Schema parser & printer

[jhgg]

PR for issue #11

• Adds the schema type definitions to ast.ast to have generate_ast() generate classes for them.
• parse_type_definition will now parse schema types.
• Implement parsers for each schema type.
• Improve error message in Executor.complete_value when runtime type doesn't match.
• Implement more missing executor tests.
• The visit function will no longer attempt to deep copy nodes when there is an edit. Shallow copies are perfectly acceptable, since we are only overwriting attributes on edit.
• The Visitor class now pre-calculates an AST Type to Handler map in the metaclass, allowing for a much quicker enter/leave implementation.
• Install six.
• Use VisitorMeta to power TypeInfo so that it's not a huge isinstance chain.

Checklist to merge:

• Implement remaining schema parser tests.
• Implement schema printer and tests.

[syrusakbary]

Why use this renamed functions instead of calling directly?

[jhgg]

It's a slight performance gain. In a tight loop, list.pop() vs list_pop, the latter will be faster.

def foo():
    l = []
    while True:
        l.append(1)

The disassembly produces:

  2           0 BUILD_LIST               0
              3 STORE_FAST               0 (l)

  3           6 SETUP_LOOP              23 (to 32)
        >>    9 LOAD_GLOBAL              0 (True)
             12 POP_JUMP_IF_FALSE       31

  4          15 LOAD_FAST                0 (l)
             18 LOAD_ATTR                1 (append)
             21 LOAD_CONST               1 (1)
             24 CALL_FUNCTION            1
             27 POP_TOP
             28 JUMP_ABSOLUTE            9
        >>   31 POP_BLOCK
        >>   32 LOAD_CONST               0 (None)
             35 RETURN_VALUE

As for:

def foo():
    l = []
    append = l.append
    while True:
        append(1)

  2           0 BUILD_LIST               0
              3 STORE_FAST               0 (l)

  3           6 LOAD_FAST                0 (l)
              9 LOAD_ATTR                0 (append)
             12 STORE_FAST               1 (append)

  4          15 SETUP_LOOP              20 (to 38)
        >>   18 LOAD_GLOBAL              1 (True)
             21 POP_JUMP_IF_FALSE       37

  5          24 LOAD_FAST                1 (append)
             27 LOAD_CONST               1 (1)
             30 CALL_FUNCTION            1
             33 POP_TOP
             34 JUMP_ABSOLUTE           18
        >>   37 POP_BLOCK
        >>   38 LOAD_CONST               0 (None)
             41 RETURN_VALUE

Notice that in the loop body, LOAD_FAST and then LOAD_ATTR has to be called. Whereas, LOAD_FAST is only called in the second version.

It's a small performance gain. For more, check out this article: https://pymotw.com/2/dis/

[syrusakbary]

Great response!

[syrusakbary]

Use node_pop perhaps?

[jhgg]

In this case, it's not worth it. node is always changing. Also REMOVE is very rarely returned from a visitor.