Skip to content

James-Ansley/codist

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CoDist

CoDist (Code Distance) is a library that provides functions to calculate the edit distance and edit paths of abstract syntax trees.

While this library is primarily concerned with AST edit distances, it can handle any generic tree of the form: Tree[T] = tuple[T, tuple[Tree[T], ...]]

Install

pip install codist

Usage

Tree Edit Distance

Trees are represented as tuples: Tree[T] = tuple[T, tuple[Tree[T], ...]].

For example, the following two trees from the original Zhang Shasha paper:

Can be represented with the following tuples:

tree1 = ("f", (("d", (("a", ()), ("c", (("b", ()),)))), ("e", ())))
tree2 = ("f", (("c", (("d", (("a", ()), ("b", ())),),),), ("e", ())))

A small helper function, t, has been provided to make tree construction less verbose:

from codist import t

tree1 = t("f", t("d", t("a"), t("c", t("b"))), t("e"))
tree2 = t("f", t("c", t("d", t("a"), t("b"))), t("e"))

The distance between these two trees can be taken with the tree_dist function:

from codist import tree_dist

dist = tree_dist(tree1, tree2)
print("The distance is:", dist)  # The distance is 2

A custom set of cost functions can be provided with a Cost object:

from codist import Cost, tree_dist

cost = Cost(
    delete=lambda n: 3,
    insert=lambda n: 3,
    relabel=lambda n1, n2: 0 if n1 == n2 else 2,
)

dist = tree_dist(tree1, tree2, cost=cost)
print("The distance is:", dist)  # The distance is 6

By default, all change operations have a cost of 1 except for the case of γ(a -> a) which is 0.

The edit path can be obtained with the tree_edit function which returns the tree distance and a tuple of change operations:

from codist import tree_edit

dist, path = tree_edit(tree1, tree2)
path = tuple(c for c in path if c[0] != c[1])
print("The distance is:", dist)  # The distance is 2

# The changes are: (('c', 'Λ', 2), ('Λ', 'c', 5))
print("The changes are:", path)

Change operations are 3-tuples of the form: tuple[T | Lambda, T | Lambda, int | Lambda] where Lambda is a singleton defined in the tree module.

The third element of the tuple is a postorder index that provides context for the change operation. For insertion operations, the context index is the index of the parent node in Tree2 under which the node T is being inserted. For deletion and relabel operations, the context is the index of the node in Tree1 that is being deleted or relabelled.

In the above example, ('c', 'Λ', 2) indicates a deletion change operation, where the node at index 2 of tree1, "c" is being deleted. ('Λ', 'c', 5) indicates a new node labelled "c" is being inserted under the node at index 5 of tree2, "f".

The context index does not contain all information about the changes, for example, for insertion changes, the context index provides no information about which siblings of the parent node are pulled down as children of the inserted node.

The tree_edit function can also take a cost object.

AST Edit Distance

Currently, only AST node type information is compared. A silhouette of an AST (an AST containing only type information) is constructed with the parse_ast_silhouette function.

from pprint import pprint
from codist.ast import parse_ast_silhouette

code1 = """
def process(data):
    result = []
    for x in data:
        if x > 5:
            result.append(x)
    return result
"""

code2 = """
def process(data):
    result = []
    for x in data:
        if x >= 6:
            result += [x]
    return result
"""

ast1 = parse_ast_silhouette(code1)
ast2 = parse_ast_silhouette(code2)

pprint(ast1)
pprint(ast2)

Which prints:

('Module',
 (('FunctionDef',
   (('arguments', (('arg', ()),)),
    ('Assign', (('Name', (('Store', ()),)), ('List', (('Load', ()),)))),
    ('For',
     (('Name', (('Store', ()),)),
      ('Name', (('Load', ()),)),
      ('If',
       (('Compare', (('Name', (('Load', ()),)), ('Gt', ()), ('Constant', ()))),
        ('Expr',
         (('Call',
           (('Attribute', (('Name', (('Load', ()),)), ('Load', ()))),
            ('Name', (('Load', ()),)))),)))))),
    ('Return', (('Name', (('Load', ()),)),)))),))
('Module',
 (('FunctionDef',
   (('arguments', (('arg', ()),)),
    ('Assign', (('Name', (('Store', ()),)), ('List', (('Load', ()),)))),
    ('For',
     (('Name', (('Store', ()),)),
      ('Name', (('Load', ()),)),
      ('If',
       (('Compare', (('Name', (('Load', ()),)), ('GtE', ()), ('Constant', ()))),
        ('AugAssign',
         (('Name', (('Store', ()),)),
          ('Add', ()),
          ('List', (('Name', (('Load', ()),)), ('Load', ()))))))))),
    ('Return', (('Name', (('Load', ()),)),)))),))

The distance between these ASTs can be taken as above.

About

Tree edit distances using the Zhang-Shasha algorithm

Topics

Resources

License

Stars

Watchers

Forks

Languages