Conrad Buck

I think I misunderstood this when I first read it. I see now that we're in line.

Why not just introduce a flag called, say, `--no-initial-snapshots` that refuses to create initial snapshots? Then tests which would otherwise result in the creation of a new snapshot would instead...

I don't understand. You're saying I should reinvent *prettier*? What I am inventing is a tool that would make it possible for prettier to support this.

In particular what I haven't built yet but could offer is a `tokensLookahead(node)` iterator and a `tokensLookbehind(node)` iterator. This is what would make it easy for prettier to answer queries...

Oh yeah. I don't think the memory costs of tokens should be too bad. There's a lot of places in this code that I could cache the token objects. For...

It looks like a pretty big/complex API but I could try implementing some parts of it over the CST. I should also definitely change `node.tokens` to something like `node[Symbol.for('cst.tokens')]`. Another...

@alexander-akait What I'm talking about isn't a pure CST, it's a CST purely inside a normal AST. How you know about your parents and siblings is no different than it...

I want the behavior because I believe there should be first-class support for writing codemods. Your code is data, and it should be as easy as possible to write change...

@fisker Why do you definitely not want tokens on node?

By first-class support I really just mean that there should be a `formatAst(ast)` or a `formatCst(cst)` method. It does seem there's interest in that right? I think it actually makes...