem = etree.xpath("//ul")[0]
        else:
            ul_elem = ET.Element("ul")

        try:
            for c in pfe.content:
                e = panflute_to_bike_etree(c)
                li_elem = ET.SubElement(ul_elem, "li")
                li_elem.append(e)
        except AttributeError:
            pass
        else:
            if pfe.tag == "Doc":
                etree = add_ids_to_bike_etree(etree)
                return etree
            else:
                return ul_elem

Until today, I didn't worry about the specific version of pandoc I used. However, as pandoc continues to evolve, there might be differences that affect the computations I conduct with pandoc, pypandoc, panflute, and the pandoc Python library. I don't have the energy this evening to write what I learned. Stay tuned.

Not much to say except to say that I'm on track.

Today, I am focused on creating a first cut of a pandoc writer for Bike. I will start by having a writer emit the simple output of a valid but empty Bike document. The next step would be to have the writer flatten a pandoc AST into a single list of plain text Bike nodes. I defer dealing with the rich text and hierarchical structures until I have the simple writer in place. Dealing with rich text and containment of elements requires me to better grok the pandoc AST, particularly what can contain which other element. I will use pandoc (Python library) and panflute interactively in a Jupyter notebook to facilitate that grokking. I aim for good coverage of pandoc AST examples by writing pandoc markdown, which I then convert to pandoc JSON (note the documentation on pandoc markdown.

Ideally, my pandoc Bike reader and writer will enable lossless roundtripping of Bike conversion. In other words, if I use the reader to convert a bike document to pandoc JSON and then use my pandoc Bike writer to convert the JSON to a Bike document, the resulting Bike document should essentially be the same the original document.

A question to answer as I dive into the pandoc writer for Bike: how to handle parts of the pandoc AST that I don\'t know how to translate into Bike elements but would nonetheless like to park in some placeholding structure?

I\'m pleased that I reached a programming milestone. Now whenever I save my main Bike outline (overall.bike), my CLI for interacting with Bike (and Bike outlines) is fired off (specifically, bike.py pandoc). I use Hammerspoon to look for changes in the enclosing directory for overall.bike and then I have some logic to check for telltale changes in overall.bike. I was stuck for a while: It turns out that the tricky part is detecting reliably whether there is data on stdin. My solution is good enough for now, but I will probably revisit it.

I've been writing all day, with a fluency that I've not had in a long while. What I've written isn't ready for sharing with the public, however. I've also made some good progress in getting my script to run but I'm stuck on an issue that might take 10 more minutes to solve -- or hours. I'm surprised that I'm serenely accepting however long it ends up taking. What choice do I have?

I set the stage for command groups in bike.py with [my latest change to bike.py](added sel2pb command to bike.py · rdhyee/python-learning@a161fe9).

It seems that Keyboard Maestro doesn't allow for watching for changes in files that already exist. Maybe, more likely, I've just not been able to figure out how. I have some progress using hammerspoon though. Here's a relevant part of my init.lua:

function list_true_flags(flags)
  local true_flags = " "
  for key, value in pairs(flags) do
    if value then
      true_flags = true_flags .. key .. " | "
    end
  end
  return true_flags
end

function fileModifiedCallback(paths, flagTables)
  for i, file in ipairs(paths) do
    if file == "/Volumes/ryvault1/MacHD/Users/raymondyee/obsidian/MainRY/bike/overall.bike" then
      local flags = flagTables[i]
      print(i, file)
      print_true_flags(flags)

      local timestamp = os.date("%Y-%m-%d %H:%M:%S")
      -- local message = timestamp .. " overall.bike changed\n"

      local logFile = io.open("/tmp/hschange.log", "a")
      logFile:write(timestamp .. list_true_flags(flags) .. "\n")
      logFile:close()
    end
  end
end

watcher = hs.pathwatcher.new("/Volumes/ryvault1/MacHD/Users/raymondyee/obsidian/MainRY/bike", fileModifiedCallback)
watcher:start()