Starting the debugger#

To start debugging, simply insert:

from pudb import set_trace; set_trace()

A shorter alternative to this is:

import pudb; pu.db

Or, if pudb is already imported, just this will suffice:

pu.db

If you are using Python 3.7 or newer, you can add:

# Set breakpoint() in Python to call pudb
export PYTHONBREAKPOINT="pudb.set_trace"

in your ~/.bashrc. Then use:

breakpoint()

to start pudb.

Insert one of these snippets into the piece of code you want to debug, or run the entire script with:

python -m pudb my-script.py

which is useful if you want to run PuDB in a version of Python other than the one you most recently installed PuDB with.

Debugging from a separate terminal#

It’s possible to control the debugger from a separate terminal. This is useful if there are several threads running that are printing to stdout while you’re debugging and messing up the terminal, or if you want to keep the original terminal available for any other reason.

Open a new terminal. First, you need to get the path of the tty of the terminal you want to debug from. To do that, use the standard unix command tty. It will print something like /dev/pts/3.

Then you need to make sure that your terminal doesn’t have a shell actively reading and possibly capturing some of the input that should go to pudb. To do that run a placeholder command that does nothing, such as perl -MPOSIX -e pause.

Then set the PUDB_TTY environment variable to the path tty gave you, for example:

PUDB_TTY=/dev/pts/3 pudb my-script.py

Now instead of using the current terminal, pudb will use this tty for its UI. You may want to use the internal shell in pudb, as others will still use the original terminal.

Logging Internal Errors#

Some kinds of internal exceptions encountered by pudb will be logged to the terminal window when the debugger is active. To send these messages to a file instead, use the --log-errors flag:

python -m pudb --log-errors pudberrors.log

Remote debugging#

Rudimentary remote debugging is also supported. To break into the debugger, enabling you to connect via telnet, use the following code:

from pudb.remote import set_trace
set_trace(term_size=(80, 24))

The terminal size can be defined via the environment variable as well:

export PUDB_TERM_SIZE=80x24

The following precedence (from highest to lowest) is used to determine the terminal size:

  1. term_size keyword argument

  2. PUDB_TERM_SIZE environment variable

  3. Size of the terminal in which the debugged program is running (as returned by os.get_terminal_size())

  4. Default fallback value of (80, 20)

At this point, the debugger will look for a free port and wait for a telnet connection:

pudb:6899: Please start a telnet session using a command like:
telnet 127.0.0.1 6899
pudb:6899: Waiting for client...

The host and port can be specified as keyword arguments to set_trace(), or via the PUDB_RDB_HOST and PUDB_RDB_PORT env vars.

To debug a function in a remote debugger (and examine any exceptions that may occur), use code like the following:

def debugged_function(x):
    y = x + fail  # noqa: F821
    return y


from pudb.remote import debugger
dbg = debugger()
dbg.runcall(debugged_function, 5)

Upon running this, again, the debugger will wait for a telnet connection.

The following programming interface is available for the remote debugger:

class pudb.remote.RemoteDebugger(host='127.0.0.1', port=6899, port_search_limit=100, out=<_io.TextIOWrapper name='<stdout>' mode='w' encoding='utf-8'>, term_size=None, reverse=False)[source]#
__init__(host='127.0.0.1', port=6899, port_search_limit=100, out=<_io.TextIOWrapper name='<stdout>' mode='w' encoding='utf-8'>, term_size=None, reverse=False)[source]#
Parameters:

term_size

A two-tuple (columns, rows), or None. If None, try to determine the terminal size automatically.

Currently, this uses a heuristic: It uses the terminal size of the debuggee as that for the debugger. The idea is that you might be running both in two tabs of the same terminal window, hence using terminals of the same size.

pudb.remote.set_trace(frame=None, term_size=None, host='127.0.0.1', port=6899, reverse=False)[source]#

Set breakpoint at current location, or a specified frame

pudb.remote.debugger(term_size=None, host='127.0.0.1', port=6899, reverse=False)[source]#

Return the current debugger instance (if any), or creates a new one.

pudb.remote.debug_remote_on_single_rank(comm: Any, rank: int, func: Callable, *args: Any, **kwargs: Any) None[source]#

Run a remote debugger on a single rank of an mpi4py application. func will be called on rank rank running in a RemoteDebugger, and will be called normally on all other ranks.

Parameters:
  • comm – an mpi4py Comm object.

  • rank – the rank to debug. All other ranks will spin until this rank exits.

  • func – the callable to debug.

  • args – the arguments passed to func.

  • kwargs – the kwargs passed to func.

“Reverse” remote debugging#

In “reverse” remote debugging, pudb connects to a socket, rather than listening to one.

First open the socket and listen using the netcat(nc), as below. Netcat of course is not a telnet client, so it can behave differently than a telnet client. By using the `stty` with “no echo: and “no buffering” input options, we can make a socket that nonetheless behave simillarly:

stty -echo -icanon && nc -l -p 6899

When using the BSD version netcat that ships with MacOS, a server can be started like this:

stty -echo -icanon && nc -l 6899

Specify host and port in set_trace and set the reverse parameter to True:

from pudb.remote import set_trace
set_trace(reverse=True)

The “reverse” mode can also be enabled by setting the environment variable to a non-empty value (the keyword argument has priority over the env var):

export PUDB_RDB_REVERSE=1

Then watch the debugger connect to netcat:

pudb:9999: Now in session with 127.0.0.1:6899.

Using the debugger after forking#

In a forked process, no TTY is usually attached to stdin/stdout, which leads to errors when debugging with standard pudb. E.g. consider this script.py:

from multiprocessing import Process
def f(name):
    # breakpoint was introduced in Python 3.7
    breakpoint()
    print('hello', name)

p = Process(target=f, args=('bob',))
p.start()
p.join()

Running it with standard pudb breaks:

PYTHONBREAKPOINT=pudb.set_trace python script.py

However, on Unix systems, e.g. Linux & MacOS, debugging a forked process is supported using pudb.forked.set_trace:

PYTHONBREAKPOINT=pudb.forked.set_trace python script.py

Usage with pytest#

To use PuDB with pytest, consider using the pytest-pudb plugin, which provides a --pudb option that simplifies the procedure below.

Alternatively, as of version 2017.1.2, pudb can be used to debug test failures in pytest, by running the test runner like so:

$ pytest --pdbcls pudb.debugger:Debugger --pdb --capture=no

Note the need to pass –capture=no (or its synonym -s) as otherwise pytest tries to manage the standard streams itself. (contributed by Antony Lee)