Debugging your Cython program

Cython comes with an extension for the GNU Debugger that helps users debug Cython code. To use this functionality, you will need to install gdb 7.2 or higher, built with Python support (linked to Python 2.6 or higher). The debugger supports debuggees with versions 2.6 and higher. For Python 3, code should be built with Python 3 and the debugger should be run with Python 2 (or at least it should be able to find the Python 2 Cython installation). Note that in recent versions of Ubuntu, for instance, gdb installed with apt-get is configured with Python 3. On such systems, the proper configuration of gdb can be obtained by downloading the gdb source, and then running:

  1. ./configure --with-python=python2
  2. make
  3. sudo make install

The debugger will need debug information that the Cython compiler can export. This can be achieved from within the setup script by passing gdb_debug=True to cythonize():

  1. from distutils.core import setup
  2. from distutils.extension import Extension
  3. extensions = [Extension('source', ['source.pyx'])]
  4. setup(..., ext_modules=cythonize(extensions, gdb_debug=True))

For development it’s often helpful to pass the --inplace flag to the setup.py script, which makes distutils build your project “in place”, i.e., not in a separate build directory.

When invoking Cython from the command line directly you can have it write debug information using the --gdb flag:

  1. cython --gdb myfile.pyx

Running the Debugger

To run the Cython debugger and have it import the debug information exported by Cython, run cygdb in the build directory:

  1. $ python setup.py build_ext --inplace
  2. $ cygdb
  3. GNU gdb (GDB) 7.2
  4. ...
  5. (gdb)

When using the Cython debugger, it’s preferable that you build and run your code with an interpreter that is compiled with debugging symbols (i.e. configured with --with-pydebug or compiled with the -g CFLAG). If your Python is installed and managed by your package manager you probably need to install debug support separately. If using NumPy then you also need to install numpy debugging, or you’ll see an import error for multiarray. E.G. for ubuntu:

  1. $ sudo apt-get install python-dbg python-numpy-dbg
  2. $ python-dbg setup.py build_ext --inplace

Then you need to run your script with python-dbg also. Ensure that when building your package with debug symbols that cython extensions are re-compiled if they had been previously compiled. If your package is version controlled, you might want to perform git clean -fxd or hg purge --all before building.

You can also pass additional arguments to gdb:

  1. $ cygdb /path/to/build/directory/ GDBARGS

i.e.:

  1. $ cygdb . -- --args python-dbg mainscript.py

To tell cygdb not to import any debug information, supply -- as the first argument:

  1. $ cygdb --

Using the Debugger

The Cython debugger comes with a set of commands that support breakpoints, stack inspection, source code listing, stepping, stepping over, etc. Most of these commands are analogous to their respective gdb command.

cy break breakpoints...

Break in a Python, Cython or C function. First it will look for a Cython function with that name, if cygdb doesn’t know about a function (or method) with that name, it will set a (pending) C breakpoint. The -p option can be used to specify a Python breakpoint.

Breakpoints can be set for either the function or method name, or they can be fully “qualified”, which means that the entire “path” to a function is given:

  1. (gdb) cy break cython_function_or_method
  2. (gdb) cy break packagename.cython_module.cython_function
  3. (gdb) cy break packagename.cython_module.ClassName.cython_method
  4. (gdb) cy break c_function

You can also break on Cython line numbers:

  1. (gdb) cy break :14
  2. (gdb) cy break cython_module:14
  3. (gdb) cy break packagename.cython_module:14

Python breakpoints currently support names of the module (not the entire package path) and the function or method:

  1. (gdb) cy break -p python_module.python_function_or_method
  2. (gdb) cy break -p python_function_or_method

Note

Python breakpoints only work in Python builds where the Python frame information can be read from the debugger. To ensure this, use a Python debug build or a non-stripped build compiled with debug support.

cy step

Step through Python, Cython or C code. Python, Cython and C functions called directly from Cython code are considered relevant and will be stepped into.

cy next

Step over Python, Cython or C code.

cy run

Run the program. The default interpreter is the interpreter that was used to build your extensions with, or the interpreter cygdb is run with in case the “don’t import debug information” option was in effect. The interpreter can be overridden using gdb’s file command.

cy cont

Continue the program.

cy up

cy down

Go up and down the stack to what is considered a relevant frame.

cy finish

Execute until an upward relevant frame is met or something halts execution.

cy bt

cy backtrace

Print a traceback of all frames considered relevant. The -a option makes it print the full traceback (all C frames).

cy select

Select a stack frame by number as listed by cy backtrace. This command is introduced because cy backtrace prints a reversed stack trace, so frame numbers differ from gdb’s bt.

cy print varname

Print a local or global Cython, Python or C variable (depending on the context). Variables may also be dereferenced:

  1. (gdb) cy print x
  2. x = 1
  3. (gdb) cy print *x
  4. *x = (PyObject) {
  5. _ob_next = 0x93efd8,
  6. _ob_prev = 0x93ef88,
  7. ob_refcnt = 65,
  8. ob_type = 0x83a3e0
  9. }

cy set cython_variable = value

Set a Cython variable on the Cython stack to value.

cy list

List the source code surrounding the current line.

cy locals

cy globals

Print all the local and global variables and their values.

cy import FILE...

Import debug information from files given as arguments. The easiest way to import debug information is to use the cygdb command line tool.

cy exec code

Execute code in the current Python or Cython frame. This works like Python’s interactive interpreter.

For Python frames it uses the globals and locals from the Python frame, for Cython frames it uses the dict of globals used on the Cython module and a new dict filled with the local Cython variables.

Note

cy exec modifies state and executes code in the debuggee and is therefore potentially dangerous.

Example:

  1. (gdb) cy exec x + 1
  2. 2
  3. (gdb) cy exec import sys; print sys.version_info
  4. (2, 6, 5, 'final', 0)
  5. (gdb) cy exec
  6. >global foo
  7. >
  8. >foo = 'something'
  9. >end

Convenience functions

The following functions are gdb functions, which means they can be used in a gdb expression.

cy_cname(varname)

Returns the C variable name of a Cython variable. For global variables this may not be actually valid.

cy_cvalue(varname)

Returns the value of a Cython variable.

cy_eval(expression)

Evaluates Python code in the nearest Python or Cython frame and returns the result of the expression as a gdb value. This gives a new reference if successful, NULL on error.

cy_lineno()

Returns the current line number in the selected Cython frame.

Example:

  1. (gdb) print $cy_cname("x")
  2. $1 = "__pyx_v_x"
  3. (gdb) watch $cy_cvalue("x")
  4. Hardware watchpoint 13: $cy_cvalue("x")
  5. (gdb) cy set my_cython_variable = $cy_eval("{'spam': 'ham'}")
  6. (gdb) print $cy_lineno()
  7. $2 = 12

Configuring the Debugger

A few aspects of the debugger are configurable with gdb parameters. For instance, colors can be disabled, the terminal background color and breakpoint autocompletion can be configured.

cy_complete_unqualified

Tells the Cython debugger whether cy break should also complete plain function names, i.e. not prefixed by their module name. E.g. if you have a function named spam, in module M, it tells whether to only complete M.spam or also just spam.

The default is true.

cy_colorize_code

Tells the debugger whether to colorize source code. The default is true.

cy_terminal_background_color

Tells the debugger about the terminal background color, which affects source code coloring. The default is “dark”, another valid option is “light”.

This is how these parameters can be used:

  1. (gdb) set cy_complete_unqualified off
  2. (gdb) set cy_terminal_background_color light
  3. (gdb) show cy_colorize_code