I’m running into an issue on parsing non-standard nested lists from html to the PM model (based on https://www.w3.org/wiki/HTML_lists#Nesting_lists ). The parsing does just fine with the standard (with LI tags wrapping the nest, rather than as seen below).
Which is kind of interesting because both render identically in the browser (normally … like the below markdown conversion … but I’d understand why it’s an inconvenience for ProseMirror to deal with this due to the tree structure it uses).
one
two
three
And the following is a spec I added to ProseMirror’s tests:
t("unordered_list_nest_nonstandard",
doc(ul(li(p("one")), ul(li(p("two"))), li(p("three")))),
"<ul>" +
"<li><p>one</p></li>" + // <= nonstandard form has </li> here.
"<ul>" +
"<li><p>two</p></li>" +
"</ul>" +
"" + // <= Where the </li> *should* be according to the standard.
"<li><p>three</p></li>" +
"</ul>")
Fails with the following:
dom_unordered_list_nest_nonstandard: Failure: types differ at doc.0.1
in doc(bullet_list(list_item(paragraph("one")), bullet_list(list_item(paragraph("two"))), list_item(paragraph("three"))))
vs doc(bullet_list(list_item(paragraph("one")), list_item(bullet_list(list_item(paragraph("two")))), list_item(paragraph("three"))))
The reason I’m posting this is because I have a ton of old HTML code from another text editor that sadly used the non-standard so I’d have to figure out how to parse and fix all of the conditions. (But also, others will run into this issue if they are scraping html out in the web)
So I’m wondering how difficult it’d be for prose mirror to be able to deal with this type of nested structure appropriately?
ProseMirror uses the browser’s HTML parser to parse HTML into a DOM tree before converting it, so, since HTML5 specifies that that inner <ul> will get an implicit <li>, you’re out of luck here, and will have to use some external tool to clean up your HTML.
I could hook into the transformPasted event and try to undo the incorrect node setup, but I’m wondering if that isn’t the right approach, or rather there is something fundamentally broken about the parsing of bulleted lists, as I don’t believe Google Doc uses the incorrect specification for ul/li/ol. Other editors handle the pasted content correctly (Ex: CKEditor, textAngular, Redactor)
On closer look, it appears that I got this wrong. The standard, as far as I can see, disallows lists as direct children of other lists, but MDN states the content of a list can be “zero or more <li> elements, eventually mixed with <ol> and <ul> elements.”, where I suppose “eventually” means “if your browser feels like it”. Browsers do seem to allow such structure. I’ve pushed this patch, which adds a somewhat ugly kludge to ‘normalize’ this kind of input.
Yes. Since the nested list isn’t a child of the first list item in the HTML, ProseMirror’s parser won’t turn it into a child of that item in the ProseMirror document. If your schema allows list items to start with lists, you’ll get a somewhat different structure for this (<ul><li>Depth1</li><li><ul><li>Depth 2</li></ul></li></ul>) but that still doesn’t really reflect what you’re looking for here.
Thanks for letting me know. I’ll start on that now…
