New Parser

[hit9]

Main changes: (All changes are parser part.)

1. include_dirs to include_dir, parser will try to find all child thrifts files inside this directory during a parsing progress. (Non-Backward-Compatible)
2. Fix referenced types and values, the features were not supported or not well supported in old parser. ( Add support for default enum values that reference the original enum. #69 "const map<EnumType, ?>" fails to parse #75)
3. Ply will parse thrift to python module directly, skipping the original to-json-data-parsing.
4. All constants and ttypes (structs/unions/exceptions/..) are using the same one namespace now, each will be parsed as a member of the parsed thrift module.
5. Constants (or as default field value) can reference to constants or named enum values now.
6. The new parser is more strict on types. Definitions like const i32 i = 1.5 will be rejected now.
7. Including child thrift files perfects well now, all child thrifts will be parsed during a single parsing progress. (via a thrift stack)
8. Strict fields checking on defining a struct instance. struct Person {1: required string name}; const Person tom = {} will be rejected now.
9. Namespaces are ignored.
10. Same with old load, we can enable cache on parsing (default enabled)
11. Cycling including checking (or dead including checking). Inluding case like a->b->c->a will be rejected.

The changes above contains 1 non-backward-compat change.

The parser makes thriftpy behaves closer to apache thrift.

@maralla will merge this pull request.

[hit9]

Dont merge now, I will add some test cases.

[tmehlinger]

These are excellent changes. Really looking forward to having this mainlined. 😄

[tmehlinger]

This should test for basestring on Python 2.6/2.7, else it will fail on unicode strings.

[hit9]

thriftpy always read thrift files as strings, not unicode strings.

[hit9]

Can be merged now.

Better to upgrade thriftpy after merge this pr.

[lxyu]

Excellent! 😄