Markdown Part 4 – Outlining

This is part of the “Markdown mode” series:

Part 3.5 – Posted on the VS Gallery
Part 3 – A live tool window previewer
Part 2 – Writing a classifier
Part 1 – Markdown!

A few nights ago, around midnight, I was in that I’m-pretty-tired-but-too-tired-to-fall-asleep-now state, and so decided to find something small that I could add to one of my extensions. Since Markdown Mode is the ripest for this type of thing, I decided to add outlining (a.k.a. folding) support for Markdown files:

A collapsed region, while editing this blog article


Outlining regions tend to follow some type of structural decomposition of a file; in C#/VB/C++/etc., you get outlining regions for structural units like namespaces, structs/classes, methods/functions, and (sometimes) blocks. In Markdown, while there isn’t an exact equivalent of these groupings, you can use the headers in a Markdown file to roughly break the file into sections.

The logic for this is fairly simple:

  1. A section starts at a header, such as an H1 or H3.
  2. A section ends when you run into either:
    • The end of the file, or
    • A section of equal or greater “size”, where H1 > H2 > … > H6.
  3. A section contains all sections between the start and endpoint (much as a namespace contains classes which contain methods which contain statements).

This code lives in MarkdownParser.ParseMarkdownSections, which is in MarkdownParser.cs. It returns an enumeration of these sections from first/largest to last/smallest.


Once the decomposition logic exists, the outlining code is simple. Like many of the other features we’ve dealt with, this also uses tagging, exporting an ITagger<IOutliningRegionTag>. There’s also a bit of logic for computing the collapsed form (what shows up in the editor when you collapse the region) and the collapsed hint form (what you see in a tooltip when hovering over the collapsed form). Still, the total size of OutliningTagger.cs is just under 100 lines, since the hard part is computing the sections.


Since we already have the structure, it’s nice to be able to see the structure and quickly navigate to different sections. In regular ol’ code files, we get this in the navigation bar at the top of the file. Since I only need one navigation bar (there isn’t a module/method split for Markdown files) and I don’t want to waste vertical space, I just stuck a simple combo box onto the existing margin. It’s exactly exactly pretty, but it gets the job done. I suppose there’s an interesting conversation to be had here, in how much time/effort I should spend to make this visually congruous with the navigation bar; for my lack of talent in this area, it would probably take me longer to figure that out than it did to write this whole feature (which was about 45 minutes), so it isn’t high on my list.

What’s left?

Since outlining wasn’t exactly on the list of features, I haven’t exactly knocked anything off the TODO block. From this point on, however, I think I’m just going to add new features as I find them useful, so I’ve gotten rid of the “TODO” list and kept only list of things I’ve finished. I’ve also dropped the “no longer doing” portion of the list.

  • Outlining support
  • Spell checking support – my teammate Roman built a sample that does spell checking, so that would be nice to add, especially since I’m writing prose.
  • Put the extension out on the VS Gallery!
  • Syntax coloring (the obvious one); things like bold, italics, links, pre blocks and code blocks, etc.
  • Turn URL labels into clickable hyperlinks – this is pretty easy to do with the new URL tagging support in the editor.
  • A tool window for showing the current output (as a webpage) of the buffer I’m editing, updating live while I’m editing it. Alternatively, this may be just another document or a margin, depending on where I end up. I’m not very good at visual schtuff, so it’ll probably be as simple as it can be and still get the job done.
  • Syntax support for regular html – Markdown lets you use intersperse regular html into your Markdown files, so it would be nice to get these to be correctly syntax highlighted. This may actually be really simple (just using a ContentType that derives from html), but if it isn’t, it isn’t high on my list, mostly because I don’t tend to use it for anything other than anchor tags.
Comments (6)

  1. Cameron Peters says:

    For outlining, when the buffer changes, you just dirty the entire file… the comment in the code suggests that you are planning to revisit the code and provide a better implementation.  What is the best way to determine the differences between two snapshots so you only need to dirty a portion of the text??

  2. If your question is how I would provide a better implementation for this code:

    NormalizedSnapshotSpanCollection has some handy static methods to help out: Difference, Intersection, Overlap, Union (  You can use the Difference between the before and after collection to figure out what areas have changed, and send out a change event for those regions.  The code for that would look roughly like this:

    A slightly different way is to compute the difference between the sections and then create a single snapshot span from the start of the first difference to the end of the last difference:

    It’s less minimal (in terms of the total amount of the snapshot that is dirtied), but it may end up being faster, since there is a cost associated with both the total area dirtied and the number of events raised, in general, with taggers.

    However, if your question is just "how do I determine the difference between two snapshots" (as you said; I just wasn’t sure if that is what you meant, sorry), then the answer is that you can get the earlier of the two versions for the snapshots and walk the Changes on each version to figure out what has changed, stepping along the version’s Next until you end up at the newer version.



  3. Cameron Peters says:

    Noah —

    Thanks!  This is awesome and is exactly what I needed.

    Much appreciated.

    ~ Cameron

  4. Cameron Peters says:

    Hey Noah —

    I tried your sample code, but unfortunately it is not working, yet.  The problem appears to be in the NormalizedSpanCollections.Difference helper method.  I wrote the following short test case to illustrate the problem:

           private  void TestDifference()


               SnapshotSpan[] prior = {

                                        new SnapshotSpan(_buffer.CurrentSnapshot, 1, 6),

                                        new SnapshotSpan(_buffer.CurrentSnapshot, 10, 16),


               var priorSections = new NormalizedSnapshotSpanCollection(prior);

               SnapshotSpan[] current = {

                                        new SnapshotSpan(_buffer.CurrentSnapshot, 1, 6),

                                        new SnapshotSpan(_buffer.CurrentSnapshot, 10, 16),

                                        new SnapshotSpan(_buffer.CurrentSnapshot, 20, 26),


               var currentSections = new NormalizedSnapshotSpanCollection(current);

               var difference = NormalizedSnapshotSpanCollection.Difference(priorSections, currentSections);

               Debug.Assert(difference.Count > 0);


    I would expect that give the second group of sections has three spans, and the first has two spans, that the difference would return at least one span.  In my case, it always returns 0 differences.  I am running on VS2010 RC.

    Is this a bug, or am I using these methods improperly?

    Thanks in advance for any help.

    ~ Cameron

  5. Ah, yeah, my bad.  The set stuff isn’t immediately obvious (I screwed it up in my description and code), and the conventional meaning of "difference" doesn’t really help.

    The difference is all the elements in the first set that *aren’t* in the second set.  So {A, B, C} – {B, C} = {A}.  However, {A, B} – {B, C} = {A}, and does *not* include C (as I led you to believe).  Also, I’m using "-" here just for the sake of clarity, but set difference is usually denoted with "∖".

    What you actually want is everything in A that isn’t in B and everything in B that isn’t in A (it’s called the symmetric difference).  Just like that sounds, that is:

    (A ∖ B) ∪ (B ∖ A)

    Here’s the wikipedia page, if you are curious:

    Here’s what that looks like for the outlining tagger:

    This should be a method that does just the symmetric difference (untested, sorry):

    Let me know if that works.



  6. Cameron Peters says:

    Noah —

    Thanks, this is very helpful.  It works and is exactly what I needed.

    ~ Cameron

Skip to main content