Skip to main content

Get the currently executing AST node of a frame, and other information

Project description

executing

Build Status Coverage Status Supports Python versions 2.7 and 3.4+, including PyPy

This mini-package lets you get information about what a frame is currently doing, particularly the AST node being executed.

Usage

Getting the AST node

import executing

node = executing.Source.executing(frame).node

Then node will be an AST node (from the ast standard library module) or None if the node couldn't be identified (which may happen often and should always be checked).

node will always be the same instance for multiple calls with frames at the same point of execution.

If you have a traceback object, pass it directly to Source.executing() rather than the tb_frame attribute to get the correct node.

Getting the source code of the node

For this you will need to separately install the asttokens library, then obtain an ASTTokens object:

executing.Source.executing(frame).source.asttokens()

or:

executing.Source.for_frame(frame).asttokens()

or use one of the convenience methods:

executing.Source.executing(frame).text()
executing.Source.executing(frame).text_range()

Getting the __qualname__ of the current function

executing.Source.executing(frame).code_qualname()

or:

executing.Source.for_frame(frame).code_qualname(frame.f_code)

The Source class

Everything goes through the Source class. Only one instance of the class is created for each filename. Subclassing it to add more attributes on creation or methods is recommended. The classmethods such as executing will respect this. See the source code and docstrings for more detail.

Installation

pip install executing

If you don't like that you can just copy the file executing.py, there are no dependencies (but of course you won't get updates).

How does it work?

Suppose the frame is executing this line:

self.foo(bar.x)

and in particular it's currently obtaining the attribute self.foo. Looking at the bytecode, specifically frame.f_code.co_code[frame.f_lasti], we can tell that it's loading an attribute, but it's not obvious which one. We can narrow down the statement being executed using frame.f_lineno and find the two ast.Attribute nodes representing self.foo and bar.x. How do we find out which one it is, without recreating the entire compiler in Python?

The trick is to modify the AST slightly for each candidate expression and observe the changes in the bytecode instructions. We change the AST to this:

(self.foo ** 'longuniqueconstant')(bar.x)

and compile it, and the bytecode will be almost the same but there will be two new instructions:

LOAD_CONST 'longuniqueconstant'
BINARY_POWER

and just before that will be a LOAD_ATTR instruction corresponding to self.foo. Seeing that it's in the same position as the original instruction lets us know we've found our match.

Is it reliable?

Yes - if it identifies a node, you can trust that it's identified the correct one. The tests are very thorough - in addition to unit tests which check various situations directly, there are property tests against a large number of files (see the filenames printed in this build) with real code. Specifically, for each file, the tests:

  1. Identify as many nodes as possible from all the bytecode instructions in the file, and assert that they are all distinct
  2. Find all the nodes that should be identifiable, and assert that they were indeed identified somewhere

In other words, it shows that there is a one-to-one mapping between the nodes and the instructions that can be handled. This leaves very little room for a bug to creep in.

Furthermore, executing checks that the instructions compiled from the modified AST exactly match the original code save for a few small known exceptions. This accounts for all the quirks and optimisations in the interpreter.

Which nodes can it identify?

Currently it works in almost all cases for the following ast nodes:

  • Call, e.g. self.foo(bar)
  • Attribute, e.g. point.x
  • Subscript, e.g. lst[1]
  • BinOp, e.g. x + y (doesn't include and and or)
  • UnaryOp, e.g. -n (includes not but only works sometimes)
  • Compare e.g. a < b (not for chains such as 0 < p < 1)

The plan is to extend to more operations in the future.

Libraries that use this

My libraries

  • stack_data: Extracts data from stack frames and tracebacks, particularly to display more useful tracebacks than the default. Also uses another related library of mine: pure_eval.
  • snoop: A feature-rich and convenient debugging library. Uses executing to show the operation which caused an exception and to allow the pp function to display the source of its arguments.
  • heartrate: A simple real time visualisation of the execution of a Python program. Uses executing to highlight currently executing operations, particularly in each frame of the stack trace.
  • sorcery: Dark magic delights in Python. Uses executing to let special callables called spells know where they're being called from.

Libraries I've contributed to

  • IPython: Highlights the executing node in tracebacks using executing via stack_data.
  • icecream: 🍦 Sweet and creamy print debugging. Uses executing to identify where ic is called and print its arguments.
  • sentry_sdk: Add the integration sentry_sdk.integrations.executingExecutingIntegration() to show the function __qualname__ in each frame in sentry events. Highlighting the executing node is hopefully coming soon.
  • varname: Dark magics about variable names in python. Uses executing to find where its various magical functions like varname and nameof are called from.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for executing, version 0.6.0
Filename, size File type Python version Upload date Hashes
Filename, size executing-0.6.0-py2.py3-none-any.whl (12.8 kB) File type Wheel Python version py2.py3 Upload date Hashes View
Filename, size executing-0.6.0.tar.gz (329.0 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring DigiCert DigiCert EV certificate Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page