PEP: 534 Title: Improved Errors for Missing Standard Library Modules Author: Tomáš Orsava , Petr Viktorin , Nick Coghlan Status: Deferred Type: Standards Track Content-Type: text/x-rst Created: 05-Sep-2016 Post-History: Abstract ======== Python is often being built or distributed without its full standard library. However, there is as of yet no standard, user friendly way of properly informing the user about the failure to import such missing standard library modules. This PEP proposes a mechanism for identifying expected standard library modules and providing more informative error messages to users when attempts to import standard library modules fail. PEP Deferral ============ The PEP authors aren't actively working on this PEP, so if improving these error messages is an idea that you're interested in pursuing, please get in touch! (e.g. by posting to the python-dev mailing list). The key piece of open work is determining how to get the autoconf and Visual Studio build processes to populate the sysconfig metadata file with the lists of expected and optional standard library modules. Motivation ========== There are several use cases for including only a subset of Python's standard library. However, there is so far no user-friendly mechanism for informing the user *why* a stdlib module is missing and how to remedy the situation appropriately. CPython ------- When one of Python's standard library modules (such as ``_sqlite3``) cannot be compiled during a CPython build because of missing dependencies (e.g. SQLite header files), the module is simply skipped. If you then install this compiled Python and use it to try to import one of the missing modules, Python will fail with a ModuleNotFoundError_. .. _ModuleNotFoundError: https://docs.python.org/3.7/library/exceptions.html#ModuleNotFoundError For example, after deliberately removing ``sqlite-devel`` from the local system:: $ ./python -c "import sqlite3" Traceback (most recent call last): File "", line 1, in File "/home/ncoghlan/devel/cpython/Lib/sqlite3/__init__.py", line 23, in from sqlite3.dbapi2 import * File "/home/ncoghlan/devel/cpython/Lib/sqlite3/dbapi2.py", line 27, in from _sqlite3 import * ModuleNotFoundError: No module named '_sqlite3' This can confuse users who may not understand why a cleanly built Python is missing standard library modules. Linux and other distributions ----------------------------- Many Linux and other distributions are already separating out parts of the standard library to standalone packages. Among the most commonly excluded modules are the ``tkinter`` module, since it draws in a dependency on the graphical environment, ``idlelib``, since it depends on ``tkinter`` (and most Linux desktop environments provide their own default code editor), and the ``test`` package, as it only serves to test Python internally and is about as big as the rest of the standard library put together. The methods of omission of these modules differ. For example, Debian patches the file ``Lib/tkinter/__init__.py`` to envelop the line ``import _tkinter`` in a *try-except* block and upon encountering an ``ImportError`` it simply adds the following to the error message: ``please install the python3-tk package`` [#debian-patch]_. Fedora and other distributions simply don't include the omitted modules, potentially leaving users baffled as to where to find them. An example from Fedora 29:: $ python3 -c "import tkinter" Traceback (most recent call last): File "", line 1, in ModuleNotFoundError: No module named 'tkinter' Specification ============= APIs to list expected standard library modules ---------------------------------------------- To allow for easier identification of which module names are *expected* to be resolved in the standard library, the `sysconfig`_ module will be extended with two additional functions: * ``sysconfig.get_stdlib_modules()``, which will provide a list of the names of all top level Python standard library modules (including private modules) * ``sysconfig.get_optional_modules()``, which will list optional public top level standard library module names The results of ``sysconfig.get_optional_modules()`` and the existing ``sys.builtin_module_names`` will both be subsets of the full list provided by the new ``sysconfig.get_stdlib_modules()`` function. These added lists will be generated during the Python build process and saved in the ``_sysconfigdata-*.py`` file along with other `sysconfig`_ values. Possible reasons for modules being in the "optional" list will be: * the module relies on an optional build dependency (e.g. ``_sqlite3``, ``tkinter``, ``idlelib``) * the module is private for other reasons and hence may not be present on all implementations (e.g. ``_freeze_importlib``, ``_collections_abc``) * the module is platform specific and hence may not be present in all installations (e.g. ``winreg``) * the ``test`` package may also be freely omitted from Python runtime installations, as it is intended for use in testing Python implementations, not as a runtime library for Python projects to use (the public API offering testing utilities is ``unittest``) (Note: the ``ensurepip``, ``venv``, and ``distutils`` modules are all considered mandatory modules in this PEP, even though not all redistributors currently adhere to that practice) .. _`sysconfig`: https://docs.python.org/3/library/sysconfig.html Changes to the default ``sys.excepthook`` implementation -------------------------------------------------------- The default implementation of the `sys.excepthook`_ function will then be modified to dispense an appropriate message when it detects a failure to import a module identified by one of the two new `sysconfig`_ functions as belonging to the Python standard library. .. _`sys.excepthook`: https://docs.python.org/3/library/sys.html#sys.excepthook Revised error message for a module that relies on an optional build dependency or is otherwise considered optional when Python is installed:: $ ./python -c "import sqlite3" Traceback (most recent call last): File "", line 1, in File "/home/ncoghlan/devel/cpython/Lib/sqlite3/__init__.py", line 23, in from sqlite3.dbapi2 import * File "/home/ncoghlan/devel/cpython/Lib/sqlite3/dbapi2.py", line 27, in from _sqlite3 import * ModuleNotFoundError: Optional standard library module '_sqlite3' was not found Revised error message for a submodule of an optional top level package when the entire top level package is missing:: $ ./python -c "import test.regrtest" Traceback (most recent call last): File "", line 1, in ModuleNotFoundError: Optional standard library module 'test' was not found Revised error message for a submodule of an optional top level package when the top level package is present:: $ ./python -c "import test.regrtest" Traceback (most recent call last): File "", line 1, in ModuleNotFoundError: No submodule named 'test.regrtest' in optional standard library module 'test' Revised error message for a module that is always expected to be available:: $ ./python -c "import ensurepip" Traceback (most recent call last): File "", line 1, in ModuleNotFoundError: Standard library module 'ensurepip' was not found Revised error message for a missing submodule of a standard library package when the top level package is present:: $ ./python -c "import encodings.mbcs" Traceback (most recent call last): File "", line 1, in ModuleNotFoundError: No submodule named 'encodings.mbcs' in standard library module 'encodings' These revised error messages make it clear that the missing modules are expected to be available from the standard library, but are not available for some reason, rather than being an indicator of a missing third party dependency in the current environment. Design Discussion ================= Modifying ``sys.excepthook`` ---------------------------- The `sys.excepthook`_ function gets called when a raised exception is uncaught and the program is about to exit or (in an interactive session) the control is being returned to the prompt. This makes it a perfect place for customized error messages, as it will not influence caught errors and thus not slow down normal execution of Python scripts. Public API to query expected standard library module names ---------------------------------------------------------- The inclusion of the functions ``sysconfig.get_stdlib_modules()`` and ``sysconfig.get_optional_modules()`` will provide a long sought-after way of easily listing the names of Python standard library modules [#stackoverflow-stdlib]_, which will (among other benefits) make it easier for code analysis, profiling, and error reporting tools to offer runtime ``--ignore-stdlib`` flags. Only including top level module names ------------------------------------- This PEP proposes that only top level module and package names be reported by the new query APIs. This is sufficient information to generate the proposed error messages, reduces the number of required entries by an order of magnitude, and simplifies the process of generating the related metadata during the build process. If this is eventually found to be overly limiting, a new ``include_submodules`` flag could be added to the query APIs. However, this is *not* part of the initial proposal, as the benefits of doing so aren't currently seen as justifying the extra complexity. There is one known consequence of this restriction, which is that the new default ``excepthook`` implementation will report incorrect submodules names the same way that it reports genuinely missing standard library submodules:: $ ./python -c "import unittest.muck" Traceback (most recent call last): File "", line 1, in ModuleNotFoundError: No submodule named 'unittest.muck' in standard library module 'unittest' Listing private top level module names as optional standard library modules --------------------------------------------------------------------------- Many of the modules that have an optional external build dependency are written as hybrid modules, where there is a shared Python wrapper around an implementation dependent interface to the underlying external library. In other cases, a private top level module may simply be a CPython implementation detail, and other implementations may not provide that module at all. To report import errors involving these modules appropriately, the new default ``excepthook`` implementation needs them to be reported by the new query APIs. Deeming packaging related modules to be mandatory ------------------------------------------------- Some redistributors aren't entirely keen on installing the Python specific packaging related modules (``distutils``, ``ensurepip``, ``venv``) by default, preferring that developers use their platform specific tooling instead. This approach causes interoperability problems for developers working on cross-platform projects and educators attempting to write platform independent setup instructions, so this PEP takes the view that these modules should be considered mandatory, and left out of the list of optional modules. Deferred Ideas ============== The ideas in this section are concepts that this PEP would potentially help enable, but they're considered out of scope for the initial proposal. Platform dependent modules -------------------------- Some standard library modules may be missing because they're only provided on particular platforms. For example, the ``winreg`` module is only available on Windows:: $ python3 -c "import winreg" Traceback (most recent call last): File "", line 1, in ModuleNotFoundError: No module named 'winreg' In the current proposal, these platform dependent modules will simply be included with all the other optional modules rather than attempting to expose the platform dependency information in a more structured way. However, the platform dependence is at least tracked at the level of "Windows", "Unix", "Linux", and "FreeBSD" for the benefit of `the documentation`_, so it seems plausible that it could potentially be exposed programmatically as well. .. _the documentation: https://docs.python.org/3/py-modindex.html Emitting a warning when ``__main__`` shadows a standard library module ---------------------------------------------------------------------- Given the new query APIs, the new default ``excepthook`` implementation could potentially detect when ``__main__.__file__`` or ``__main__.__spec__.name`` match a standard library module, and emit a suitable warning. However, actually doing anything along this lines should review more cases where uses actually encounter this problem, and the various options for potentially offering more information to assist in debugging the situation, rather than needing to be incorporated right now. Recommendation for Downstream Distributors ========================================== By patching `site.py`_ [*]_ to provide their own implementation of the `sys.excepthook`_ function, Python distributors can display tailor-made error messages for any uncaught exceptions, including informing the user of a proper, distro-specific way to install missing standard library modules upon encountering a `ModuleNotFoundError`_. Some downstream distributors are already using this method of patching `sys.excepthook` to integrate with platform crash reporting mechanisms. .. _`site.py`: https://docs.python.org/3.7/library/site.html .. _`sitecustomize.py`: `site.py`_ Backwards Compatibility ======================= No problems with backwards compatibility are expected. Distributions that are already patching Python modules to provide custom handling of missing dependencies can continue to do so unhindered. Reference and Example Implementation ==================================== TBD. The finer details will depend on what's practical given the capabilities of the CPython build system (other implementations should then be able to use the generated CPython data, rather than having to regenerate it themselves). Notes and References ==================== .. [*] Or `sitecustomize.py`_ for organizations with their own custom Python variant. .. [#debian-patch] http://bazaar.launchpad.net/~doko/python/pkg3.5-debian/view/head:/patches/tkinter-import.diff .. [#stackoverflow-stdlib] http://stackoverflow.com/questions/6463918/how-can-i-get-a-list-of-all-the-python-standard-library-modules Ideas leading up to this PEP were discussed on the `python-dev mailing list`_ and subsequently on `python-ideas`_. .. _`python-dev mailing list`: https://mail.python.org/pipermail/python-dev/2016-July/145534.html .. _`python-ideas`: https://mail.python.org/pipermail/python-ideas/2016-December/043907.html Copyright ========= This document has been placed in the public domain.