Python module dependency visualization.

This package is primarily intended to be used from the command line through the pydeps command.

Feature requests and bug reports:

Please report bugs and feature requests on GitHub at https://github.com/thebjorn/pydeps/issues

pip install pydeps

To create graphs with pydeps you also need to install Graphviz. Please follow the installation instructions provided in the Graphviz link (and make sure the dot command is on your path).

usage: pydeps [-h] [--debug] [--config FILE] [--no-config] [--version] [-L LOG]
              [--find-package] [-v] [-o file] [-T FORMAT] [--display PROGRAM]
              [--noshow] [--show-deps] [--show-raw-deps] [--deps-output DEPS_OUT]
              [--show-dot] [--dot-output DOT_OUT] [--nodot] [--no-output]
              [--show-cycles] [--debug-mf INT] [--noise-level INT]
              [--max-bacon INT] [--max-module-depth INT] [--pylib] [--pylib-all]
              [--include-missing] [-x PATTERN [PATTERN ...]]
              [-xx MODULE [MODULE ...]] [--only MODULE_PATH [MODULE_PATH ...]]
              [--externals] [--reverse] [--rankdir {TB,BT,LR,RL}] [--cluster]
              [--min-cluster-size INT] [--max-cluster-size INT]
              [--keep-target-cluster] [--collapse-target-cluster]
              [--rmprefix PREFIX [PREFIX ...]] [--start-color INT]
              fname

positional arguments:
  fname                 filename

optional arguments:
  -h, --help                             show this help message and exit
  --debug                                turn on all the show and verbose options (mainly for debugging pydeps itself)
  --config FILE                          specify config file
  --no-config                            disable processing of config files
  --version                              print pydeps version
  -L LOG, --log LOG                      set log-level to one of CRITICAL, ERROR, WARNING, INFO, DEBUG, NOTSET.
  --find-package                         tries to automatically find the name of the current package.
  -v, --verbose                          be more verbose (-vv, -vvv for more verbosity)
  -o file                                write output to 'file'
  -T FORMAT                              output format (svg|png)
  --display PROGRAM                      program to use to display the graph (png or svg file depending on the T parameter)
  --noshow, --no-show                    don't call external program to display graph
  --show-deps                            show output of dependency analysis
  --show-raw-deps                        show output of dependency analysis before removing skips
  --deps-output                          write output of dependency analysis to file (instead of screen)
  --show-dot                             show output of dot conversion
  --dot-output                           write dot code to file (instead of screen)
  --nodot, --no-dot                      skip dot conversion
  --no-output                            don't create .svg/.png file, implies --no-show (-t/-o will be ignored)
  --show-cycles                          show only import cycles
  --debug-mf INT                         set the ModuleFinder.debug flag to this value
  --noise-level INT                      exclude sources or sinks with degree greater than noise-level
  --max-bacon INT                        exclude nodes that are more than n hops away (default=2, 0 -> infinite)
  --max-module-depth INT                 coalesce deep modules to at most n levels
  --pylib                                include python std lib modules
  --pylib-all                            include python all std lib modules (incl. C modules)
  --include-missing                      include modules that are not installed (or can't be found on sys.path)
  --only MODULE_PATH                     only include modules that start with MODULE_PATH, multiple paths can be provided
  --externals                            create list of direct external dependencies
  --reverse                              draw arrows to (instead of from) imported modules
  --rankdir                              set the direction of the graph, legal values are TB (default, imported modules above importing modules), BT (opposite direction of TB), LR (left-to-right), and RL (right-to-left)
  --cluster                              draw external dependencies as separate clusters
  --min-cluster-size INT                 the minimum number of nodes a dependency must have before being clustered (default=0)
  --max-cluster-size INT                 the maximum number of nodes a dependency can have before the cluster is collapsed to a single node (default=0)
  --keep-target-cluster                  draw target module as a cluster
  --collapse-target-cluster              collapse target module (this implies --cluster)
  --rmprefix PREFIX                      remove PREFIX from the displayed name of the nodes (multiple prefixes can be provided)
  -x PATTERN, --exclude PATTERN          input files to skip (e.g. `foo.*`), multiple patterns can be provided
  --exclude-exact MODULE                 (shorthand -xx MODULE) same as --exclude, except requires the full match. `-xx foo.bar` will exclude foo.bar, but not foo.bar.blob

Note: if an option with a variable number of arguments (like -x) is provided before fname, separate the arguments from the filename with -- otherwise fname will be parsed as an argument of the option. Example: $ pydeps -x os sys -- pydeps.

You can of course also import pydeps from Python and use it as a library, look in tests/test_relative_imports.py for examples.

This is the result of running pydeps on itself (pydeps pydeps):

(full disclosure: this is for an early version of pydeps)

pydeps finds imports by looking for import-opcodes in python bytecodes (think .pyc files). Therefore, only imported files will be found (ie. pydeps will not look at files in your directory that are not imported). Additionally, only files that can be found using the Python import machinery will be considered (ie. if a module is missing or not installed, it will not be included regardless if it is being imported). This can be modified by using the --include-missing flag.

Displaying the graph:

To display the resulting .svg or .png files, pydeps by default calls an appropriate opener for the platform, like xdg-open foo.svg.

This can be overridden with the --display PROGRAM option, where PROGRAM is an executable that can display the image file of the graph.

You can also export the name of such a viewer in either the PYDEPS_DISPLAY or BROWSER environment variable, which changes the default behaviour when --display is not used.

All options can also be set in a .pydeps file using .ini file syntax (parsable by ConfigParser). Command line options override options in the .pydeps file in the current directory, which again overrides options in the user's home directory (%USERPROFILE%\.pydeps on Windows and ${HOME}/.pydeps otherwise).

An example .pydeps file:

[pydeps]
max_bacon = 2
no_show = True
verbose = 0
pylib = False
exclude =
    os
    re
    sys
    collections
    __future__

pydeps will also look for configuration data in pyproject.toml (under [tool.pydeps]) and setup.cfg (under [pydeps]).

pydeps also contains an Erdős-like scoring function (a.k.a. Bacon number, from Six degrees of Kevin Bacon (http://en.wikipedia.org/wiki/Six_Degrees_of_Kevin_Bacon) that lets you filter out modules that are more than a given number of 'hops' away from the module you're interested in. This is useful for finding the interface a module has to the rest of the world.

To find pydeps' interface to the Python stdlib (less some very common modules).

shell> pydeps pydeps --show --max-bacon 2 --pylib -x os re types _* enum

--max-bacon 2 (the default) gives the modules that are at most 2 hops away, and modules that belong together have similar colors. Compare that to the output with the --max-bacon=0 (infinite) filter:

pydeps can detect and display cycles with the --show-cycles parameter. This will only display the cycles, and for big libraries it is not a particularly fast operation. Given a folder with the following contents (this uses yaml to define a directory structure, like in the tests):

relimp:
    - __init__.py
    - a.py: |
        from . import b
    - b.py: |
        from . import a

pydeps relimp --show-cycles displays:

Running pydeps pydeps --max-bacon=4 on version 1.8.0 of pydeps gives the following graph:

If you are not interested in the internal structure of external modules, you can add the --cluster flag, which will collapse external modules into folder-shaped objects:

shell> pydeps pydeps --max-bacon=4 --cluster

To see the internal structure and delineate external modules, use the --max-cluster-size flag, which controls how many nodes can be in a cluster before it is collapsed to a folder icon:

shell> pydeps pydeps --max-bacon=4 --cluster --max-cluster-size=1000

or, using a smaller max-cluster-size:

shell> pydeps pydeps --max-bacon=4 --cluster --max-cluster-size=3

To remove clusters with too few nodes, use the --min-cluster-size flag:

shell> pydeps pydeps --max-bacon=4 --cluster --max-cluster-size=3 --min-cluster-size=2

In some situations it can be useful to draw the target module as a cluster:

shell> pydeps pydeps --max-bacon=4 --cluster --max-cluster-size=3 --min-cluster-size=2 --keep-target-cluster

..and since the cluster boxes include the module name, we can remove those prefixes:

shell> pydeps pydeps --max-bacon=4 --cluster --max-cluster-size=3 --min-cluster-size=2 --keep-target-cluster --rmprefix pydeps. stdlib_list.

For Python packages that have a module structure more than two levels deep, the graph can easily become overwhelmingly complex. Use the --max-module-depth=n flag to examine the internal dependencies of a package while limiting the module depth (private and testing-related modules are removed to further simplify the graph using -x ...):

shell> pydeps pandas --only pandas --max-module-depth=2 -x pandas._* pandas.test* pandas.conftest

The direction of the graph can be specified using the --rankdir flag.

Top to bottom (default):

shell> pydeps pydeps --rankdir TB

Bottom to top:

shell> pydeps pydeps --rankdir BT

Left to right:

shell> pydeps pydeps --rankdir LR

Right to left:

shell> pydeps pydeps --rankdir RL

When internal target package dependencies are unimportant, they can be collapsed using the --collapse-target-cluster flag. This option also implies --cluster:

shell> pydeps pydeps --collapse-target-cluster

An attempt has been made to keep the intermediate formats readable, eg. the output from pydeps --show-deps .. looks like this:

...
"pydeps.mf27": {
    "imported_by": [
        "__main__",
        "pydeps.py2depgraph"
    ],
    "kind": "imp.PY_SOURCE",
    "name": "pydeps.mf27",
    "path": "pydeps\\mf27.py"
},
"pydeps.py2depgraph": {
    "imported_by": [
        "__main__",
        "pydeps.pydeps"
    ],
    "imports": [
        "pydeps.depgraph",
        "pydeps.mf27"
    ],
    "kind": "imp.PY_SOURCE",
    "name": "pydeps.py2depgraph",
    "path": "pydeps\\py2depgraph.py"
}, ...

Version 2.0.0 Python 3.8+ only. Thanks to timhoffm for the PR.

Version 1.12.19 Thanks to wiguwbe for a PR that fixes an inconsistency with the --no-dot flag.

Version 1.12.13 Better docs for larger packages. See maximum_module_depth for an example. Thanks to sheromon for the PR.

Version 1.12.5 Pydeps can now read configuration data from pyproject.toml. Thanks to septatrix for pushing the idea and for countering my toml-rant with an informative argument.

Version 1.11.0 drop support for Python 3.6. Thanks to pawamoy for removing imports of the deprecated imp module. (Parts of it has been vendorized due to a Python bug, see the code for details.)

Version 1.10.1 Thanks to vector400 for a new option --rankdir which renders the graph in different directions.

Version 1.10.0 supports Python 3.10.

Version 1.9.15 Thanks to Pipeline Foundation for a very much improved CI pipeline, and a CD pipeline as well.

Version 1.9.14 Thanks to poneill for fixing a cryptic error message when run in a directory without an __init__.py file.

Version 1.9.13 Thanks to glumia and SimonBiggs for improving the documentation.

Version 1.9.10 no_show is now honored when placed in .pydeps file. Thanks to romain-dartigues for the PR.

Version 1.9.8 Fix for maximum recursion depth exceeded when using large frameworks (like sympy). Thanks to tanujkhattar for finding the fix and to balopat for reporting it.

Version 1.9.7 Check PYDEPS_DISPLAY and BROWSER for a program to open the graph, PR by jhermann

Version 1.9.1 graphs are now stable on Python 3.x as well - this was already the case for Py2.7 (thanks to pawamoy for reporting and testing the issue and to kinow for helping with testing).

Version 1.9.0 supports Python 3.8.

Version 1.8.7 includes a new flag --rmprefix which lets you remove prefixes from the node-labels in the graph. The name of the nodes are not effected so this does not cause merging of nodes, nor does it change coloring - but it can lead to multiple nodes with the same label (hovering over the node will give the full name). Thanks to aroberge for the enhancement request.

Version 1.8.5 With svg as the output format (which is the default), paths are now hilighted on mouse hover (thanks to tomasito665 for the enhancement request).

Version 1.8.2 incldes a new flag --only that causes pydeps to only report on the paths specified:

shell> pydeps mypackage --only mypackage.a mypackage.b

Version 1.8.0 includes 4 new flags for drawing external dependencies as clusters. See clustering for examples. Additionally, the arrowheads now have the color of the source node.

Version 1.7.3 includes a new flag -xx or --exclude-exact which matches the functionality of the --exclude flag, except it requires an exact match, i.e. -xx foo.bar will exclude foo.bar, but not foo.bar.blob (thanks to AvenzaOleg for the PR).

Version 1.7.2 includes a new flag, --no-output, which prevents creation of the .svg/.png file.

Version 1.7.1 fixes excludes in .pydeps files (thanks to eqvis for the bug report).

Version 1.7.0 The new --reverse flag reverses the direction of the arrows in the dependency graph, so they point to the imported module instead of from the imported module (thanks to goetzk for the bug report and tobiasmaier for the PR!).

Version 1.5.0 Python 3 support (thanks to eight04 for the PR).

Version 1.3.4 --externals will now include modules that haven't been installed (what modulefinder calls badmodules).

Version 1.2.8 A shortcut for finding the direct external dependencies of a package was added:

pydeps --externals mypackage

which will print a json formatted list of module names to the screen, e.g.:

(dev) go|c:\srv\lib\dk-tasklib> pydeps --externals dktasklib
[
    "dkfileutils"
]

which means that the dktasklib package only depends on the dkfileutils package.

This functionality is also available programmatically:

import os
from pydeps.pydeps import externals
# the directory that contains setup.py (one level up from actual package):
os.chdir('package-directory')
print externals('mypackage')

Version 1.2.5: The defaults are now sensible, such that:

shell> pydeps mypackage

will likely do what you want. It is the same as pydeps --show --max-bacon=2 mypackage which means display the dependency graph in your browser, but limit it to two hops (which includes only the modules that your module imports -- not continuing down the import chain). The old default behavior is available with pydeps --noshow --max-bacon=0 mypackage.

1. Fork it
2. It is appreciated (but not required) if you raise an issue first: https://github.com/thebjorn/pydeps/issues
3. Create your feature branch (git checkout -b my-new-feature)
4. Commit your changes (git commit -am 'Add some feature')
5. Push to the branch (git push origin my-new-feature)
6. Create new Pull Request