chapters/49.markdown @ c639d2c11892 typo-the-then

Close branch.
author Steve Losh <steve@stevelosh.com>
date Fri, 07 Oct 2016 13:11:30 +0000
parents fcbfb7d92770
children (none)
Advanced Folding
================

In the last chapter we used Vim's `indent` folding to add some quick and dirty
folding to Potion files.

Open `factorial.pn` and make sure all the folds are closed with `zM`.  The file
should now look something like this:

    :::text
    factorial = (n):
    +--  5 lines: total = 1

    10 times (i):
    +--  4 lines: i string print

Toggle the first fold and it will look like this:

    :::text
    factorial = (n):
        total = 1
        n to 1 (i):
    +---  2 lines: # Multiply the running total.
        total.

    10 times (i):
    +--  4 lines: i string print

This is pretty nice, but I personally prefer to fold the first line of a block
with its contents.  In this chapter we'll write some custom folding code, and
when we're done our folds will look like this:

    :::text
    factorial = (n):
        total = 1
    +---  3 lines: n to 1 (i):
        total.

    +--  5 lines: 10 times (i):

This is more compact and (to me) easier to read.  If you prefer the `indent`
method that's okay, but do this chapter anyway just to get some practice writing
Vim folding expressions.

Folding Theory
--------------

When writing custom folding code it helps to have an idea of how Vim "thinks" of
folding.  Here are the rules in a nutshell:

* Each line of code in a file has a "foldlevel".  This is always either zero or
  a positive integer.
* Lines with a foldlevel of zero are *never* included in any fold.
* Adjacent lines with the same foldlevel are folded together.
* If a fold of level X is closed, any subsequent lines with a foldlevel greater
  than or equal to X are folded along with it until you reach a line with
  a level less than X.

It's easiest to get a feel for this with an example.  Open a Vim window and
paste the following text into it.

    :::text
    a
        b
        c
            d
            e
        f
    g

Turn on `indent` folding by running the following command:

    :::vim
    :setlocal foldmethod=indent

Play around with the folds for a minute to see how they behave.

Now run the following command to view the foldlevel of line 1:

    :::vim
    :echom foldlevel(1)

Vim will display `0`.  Now let's find the foldlevel of line 2:

    :::vim
    :echom foldlevel(2)

Vim will display `1`.  Let's try line 3:

    :::vim
    :echom foldlevel(3)

Once again Vim displays `1`.  This means that lines 2 and 3 are part of a level
1 fold.

Here are the foldlevels for each line:

    :::text
    a           0
        b       1
        c       1
            d   2
            e   2
        f       1
    g           0

Reread the rules at the beginning of this section.  Open and close each fold in
this file, look at the foldlevels, and make sure you understand why the folds
behave as they do.

Once you're confident that you understand how every line's foldlevel works to
create the folding structure, move on to the next section.


First: Make a Plan
------------------

Before we dive into writing code, let's try to sketch out some rough "rules" for
our folding.

First, lines that are indented should be folded together.  We also want the
*previous* line folded with them, so that something like this:

    :::text
    hello = (name):
        'Hello, ' print
        name print.

Will fold like this:

    :::text
    +--  3 lines: hello = (name):

Blank lines should be at the same level as *later* lines, so blank lines at the
end of a fold won't be included in it.  This means that this:

    :::text
    hello = (name):
        'Hello, ' print
        name print.

    hello('Steve')

Will fold like this:

    :::text
    +--  3 lines: hello = ():

    hello('Steve')

And *not* like this:

    :::text
    +--  4 lines: hello = ():
    hello('Steve')

These rules are a matter of personal preference, but for now this is the way
we're going to implement folding.

Getting Started
---------------

Let's get started on our custom folding code by opening Vim with two splits.
One should contain our `ftplugin/potion/folding.vim` file, and the other should
contain our sample `factorial.pn`.

In the previous chapter we closed and reopened Vim to make our changes to
`folding.vim` take effect, but it turns out there's an easier way to do that.

Remember that any files inside `ftplugin/potion/` will be run whenever the
`filetype` of a buffer is set to `potion`.  This means you can simply run `:set
ft=potion` in the split containing `factorial.pn` and Vim will reload the
folding code!

This is much faster than closing and reopening the file every time.  The only
thing you need to remember is that you have to *save* `folding.vim` to disk,
otherwise your unsaved changes won't be taken into account.

Expr Folding
------------

We're going to use Vim's `expr` folding to give us unlimited flexibility in how
our code is folded.

We can go ahead and remove the `foldignore` from `folding.vim` because it's only
relevant when using `indent` folding.  We also want to tell Vim to use `expr`
folding, so change the contents of `folding.vim` to look like this:

    :::vim
    setlocal foldmethod=expr
    setlocal foldexpr=GetPotionFold(v:lnum)

    function! GetPotionFold(lnum)
        return '0'
    endfunction

The first line simply tells Vim to use `expr` folding.

The second line defines the expression Vim should use to get the foldlevel of
a line.  When Vim runs the expression it will set `v:lnum` to the line number of
the line it wants to know about.  Our expression will call a custom function
with this number as an argument.

Finally we define a dummy function that simply returns `'0'` for every line.
Note that it's returning a String and not an Integer.  We'll see why shortly.

Go ahead and reload the folding code by saving `folding.vim` and running `:set
ft=potion` in `factorial.pn`.  Our function returns `'0'` for every line, so
Vim won't fold anything at all.

Blank Lines
-----------

Let's take care of the special case of blank lines first.  Modify the
`GetPotionFold` function to look like this:

    :::vim
    function! GetPotionFold(lnum)
        if getline(a:lnum) =~? '\v^\s*$'
            return '-1'
        endif

        return '0'
    endfunction

We've added an `if` statement to take care of the blank lines.  How does it
work?

First we use `getline(a:lnum)` to get the content of the current line as
a String.

We compare this to the regex `\v^\s*$`.  Remember that `\v` turns on "very
magic" ("sane") mode.  This regex will match "beginning of line, any number
of whitespace characters, end of line".

The comparison is using the case-insensitive match operator `=~?`.  Technically
we don't have to be worried about case since we're only matching whitespace, but
I prefer to be more explicit when using comparison operators on Strings.  You
can use `=~` instead if you prefer.

If you need a refresher on using regular expressions in Vim you should go back
and reread the "Basic Regular Expressions" chapter and the chapters on the "Grep
Operator".

If the current line has some non-whitespace characters it won't match and we'll
just return `'0'` as before.

If the current line *does* match the regex (i.e. if it's empty or just
whitespace) we return the string `'-1'`.

Earlier I said that a line's foldlevel can be zero or a positive integer, so
what's happening here?

Special Foldlevels
------------------

Your custom folding expression can return a foldlevel directly, or return one of
a few "special" strings that tell Vim how to fold the line without directly
specifying its level.

`'-1'` is one of these special strings.  It tells Vim that the level of this
line is "undefined".  Vim will interpret this as "the foldlevel of this line is
equal to the foldlevel of the line above or below it, whichever is smaller".

This isn't *exactly* what our plan called for, but we'll see that it's close
enough and will do what we want.

Vim can "chain" these undefined lines together, so if you have two in a row
followed by a line at level 1, it will set the last undefined line to 1, then
the next to last to 1, then the first to 1.

When writing custom folding code you'll often find a few types of line that you
can easily set a specific level for.  Then you'll use `'-1'` (and some other
special foldlevels we'll see soon) to "cascade" the proper folding levels to the
rest of the file.

If you reload the folding code for `factorial.pn` Vim *still* won't fold any
lines together.  This is because all the lines have a foldlevel of either zero
or "undefined".  The level `0` will "cascade" through the undefined lines and
eventually all the lines will have their foldlevel set to `0`.

An Indentation Level Helper
---------------------------

To tackle non-blank lines we'll need to know their indentation level, so let's
create a small helper function to calculate it for us.  Add the following
function above `GetPotionFold`:

    :::vim
    function! IndentLevel(lnum)
        return indent(a:lnum) / &shiftwidth
    endfunction

Reload the folding code.  Test out your function by running the following
command in the `factorial.pn` buffer:

    :::vim
    :echom IndentLevel(1)

Vim displays `0` because line 1 is not indented.  Now try it on line 2:

    :::vim
    :echom IndentLevel(2)

This time Vim displays `1`.  Line two has 4 spaces at the beginning, and
`shiftwidth` is set to 4, so 4 divided by 4 is 1.

`IndentLevel` is fairly straightforward.  The `indent(a:lnum)` returns the
number of spaces at the beginning of the given line number.  We divide that by
the `shiftwidth` of the buffer to get the indentation level.

Why did we use `&shiftwidth` instead of just dividing by 4?  If someone prefers
two-space indentation in their Potion files, dividing by 4 would produce an
incorrect result.  We use the `shiftwidth` setting to allow for any number of
spaces per level.

One More Helper
---------------

It might not be obvious where to go from here.  Let's stop and think about what
type of information we need to have to figure out how to fold a non-blank line.

We need to know the indentation level of the line itself.  We've got that
covered with the `IndentLevel` function, so we're all set there.

We'll also need to know the indentation level of the *next non-blank line*,
because we want to fold the "header" lines with their indented bodies.

Let's write a helper function to get the number of the next non-blank line after
a given line.  Add the following function above `IndentLevel`:

    :::vim
    function! NextNonBlankLine(lnum)
        let numlines = line('$')
        let current = a:lnum + 1

        while current <= numlines
            if getline(current) =~? '\v\S'
                return current
            endif

            let current += 1
        endwhile

        return -2
    endfunction

This function is a bit longer, but is pretty simple.  Let's take it
piece-by-piece.

First we store the total number of lines in the file with `line('$')`.  Check
out the documentation for `line()` to see how this works.

Next we set the variable `current` to the number of the next line.

We then start a loop that will walk through each line in the file.

If the line matches the regex `\v\S`, which means "match a character that's
*not* a whitespace character", then it must be non-blank, so we should return
its line number.

If the line doesn't match, we loop around to the next one.

If the loop gets all the way to the end of the file without ever returning, then
there are *no* non-blank lines after the current line!  We return `-2` if that
happens to indicate this.  `-2` isn't a valid line number, so it's an easy way
to say "sorry, there's no valid result".

We could have returned `-1`, because that's not a valid line number either.
I could have even picked `0`, since line numbers in Vim start at `1`!  So why
did I pick `-2`, which seems like a strange choice?

I chose `-2` because we're working with folding code, and `'-1'` (and `'0'`) is
a special Vim foldlevel string.

When my eyes are reading over this file and I see a `-1` my brain immediately
thinks "undefined foldlevel".  The same is true with `0`.  I picked `-2` here
simply to make it obvious that it's *not* a foldlevel, but is instead an
"error".

If this feels weird to you, you can safely change the `-2` to a `-1` or a `0`.
It's just a coding style preference.

Finishing the Fold Function
---------------------------

This is turning out to be quite a long chapter, so let's wrap up the folding
function.  Change `GetPotionFold` to look like this:

    :::vim
    function! GetPotionFold(lnum)
        if getline(a:lnum) =~? '\v^\s*$'
            return '-1'
        endif

        let this_indent = IndentLevel(a:lnum)
        let next_indent = IndentLevel(NextNonBlankLine(a:lnum))

        if next_indent == this_indent
            return this_indent
        elseif next_indent < this_indent
            return this_indent
        elseif next_indent > this_indent
            return '>' . next_indent
        endif
    endfunction

That's a lot of new code!  Let's step through it to see how it all works.

### Blanks

First we have our check for blank lines.  Nothing's changed there.

If we get past that check we know we're looking at a non-blank line.

### Finding Indentation Levels

Next we use our two helper functions to get the indent level of the current
line, and the indent level of the next non-blank line.

You might wonder what happens if `NextNonBlankLine` returns our error condition
of `-2`.  If that happens, `indent(-2)` will be run.  Running `indent()` on
a nonexistent line number will just return `-1`.  Go ahead and try it yourself
with `:echom indent(-2)`.

`-1` divided by any `shiftwidth` larger than `1` will return `0`.  This may seem
like a problem, but it turns out that it won't be.  For now, don't worry about
it.

### Equal Indents

Now that we have the indentation levels of the current line and the next
non-blank line, we can compare them and decide how to fold the current line.

Here's the `if` statement again:

    :::vim
    if next_indent == this_indent
        return this_indent
    elseif next_indent < this_indent
        return this_indent
    elseif next_indent > this_indent
        return '>' . next_indent
    endif

First we check if the two lines have the same indentation level.  If they do, we
simply return that indentation level as the foldlevel!

An example of this would be:

    :::text
    a
    b
        c
        d
    e

If we're looking at the line containing "c", it has an indentation level of 1.
This is the same as the level of the next non-blank line ("d"), so we return `1`
as the foldlevel.

If we're looking at "a", it has an indentation level of 0.  This is the same as
the level of the next non-blank line ("b"), so we return `0` as the foldlevel.

This case fills in two foldlevels in this simple example:

    :::text
    a       0
    b       ?
        c   1
        d   ?
    e       ?

By pure luck this also handles the special "error" case of the last line as
well!  Remember we said that `next_indent` will be `0` if our helper function
returns `-2`.

In this example the line "e" has an indent level of `0`, and `next_indent` will
also be set to `0`, so this case matches and returns `0`.  The foldlevels now
look like this:

    :::text
    a       0
    b       ?
        c   1
        d   ?
    e       0

### Lesser Indent Levels

Once again, here's the `if` statement:

    :::vim
    if next_indent == this_indent
        return this_indent
    elseif next_indent < this_indent
        return this_indent
    elseif next_indent > this_indent
        return '>' . next_indent
    endif

The second part of the `if` checks if the indentation level of the next line is
*smaller* than the current line.  This would be like line "d" in our example.

If that's the case, we once again return the indentation level of the current
line.

Now our example looks like this:

    :::text
    a       0
    b       ?
        c   1
        d   1
    e       0

You could, of course, combine these two cases with `||`, but I prefer to keep
them separate to make it more explicit.  You might feel differently.  It's
a style issue.

Again, purely by luck, this case handles the other possible "error" case of our
helper function.  Imagine that we have a file like this:

    :::text
    a
        b
        c

The first case takes care of line "b":

    :::text
    a       ?
        b   1
        c   ?

Line "c" is the last line, and it has an indentation level of 1.  The
`next_indent` will be set to `0` thanks to our helper functions.  The second
part of the `if` matches and sets the foldlevel to the current indentation
level, or `1`:

    :::text
    a       ?
        b   1
        c   1

This works out great, because "b" and "c" will be folded together.

### Greater Indentation Levels

Here's that tricky `if` statement for the last time:

    :::vim
    if next_indent == this_indent
        return this_indent
    elseif next_indent < this_indent
        return this_indent
    elseif next_indent > this_indent
        return '>' . next_indent
    endif

And our example file:

    :::text
    a       0
    b       ?
        c   1
        d   1
    e       0

The only line we haven't figured out is "b", because:

* "b" has an indent level of `0`.
* "c" has an indent level of `1`.
* 1 is not equal to 0, nor is 1 less than 0.

The last case checks if the next line has a *larger* indentation level than the
current one.

This is the case that Vim's `indent` folding gets wrong, and it's the entire
reason we're writing this custom folding in the first place!

The final case says that when the next line is indented more than the current
one, it should return a string of a `>` character and the indentation level of
the *next* line.  What the heck is *that*?

Returning a string like `>1` from the fold expression is another one of Vim's
"special" foldlevels.  It tells Vim that the current line should *open* a fold
of the given level.

In this simple example we could have just returned the number, but we'll see
why this is important shortly.

In this case line "b" will open a fold at level 1, which makes our example look
like this:

    :::text
    a       0
    b       >1
        c   1
        d   1
    e       0

That's exactly what we want!  Hooray!

Review
------

If you've made it this far you should feel proud of yourself.  Even simple
folding code like this can be tricky and mind bending.

Before we end, let's go through our original `factorial.pn` code and see how our
folding expression fills in the foldlevels of its lines.

Here's `factorial.pn` for reference:

    :::text
    factorial = (n):
        total = 1
        n to 1 (i):
            # Multiply the running total.
            total *= i.
        total.

    10 times (i):
        i string print
        '! is: ' print
        factorial (i) string print
        "\n" print.

First, any blank lines' foldlevels will be set to undefined:

    :::text
    factorial = (n):
        total = 1
        n to 1 (i):
            # Multiply the running total.
            total *= i.
        total.
                                             undefined
    10 times (i):
        i string print
        '! is: ' print
        factorial (i) string print
        "\n" print.

Any lines where the next line's indentation is *equal* to its own are set to its
own level:

    :::text
    factorial = (n):
        total = 1                            1
        n to 1 (i):
            # Multiply the running total.    2
            total *= i.
        total.
                                             undefined
    10 times (i):
        i string print                       1
        '! is: ' print                       1
        factorial (i) string print           1
        "\n" print.

The same thing happens when the next line's indentation is *less* than the
current line's: 

    :::text
    factorial = (n):
        total = 1                            1
        n to 1 (i):
            # Multiply the running total.    2
            total *= i.                      2
        total.                               1
                                             undefined
    10 times (i):
        i string print                       1
        '! is: ' print                       1
        factorial (i) string print           1
        "\n" print.                          1

The last case is when the next line's indentation is *greater* than the current
line's.  When that happens the line's foldlevel is set to *open* a fold of the
*next* line's foldlevel:

    :::text
    factorial = (n):                         >1
        total = 1                            1
        n to 1 (i):                          >2
            # Multiply the running total.    2
            total *= i.                      2
        total.                               1
                                             undefined
    10 times (i):                            >1
        i string print                       1
        '! is: ' print                       1
        factorial (i) string print           1
        "\n" print.                          1

Now we've got a foldlevel for every line in the file.  All that's left is for
Vim to resolve any undefined lines.

Earlier I said that undefined lines will take on the smallest foldlevel of
either of their neighbors.

That's how Vim's manual describes it, but it's not entirely accurate.  If that
were the case, the blank line in our file would take foldlevel 1, because both
of its neighbors have a foldlevel of 1.

In reality, the blank line will be given a foldlevel of 0!

The reason for this is that we didn't just set the `10 times (i):` line to
foldlevel `1` directly.  We told Vim that the line *opens* a fold of level `1`.
Vim is smart enough to know that this means the undefined line should be set to
`0` instead of `1`.

The exact logic of this is probably buried deep within Vim's source code.  In
general Vim behaves pretty intelligently when resolving undefined lines against
"special" foldlevels, so it will usually do what you want.

Once Vim's resolved the undefined line it has a complete description of how to
fold each line in the file, which looks like this:

    :::text
    factorial = (n):                         1
        total = 1                            1
        n to 1 (i):                          2
            # Multiply the running total.    2
            total *= i.                      2
        total.                               1
                                             0
    10 times (i):                            1
        i string print                       1
        '! is: ' print                       1
        factorial (i) string print           1
        "\n" print.                          1

That's it, we're done!  Reload the folding code and play around with the fancy
new folding in `factorial.pn`.

Exercises
---------

Read `:help foldexpr`.

Read `:help fold-expr`.  Pay particular attention to all the "special" strings
your expression can return.

Read `:help getline`.

Read `:help indent()`.

Read `:help line()`.

Figure out why it's important that we use `.` to combine the `>` character with
the number in our folding function.  What would happen if you used `+` instead?
Why?

We defined our helper functions as global functions, but that's not a good idea.
Change them to be script-local functions.

Put this book down and go outside for a while to let your brain recover from
this chapter.