A Simple Parser That Converts HTML from OneNote to Markdown

13 January 2013

18:48

# Introduction #

[This article](http://www.codeproject.com/Tips/526908/A-Simple-Parser-That-Converts-HTML-from-OneNote-to) introduces OneNote2Markdown, a parser I made that converts the html file generated from OneNote (by sending to Word and save as html) to Markdown format, which can then be translated to a cleaner html by [any online Markdown parser](http://daringfireball.net/projects/markdown/dingus) later.

*  Written in F#, the tool works with OneNote 2010 and Word 2010. It handles normal paragraphs, headings, links, lists, inlined code and code blocks only.
*  The tool reads from "input.html" and writes to "output.txt".
*  The source code of the latest version can be viewed at [Bitbucket](https://bitbucket.org/colinfang/onenote2markdown/src). It requires [HtmlAgilityPack](http://htmlagilitypack.codeplex.com/) to compile.
*  The example pack contains this article in docx, html & Markdown formats, which would give a basic demonstration on how the tool works.

# Background #

I tend to take notes in OneNote. When I first time try to submit an article which is composed in OneNote, it is really a hassle to adapt the content to the template in [Code Project](http://www.codeproject.com) manually. So I decided to make a parser which would automate most of the formatting work for me.

# Implementation Overview #

## Preparations ##

*  An Active Pattern to pattern match if a text node has a certain ancestor such as `<b>` or `<i>`.

        let (|HasAncestor|) tag (node: HtmlNode) =
            node.Ancestors(tag) |> Seq.isEmpty |> not

*  A function to dig up a certain CSS property that a text node inherits from `style` attribute.

        let getPartialStyle cssProperty (node: HtmlNode) =
            let predicate node =
                // "property1:value1;property2:value2"
                let myMatch = Regex.Match(getStyle node, sprintf "%s:(.+?)(;|$)" cssProperty)
                if myMatch.Success then
                    Some myMatch.Groups.[1].Value
                else None
            // Gets the value for the closest cssProperty.
            node.Ancestors("span") |> Seq.tryPick predicate

*  A function to get a certain CSS property of a node from `style` attribute.

        let getPartialStyleSelf cssProperty (node: HtmlNode) =
            let myMatch = Regex.Match(getStyle node, sprintf "%s:(.+?)(;|$)" cssProperty)
            if myMatch.Success then
                Some myMatch.Groups.[1].Value
            else
                None

## Headings ##

*  Determines the heading type of a paragraph by checking its `font-size` & `color` CSS property as well as if it has `<b>` or `<i>` ancestor.

        match font, color, node with
        | Some "16.0pt", Some "#17365D", (HasAncestor "b" true)   -> H 1
        | Some "13.0pt", Some "#366092", (HasAncestor "b" true)   -> H 2
        | Some "11.0pt", Some "#366092", (HasAncestor "b" true)  & (HasAncestor "i" false) -> H 3
        | Some "11.0pt", Some "#366092", (HasAncestor "b" true)  & (HasAncestor "i" true)  -> H 4
        | Some "11.0pt", Some "#366092", (HasAncestor "b" false) & (HasAncestor "i" false) -> H 5
        | Some "11.0pt", Some "#366092", (HasAncestor "b" false) & (HasAncestor "i" true)  -> H 6
        | _ -> Normal

*  Uses` ## heading ##` syntax so that Markdown parser doesn't eat the last `#` contained in the heading.

        let headIt n text =
            String.Format("{1} {0} {1}", text, (String (Array.create n '#')))

## Code ##

*  Any text whose font is `Consolas` is considered as code, otherwise not.

        match getPartialStyle "font-family" textNode with
        | Some "Consolas" -> varIt text
        | _ -> text

*  Simplifies Markdown syntax by combining several inlined code pieces into one if they are separated by white-spaces (e.g. a b -> a b). Preserves the leading spaces so as to protect indentations and blank lines with code blocks (anything inside  is non-trivial and will not be removed later, e.g.   a ->   a). However, the limitation exists that the text itself cannot contain `.

        let simplifyVar (text: string) =

`    Regex.Replace(text, @"(?<=.)`(\s*)`", "$1") `

*  Differentiates code blocks from inlined code.

        let tryGetPureCode (text: string) =

`    let myMatch = (Regex.Match(text, @"^`([^`]*)`$"))`

        if myMatch.Success then
            Some (myMatch.Result "$1")
        else
            None

## Lists ##

*  Distinguishes between ordered lists and unordered lists by the symbol. Lists without symbols are considered as normal paragraphs without indentation.

        let listIt x text =
            match x with
            | "o" | "·" -> sprintf "*  %s" text
            | _         -> sprintf "1. %s" text

*  Gets the indentation by `margin-left:54.0pt` CSS property.

        let getIndent (node: HtmlNode) =
            let getMargin (x: string) =
                let unit = 27 // each level is 27
                let array = x.Split '.'
                let (success, x) = Int32.TryParse array.[0]
                if success then x / unit
                else failwith "indent parse error!"
            match getPartialStyleSelf "margin-left" node with
            | Some x -> getMargin x
            | None -> 0

## Links ##

*  Checks if a piece of text contains the link by looking for `<a>` in its ancestors.

        match textNode with
        | (HasAncestor "a" true) ->
            let ancestor_a = textNode.Ancestors("a") |> Seq.head
            linkIt text (ancestor_a.GetAttributeValue("href", "none"))
        | _ -> text

## Finalization ##

*  Gets the correct indentations and paragraph spacing for the whole content.

        /// Assumes in OneNote there are no spaces in front of a code block (indent by tabs).
        /// Assumes in OneNote the internal indentations of a code block are either all tabs or all spaces, never mixed.
        let review paragraphs =
            // indentOffset is used for nesting indentations.
            // If a benchmark line with indentation a, actually indents x, we set indentOffset = a - x.
            // So any line with indentation b, does actually indent b - indentOffset = b - a + x.
            let mutable listIndentOffset = 0
            let mutable codeIndentOffset = 0
            let oldCopy = paragraphs |> Seq.toArray
            let newCopy = Array.zeroCreate oldCopy.Length
            // Looks at the current paragraph and the previous paragraph.
            // I don't care about the first paragraph as it will be the title.
            // Uses "\r\n" so that Notepad reads correctly.
            for i in 1 .. oldCopy.Length - 1 do
                match oldCopy.[i - 1], oldCopy.[i] with
                | (Code _ | Listing _) , (Heading text | Basic text) ->
                    // Code block / list block ends, prepends and appends new lines, and resets both indentOffsets.
                    newCopy.[i] <- sprintf "\r\n%s\r\n" text
                    listIndentOffset <- 0
                    codeIndentOffset <- 0
        
                | (Heading _ | Basic _), (Heading text | Basic text) ->
                    // Appends a new line.
                    newCopy.[i] <- sprintf "%s\r\n" text
        
                | Code (_, a)          , Code (text, b)              ->
                    // Don't add a new line in between code blocks.
                    newCopy.[i] <- indentIt (b - codeIndentOffset) text
        
                | (Heading _ | Basic _), Code (text, b)              ->
                    // Code block starts, cache codeIndentOffset
                    // Indents 1 level only as Heading or Basic indents none.
                    newCopy.[i] <- indentIt 1 text
                    codeIndentOffset <- b - 1
        
                | Listing (_, a)       , Code (text, b)              ->
                    // Code block within a list requires 1 additional level on top of the list indentation.
                    // Code block starts, cache codeIndentOffset.
                    // Prepends a new line.
                    newCopy.[i] <- sprintf "\r\n%s" (indentIt (b - listIndentOffset + 1) text)
                    codeIndentOffset <- listIndentOffset - 1
        
                | Listing (_, a)       , Listing (text, b)           ->
                    // Don't add  a new line in between list blocks.
                    newCopy.[i] <- indentIt (b - listIndentOffset) text
        
                | Code (_, a)          , Listing (text, b)           ->
                    // Code block ends, reset codeIndentOffset.
                    // Prepends a new line.
                    codeIndentOffset <- 0
                    newCopy.[i] <- sprintf "\r\n%s" (indentIt (b - listIndentOffset) text)
        
                | (Heading _ | Basic _), Listing (text, b)           ->
                    // List block starts, cache listIndentOffset
                    listIndentOffset <- b
                    newCopy.[i] <- text
            newCopy