Crosscut

The new comments on branches required changes to the syntax. I want them to be associated with the branch in the syntax tree, so they can be used to generate documentation, for example. Or, as I've already done, to add metadata to branches for use in tests. Why did this require changes to the syntax? Consider this example:

my_function: fn
    \ 0 ->
        0
        # This comment is part of the body of the first branch.

    \ 1 ->
        1
end

Versus this:

my_function: fn
    \ 0 ->
        0

    # This comment applies to the second branch.
    \ 1 ->
        1
end

The parser ignores whitespace, and it doesn't look ahead beyond the next token it's going to process (the grammar is LL(1)). So it can't distinguish between those two cases. To fix that, I required that branches end with end, like functions:

my_function: fn
    \ 0 ->
        0
        # Now this comment can be clearly distinguished...
    end

    # ...from this other comment.
    \ 1 ->
        1
    end
end

I thought that having the end there made it look inconsistent, that branches are started using an \ token. So I replaced that with the new br keyword:

my_function: fn
    br 0 ->
        0
    end

    br 1 ->
        1
    end
end

The new syntax still looks weird to me, but that's fine. I've come to realize more and more, that everything I do syntax-wise before the code database exists, is temporary at best. But that's a topic for another day.