But yes, having a syntax tree would solve this problem quite nicely.
(But to make it work universally you have to decide on a representation of this syntax tree to be able to represent it in plain text files to make it work with all the editors and IDEs that don't support that syntax tree stuff... And then we're back to .editorconfig, aren't we?)
A VCS for syntax trees means you need the One True Data Format(tm) for syntax trees in any language. Finding consensus on formatting per project is easier than unifying all syntax tree representations globally.
(Aside: I built a cool proof of concept for what I think is the most useful "smart middle ground" here, which is to use syntax highlighting lexers (tokenizers) for a useful effect in creating smart character-based diffs. Syntax highlighting lexers are designed for intermediate code and with output as ordinary character diffs you don't have to bikeshed a universal token stream format. The tokenized diffs were cleaner and faster than the equivalent non-tokenized character diff. https://github.com/WorldMaker/tokdiff if anyone wants to play with it.)
http://news.squeak.org/2008/08/15/all-new-monticello-2/
I had no idea there was such a new rewrite (2008), but then I keep forgetting that DVCS are rather new (as in popular, free-software DVCS are pretty new - git is just over 10 years).
[1] http://www.colorforth.com/cf.htm [It would appear that the on-disk representation is rather simple and straight forward, but you're only supposed to use a ColorForth aware editor (eg: ColorForth)]
Are you saying git is not popular, is not free software, or something else?
PS: Whooop on the Windows trick! So useful.