WIP: Annotate expressions with their location

[expipiplus1]

Some initial work towards fixing #33

This is far from complete, Some functions need implementing, and there's some tidying up to do in the parser. Annotation.hs could also use a little more organisation.

It typechecks though :)

[jwiegley]

@expipiplus1 Nice! If this is a work in progress, shouldn't I wait until you're done?

[expipiplus1]

Yes! Please don't merge now!

[expipiplus1]

@jwiegley How would you suggest dealing with whitespace in the expression annotations?

It's not possible to eat the whitespace before recording the start of the span as some parsers are whitespace sensitive (the tests fail for select).

There isn't any opportunity to record trailing whitespace, as token eats any trailing whitespace.

The best option I can think of is to postprocess the source spans to remove leading and trailing whitespace.

Another option would be to split the parser and the lexer, but that'd probably be a much larger job.

Another option would be to leave it as is, and let the user deal with it :)

[jwiegley]

For now we can let the user deal with it, but I like the idea of splitting parser and lexer.

[expipiplus1]

The annoying thing is that I'm the user at the moment :)

If I find a tidy way of stripping whitespace I think I'll do that.

[expipiplus1]

Another possibility would be to modify the definition of whitespace so that it maintains some kind of beforeWhitespacePos + afterWhitespacePos state. I'm not sure if this is possible in a Trifecta parser though. Does this idea makes sense?

[expipiplus1]

This is working well enough now for me to have made this little utility to automatically update fetchgit values in nix expressions to the latest versions.

https://github.com/expipiplus1/update-nix-fetchgit

[expipiplus1]

@jwiegley aside from some tests and documentation, is there anything major preventing this from being merged