Expose parse function to plugins

[unstubbable]

This is motivated by #1512 and basically implements what @Rich-Harris was suggesting a few weeks ago.

By exposing rollup's parse function to plugins (via the transform context) these plugins can use the same acorn configuration (including injected plugins) if they need to operate on an ast. The most prominent example for this is probably rollup-plugin-commonjs, as mentioned in #1512. I opened a PR there as well to switch to the parse method of the transform context.

[lukastaegert]

Very nice! Also love the test! Found one minor comment, otherwise this is good to go!

[lukastaegert]

👍

[lukastaegert]

I think onComment should actually be added last as we are relying internally on this not being overwritten.

[unstubbable]

fixed in cd53741

[guybedford]

Is there a way for this function to share the Rollup cache at all? Otherwise the internal parsing mechanism remains "privileged" in terms of providing this.

[unstubbable]

@guybedford I tried to investigate this. Maybe you can help me to understand the question first.

Are you referring to
a) the cachedModules of a Graph or
b) the ast of a Module or
c) something else entirely?

Depending on the answer I will probably have follow-up questions. 🙂

[unstubbable]

Ahh, I just saw your reference to this issue in #1371. So you are probably talking about b) the ast of a Module.

In this case my follow-up question is this:

Right now the ast is stored in the Module after the source has been transformed by the plugins. So if I understand this correctly the "cache" currently only works in the other direction, i.e. the plugin can return the ast and rollup then does not need to parse again, right? Should this be somehow reversed or what was your idea?

[unstubbable]

Oops, I guess the aforementioned order is irrelevant when you're talking about caching across builds in watch mode. But in this case the cached result will be returned before transform is even called, right? ... Sorry for my rambling, I don't have the full picture yet.

[guybedford]

@KingHenne yes I'm referring to across builds. It sounds like you may be right in that it would only apply if a plugin for file x happens to parse file y's ast when file y has previously been transformed. Also not completely clear on how the cache layer works myself, just trying to work out the repercussions for future caching architectures of this change.

[tunnckoCore]

Wouldn't it be better to expose the AST too? So there will be only one parsing... and plugins can work on it directly.

[unstubbable]

@olstenlarck See discussion above.