asts 0.2.2

A library for programmatic software modification

• Generic Tree-Sitter AST API
• Quick Intro
• Extended Tutorial
  • AST Creation
    • Constructor
    • AST Templates
    • AST Copy
  • AST Methods
    • Common Operations
    • Source Locations
    • Functions
    • Function Callsites
  • AST Traversal
  • AST Manipulation
    • Mutation Primitives
    • Transformers
• Architecture
• License

Generic Tree-Sitter AST API

The ASTs package provides a Python API into GrammaTech's Software Evolution Library (SEL) for source code manipulation. SEL generalizes over GitHub's tree-sitter parsing libraries providing a uniform interface over multiple programming languages (primarily Python, JavaScript, and C/C++), and providing additional functionality for software inspection and modification.

Quick Intro

Here's how to create and perform some common operations on an AST:

$ python3
Python 3.8.5
Type "help", "copyright", "credits" or "license" for more information.
>>> import asts
>>> root = asts.AST("x + 88", language=asts.ASTLanguage.Python)
>>> root.children()
[<PYTHON-EXPRESSION-STATEMENT-0 0x3b2>]
>>> root.children()[0].children()
[<PYTHON-BINARY-OPERATOR 0x3b3>]
>>> root.children()[0].children()[0].children()
[<PYTHON-IDENTIFIER 0x3b4>, <PYTHON-+ 0x3b5>, <PYTHON-INTEGER 0x3b6>]
>>> root.children()[0].children()[0].children()[0].source_text()
'x'
>>> root.children()[0].children()[0].children()[1].source_text()
'+'
>>> root.children()[0].children()[0].children()[2].source_text()
'88'
>>> root.children()[0].children()[0].source_text()
'x + 88'
>>> root.children()[0].children()[0].child_slots()
[['BEFORE-ASTS', 0], ['PYTHON-LEFT', 1], ['PYTHON-OPERATOR', 1],
 ['PYTHON-RIGHT', 1], ['CHILDREN', 0], ['AFTER-ASTS', 0]]
>>> list(map(lambda x:x.source_text(), root.children()[0].children()[0].children()))
['x', '+', '88']
>>> list(map(lambda x:x.ast_type(), root.children()[0].children()[0].children()))
['PYTHON-IDENTIFIER', 'PYTHON-+', 'PYTHON-INTEGER']

Extended Tutorial

The following examples assume you have imported the asts library using import asts. See the methods provided by asts.py for more information.

AST Creation

Constructor

Creating an AST using the Python API requires source text and (optionally) a language enumeration indicating the source text language. The example below shows the creation of a simple AST:

>>> root = asts.AST("x + 88", language=asts.ASTLanguage.Python)

Language enumerations exist for Python, C, Cpp, and Javascript.

For larger examples where the language may be inferred, the language parameter may optionally be elided. For instance:

>>> text = """
... import sys
...
... def fib(n: int) -> int:
...     if n < 2:
...         return n
...     else:
...         return fib(n - 1) + fib(n - 2)
...
... def main():
...     if len(sys.argv) == 1:
...         print(fib(int(sys.argv[1])))
...
... if __name__ == '__main__':
...     main()
"""
>>> root = asts.AST(text)

Finally, by default, the AST returned is the top-level, module node of the parsed AST. However, in some cases, we may wish get the deepest subnode still encompassing all of the given source text. This allows us to create statement AST nodes, for instance. To do so, clients may use the deepest keyword, as shown below:

>>> root = asts.AST("x + 88", language=asts.ASTLanguage.Python, deepest=True)
>>> root.ast_type()
'PYTHON-BINARY-OPERATOR'

AST Templates

Templates for building ASTs

ASTs may also be created using the AST template mechanism. For instance, the following snippet creates an AST equivalent to asts.AST("x = 2", language=asts.ASTLanguage.Python, deepest=True):

>>> root = asts.AST.ast_template("$ID = 2", asts.ASTLanguage.Python, id="x")
>>> root.source_text()
'x = 2'

Metavariable names (e.g. $ID above) may contain uppercase characters, digits, or underscores. Metavariables may be scalars (e.g. $) or lists (e.g. @), as shown below:

>>> root = asts.AST.ast_template("fn(@ARGS)", asts.ASTLanguage.Python, args=[1,2,3])
>>> root.source_text()
'fn(1, 2, 3)'

Metavariables may also be positional arguments (e.g. $1, $2), as shown below:

>>> root = asts.AST.ast_template("$1 = 2", asts.ASTLanguage.Python, "x")
>>> root.source_text()
'x = 2'

However, you may not combine positional (e.g. $1) and keyword (e.g. $ID) metavariables in a single template. The corresponding metavariable values passed in as arguments to ast_template may be ASTs, literals, or lists.

Templates for building and destructuring ASTs

ASTs may also be directly created for the metavariables in an AST template. For instance, in the template "$1 = $2", we may create ASTs for $1 and $2 using asts_from_template, as shown below:

>>> asts = asts.AST.asts_from_template("$1 = $2", asts.ASTLanguage.Python, "x", 1)
>>> len(asts)
2
>>> asts[0].source_text()
'x'
>>> asts[1].source_text()
'1'

For now, only the position syntax (e.g. $1, $2) is supported by asts_from_template. One AST is returned per metavariable, in numeric order.

More information

More information on AST templates may be found in the SEL template documentation.

AST Copy

Copies of an AST may created using AST.copy or the python copy module, as shown below:

>>> root = asts.AST("x + 1", asts.ASTLanguage.Python, deepest=True)
>>> copy = asts.AST.copy(root)
>>> copy.source_text()
'x + 1'

In addition, clients may set child slots in the copy by passing in new ASTs for each slot as keyword arguments, as shown below:

>>> root = asts.AST("x + 1", asts.ASTLanguage.Python, deepest=True)
>>> y = asts.AST("y", asts.ASTLanguage.Python, deepest=True)
>>> copy = asts.AST.copy(root, python_left=y)
>>> copy.source_text()
'y + 1'

In addition to ASTs, clients may also pass in literals (e.g. strings, code snippets, numbers) as keyword values, as shown below. These are automatically parsed into an AST to be inserted.

>>> root = asts.AST("x + 1", asts.ASTLanguage.Python, deepest=True)
>>> copy = asts.AST.copy(root, python_left=1)
>>> copy.source_text()
'1 + 1'
>>> copy = asts.AST.copy(root, python_left="y")
>>> copy.source_text()
'y + 1'

To view the names of an AST's child slots, you may use the child_slots method, as shown below:

>>> root = asts.AST("x + 1", asts.ASTLanguage.Python, deepest=True)
>>> root.child_slots()
[['BEFORE-ASTS', 0], ['PYTHON-LEFT', 1], ['PYTHON-OPERATOR', 1],
 ['PYTHON-RIGHT', 1], ['CHILDREN', 0], ['AFTER-ASTS', 0]]

AST Methods

Common Operations

In practice, most clients will interact with ASTs by retrieving the AST type, source text, parent AST, and child ASTs. All of these operations are supported by the python API. To begin, let's examine retrieving AST type.

There are two ways to retrieve the AST type using the python API - ast_type and ast_types. ast_type returns a string representation of the most specialized subclass the object is an instance of; ast_types returns a list of strings representing the entire class hierarchy the object is embedded within on the Common Lisp side of the interface. This class hierarchy is rich and contains generic mixins for common AST types across languages; for instance, "IF-STATEMENT-AST", "STATEMENT-AST", "IDENTIFIER-AST", etc. For cross-language applications, it is useful to utilize these language-agnostic types. An example of the ast_type and ast_types methods is shown below; please note that CALL-AST is a generic, cross-language mixin type for all function callsite ASTs.

>>> root = asts.AST("print(x)", language=asts.ASTLanguage.Python, deepest=True)
>>> root.ast_type()
'PYTHON-CALL'
>>> root.ast_types()
['PYTHON-CALL', 'PYTHON-PRIMARY-EXPRESSION', 'PYTHON-EXPRESSION', 'PYTHON-AST',
 'TREE-SITTER-AST', 'INDENTATION', 'STRUCTURED-TEXT', 'FUNCTIONAL-TREE-AST',
 'NODE', 'IDENTITY-ORDERING-MIXIN', 'OID-OBJECT', 'STORED-HASH', 'CALL-AST',
 'EXPRESSION-AST', 'AST', 'STANDARD-OBJECT', 'SLOT-OBJECT']

Beyond AST types, retrieving the source text is another common operation. This may be accomplished using the source_text method, as shown below:

>>> root = asts.AST("print(x)", language=asts.ASTLanguage.Python, deepest=True)
>>> root.source_text()
'print(x)'

Finally, subtrees and parent trees may be accessed using the children and parent methods, as shown below. Please note that the parent method requires the root of the subtree as a parameter.

>>> root = asts.AST("print(x)", language=asts.ASTLanguage.Python, deepest=True)
>>> root.children()
[<PYTHON-IDENTIFIER 0x3b4>, <PYTHON-ARGUMENT-LIST-1 0x3b5>]
>>> root.children()[0].source_text()
'print'
>>> identifier = root.children()[1].children()[0]
>>> identifier.source_text()
'x'
>>> identifier.parent(root).source_text()
'(x)'

An AST is composed of various child slots which are concatenated together when using the children method. To view the child slots for a particular AST you may use the child_slots method, which returns a list of slot-name, arity pairs. An arity of one indicates the slot is a single AST, while an arity of zero indicates the slot is composed of zero or more ASTs. The AST(s) comprising a given slot may be accessed using the child_slot method. An example is shown below:

>>> root = asts.AST("print(x)", language=asts.ASTLanguage.Python, deepest=True)
>>> root.child_slots()
[['BEFORE-ASTS', 0], ['PYTHON-FUNCTION', 1], ['PYTHON-ARGUMENTS', 1],
 ['CHILDREN', 0], ['AFTER-ASTS', 0]]
>>> root.child_slot("PYTHON-FUNCTION").source_text()
'print'

Source Locations

For some applications, it may be useful to know the start/end locations of each AST or retrieve the AST at a given location. Clients may do both using the ast_source_ranges and ast_at_point methods respectively, as shown below. Please note that for each method the lines and columns are 1-indexed.

>>> root = asts.AST("print(x)", language=asts.ASTLanguage.Python, deepest=True)
>>> root.ast_source_ranges()
[[<PYTHON-CALL 0x3b3>, [[1, 1], [1, 9]]],
 [<PYTHON-IDENTIFIER 0x3b4>, [[1, 1], [1, 6]]],
 [<PYTHON-ARGUMENT-LIST-1 0x3b5>, [[1, 6], [1, 9]]],
 [<PYTHON-IDENTIFIER 0x3b6>, [[1, 7], [1, 8]]]]
>>> root.ast_at_point(1, 7).source_text()
'x'

Functions

Function ASTs have special consideration in the python API, and clients may retrieve various function attributes, such as name, parameters, and body, using the respective AST methods, function_name, function_parameters, and function_body, as shown below:

>>> root = asts.AST("def foo(bar: int) -> int:\n    return bar / 2",
...                 language=asts.ASTLanguage.Python,
...                 deepest=True)
>>> root.function_name()
'foo'
>>> [param.source_text() for param in root.function_parameters()]
['bar: int']
>>> root.function_body().source_text()
'    return bar / 2'

Function Callsites

In addition to function ASTs, function callsites also have special consideration in the python API. Clients may query for the library containing the function implementation (provided_by), the function portion of the callsite (call_function), and the callargs (call_arguments). An example incorporating these methods is shown below:

>>> root = asts.AST("import json\njson.dumps({})", language=asts.ASTLanguage.Python)
>>> callsite = root.children()[-1].children()[-1]
>>> callsite.provided_by(root)
'json'
>>> callsite.call_function().source_text()
'json.dumps'
>>> [callarg.source_text() for callarg in callsite.call_arguments()]
['{}']

AST Traversal

ASTs may be explictly traversed in pre-order using the traverse method which creates a generator that may be used anywhere a python iterable is required. An example usage is shown below:

>>> root = asts.AST("x + 88", language=asts.ASTLanguage.Python)
>>> for a in root.traverse():
...     print(a.ast_type())
PYTHON-MODULE
PYTHON-EXPRESSION-STATEMENT-0
PYTHON-BINARY-OPERATOR
PYTHON-IDENTIFIER
PYTHON-+
PYTHON-INTEGER

Additionally, AST objects are themselves iterators and may be used anywhere a python iterable is required, as shown below:

>>> root = asts.AST("x + 88", language=asts.ASTLanguage.Python)
>>> for a in root:
...     print(a.ast_type())
PYTHON-MODULE
PYTHON-EXPRESSION-STATEMENT-0
PYTHON-BINARY-OPERATOR
PYTHON-IDENTIFIER
PYTHON-+
PYTHON-INTEGER

As expected, ASTs may be also be used in list comprehensions as shown:

>>> root = asts.AST("x + 88", language=asts.ASTLanguage.Python)
>>> ids = [a for a in root if a.ast_type() == 'PYTHON-IDENTIFIER']
>>> len(ids)
1

AST Manipulation

ASTs may also be manipulated (mutated) using the python API. Mutation operations create a new AST distinct from the inputs. The input ASTs may continue to be used as before; however, they are unchanged objects distinct from the AST(s) created.

Mutation Primitives

Currently, clients may cut, insert, or replace AST subtrees, as shown:

CUT:

>>> root = asts.AST("x = 2\n", language=asts.ASTLanguage.Python)
>>> stmt = root.children()[0]
>>> root = asts.AST.cut(root, stmt)
>>> root.source_text()
''

INSERT:

>>> root = asts.AST("y = 3\n", language=asts.ASTLanguage.Python)
>>> stmt = root.children()[0]
>>> new_stmt = asts.AST("x = 2\n", language=asts.ASTLanguage.Python, deepest=True)
>>> root = asts.AST.insert(root, stmt, new_stmt)
>>> root.source_text()
'x = 2\ny = 3\n'

REPLACE:

>>> root = asts.AST("x = 2\n", language=asts.ASTLanguage.Python)
>>> literal = root.children()[0].children()[0].children()[-1]
>>> new_literal = asts.AST("3", language=asts.ASTLanguage.Python, deepest=True)
>>> root = asts.AST.replace(root, literal, new_literal)
>>> root.source_text()
"x = 3\n"

As a useful shortcut, for small mutations, literals may be passed as the values for insertion and replacement, as shown below:

>>> root = asts.AST("x = 2\n", language=asts.ASTLanguage.Python)
>>> literal = root.children()[0].children()[0].children()[-1]
>>> root = asts.AST.replace(root, literal, 3)
>>> root.source_text()
"x = 3\n"

Transformers

In addition to simple mutation primitives, the API also supports walking the AST tree and performing transformations on the nodes within. This mode of mutation requires the client to define a transformer function which takes an AST node parameter as input and (optionally) returns an AST or literal which should replace that node in the new tree. If no new node is provided by the transformer function, the node is not replaced in the newly created tree.

For instance, consider a scenario where you wish to rewrite an AST, replacing all x identifiers with y. To begin, you would define an x_to_y transformer function, as shown below:

>>> def x_to_y(ast: asts.AST) -> Optional[asts.LiteralOrAST]:
...     """Convert 'x' identifier ASTs to 'y'."""
...     if "IDENTIFIER-AST" in ast.ast_types() and "x" == ast.source_text():
...         return asts.AST("y", language=ast.ast_language(), deepest=True)
...

As you can see, this function returns a y identifier when it encounters an x identifier. To use the transformer to replace nodes in an AST tree, you would use the AST.transform function, as shown below:

>>> text = """
... x = 1
... print(x)
... """
>>> root = asts.AST(text, asts.ASTLanguage.Python)
>>> print(root.source_text().strip())
x = 1
print(x)
>>> transformed = asts.AST.transform(root, x_to_y)
>>> print(transformed.source_text().strip())
y = 1
print(y)

In the example above, the x_to_y transformer returned an AST. The transformer function may also return a literal value, which will be parsed as an AST. For instance, the below x_to_y transformer is functionally equivalent to the example above:

>>> def x_to_y(ast: asts.AST) -> Optional[asts.LiteralOrAST]:
...     """Convert 'x' identifier ASTs to 'y'."""
...     if "IDENTIFIER-AST" in ast.ast_types() and "x" == ast.source_text():
...         return "y"
...

Transformers may implement more complicated logic than the simple x_to_y transform above. For instance, one may wish to convert x identifiers to y, but only for the left-hand side of assignment statements, as shown below:

>>> def x_to_y_assignment_lhs(ast: asts.AST) -> Optional[asts.LiteralOrAST]:
...     """Convert 'x' identifier ASTs to 'y' on the lhs of assignments."""
...     if (
...         "PYTHON-ASSIGNMENT" in ast.ast_types()
...         and "x" == ast.child_slot("python-left").source_text()
...     ):
...         return asts.AST.copy(ast, python_left="y")
...
>>> text = """
... x = 1
... print(y)
... """
>>> root = asts.AST(text, asts.ASTLanguage.Python)
>>> transformed = asts.AST.transform(root, x_to_y_assignment_lhs)
>>> print(transformed.source_text().strip())
y = 1
print(y)

For these more complicated transforms, you may need to mutate the parent of the node you are interested in. For instance, above we create a new python assignment node with the left-hand side modified to the value we want.

Tying everything together, consider the case where wish to delete all print statements from an AST. This may be accomplished by first defining a predicate for print statements, as shown below:

>>> def is_print_statement(ast: asts.AST) -> bool:
...     """Return TRUE if AST is an statement calling the print function."""
...     if "EXPRESSION-STATEMENT-AST" in ast.ast_types():
...         fn_calls = [c.call_function().source_text() for c in ast.call_asts()]
...         return "print" in fn_calls
...     return False
...

Once this predicate is in place, we may use it to define a transformer which returns a node with the print statements immediately below it elided:

>>> def delete_print_statements(ast: asts.AST) -> Optional[asts.LiteralOrAST]:
...     """Delete all print statements from the children of AST."""
...     if "ROOT-AST" in ast.ast_types() or "COMPOUND-AST" in ast.ast_types():
...         # Build a list of new children under the AST, eliding print statements.
...         new_children = [c for c in ast.children() if not is_print_statement(c)]
...
...         # Special case; if no children remain, add a "pass" statement nop to
...         # avoid syntax errors.
...         new_children = new_children if new_children else ["pass\n"]
...
...         # Create the new node with print statements removed from the children.
...         return asts.AST.copy(ast, children=new_children)
...

Finally, as before, we may use the delete_print_statements transformer to mutate an AST, as shown below:

>>> text = """
... x = 1
... y = 2
... if x > 1:
...     x = x ** y
...     print("Test One: %d" % x)
... else:
...     print("Test Two: %d" % x)
... print("y = %d", y)
... """
>>> root = asts.AST(text, asts.ASTLanguage.Python)
>>> transformed = asts.AST.transform(root, delete_print_statements)
>>> print(transformed.source_text().strip())
x = 1
y = 2
if x > 1:
    x = x ** y
else:
    pass

Architecture

The python library is a thin wrapper around a Common Lisp program named tree-sitter-interface which calls the required pieces of the Software Evolution Library (SEL). Most API calls are delegated to this interface which we communicate with using JSON formatted input/ output over stdio/stdout or a socket.

The python AST objects contain a handle attribute representing an object id (oid) on the Common Lisp side of the interface; in essence, the python ASTs are pointers to Common Lisp memory locations. When calling a python AST method, the oid is serialized to the Common Lisp side of the interface where the underlying AST object is found (dereferenced) and the operation performed. You may get the object id using the get_oid method on python ASTs; to test for python AST equality, we check to see if the ASTs point to the same object using the oids.

To allow for garbage collection, the ASTs are manually reference counted. Whenever a python AST (pointer) is created, the reference count for the associated Common Lisp AST is incremented. Similarly, as python ASTs are garbage collected the reference counter is decremented. When the reference counter reaches zero, the Common Lisp AST is released and may be garbage collected at the appropriate time. To get the reference count for a particular python AST, you may use the ast_ref_count method.

The underlying Common Lisp ASTs are themselves treated as immutable. Therefore, when performing mutation operations (e.g. cut, replace, insert), new ASTs are created in the process.

License

GPLv3+