Refactor blocks, programs and inverses

[mmun]

I've untangled the parsing of programs and inverses in blocks. This simplifies the grammar and reduces the amount of parsing logic that has been pushed into the AST Node constructors. Namely, ProgramNode no longer is concerned with inverses and BlockNode no longer requires passing in the closing token.

In order to simplify the grammar further I also added an INVERSE token to avoid conflicts with the openInverse production. See https://github.com/wycats/handlebars.js/pull/834/files#diff-6.

[kpdecker]

This conflicts with the feature PR for standalone template handling, which I did not want to block on getting this merged. If you want to refactor against that we can discuss as I think there are some helpful things in here. Otherwise I'll try to cherry pick the changes.

A few notes if you do refactor:

• Creating the helpers module seems like unnecessary overhead (Not minimizable, etc)
• Same for the base.js changes to expose this AST/helper code to the parser
• The parser logic can be very tempermental to changes to the rules, both in terms of conflicts and the size of the generated code. It seems like you've worked through the first as it won't compile with conflicts, but will want to compare the sizes after rebase to see what the impact is.

[mmun]

Thanks for taking the time to review this PR. I don't mind resolving the conflicts. I understand that compatibility with mustache is important. I hope you agree that a stable AST is also important for the consumers of the parser / compiler, e.g. HTMLBars, Emblem.js.

The motivation behind my PR is that parsing logic should not belong in the AST. Consumers of the AST shouldn't have to jump through hoops to instantiate AST nodes.

For example, BlockNodes requires their closing token to perform white stripping and for validation. This is analogous to the browser asking you to write document.createElement('div', '</div>'). The stripping information should also not be maintained in the AST, but instead performed before the nodes are created.

As another example, ProgramNode now has non-local tracking information like isRoot which complicates moving ProgramNodes around in the AST. isRoot could be trivially tracked in the compiler.

I would be happy to make these changes in a sequence of small PRs.

To address your specific concerns, the reason I pulled out the helpers (admittedly, I could have left stripFlags) is because they make reference to other classes. For example prepareBlock references Exception and AST (link). The methods defined in the grammar preamble all have global scope and in particular do not have access to the parser or even the free variable yy. This is why I tacked the prepareBlock function onto yy. Here are the file sizes before and after:

	Before	After
handlebars.min.js	48995	48781
handlebars.amd.min.js	49772	49631
handlebars.runtime.min.js	8224	8224
handlebars.runtime.amd.min.js	8522	8522

You're right that I hit conflicts when rewriting the grammar which is why I introduced the INVERSE token. Frankly, the grammar rewrite is mostly cosmetic but I think its easier to reason about this way.

[kpdecker]

I've merged this with master. Thanks for the PR!

Note that I'm going to move further in this direction and my hope is that all stripping operations will be embedded in the content nodes rather than the parent nodes.

[mmun]

@kpdecker Wow!!! Thanks a lot, this looks great!!!

What do you think about having all the content nodes pre-stripped by the helpers? In particular, this would mean that a completed AST would not require strip annotations, and different compilers would not have to reimplement their own stripping.

[kpdecker]

Already done :) 84d646b

[kpdecker]

The strip tokens couldn't go away completely as there was some state needed, but the parser is now updating content nodes for the various stripping needs. The nodes are still left intact though and the value can be accessed through original should a particular version of the compiler find a need for that.

[kpdecker]

Released in v2.0.0-beta.1