|Title:||Procedure for reviewing and improving standard library modules|
|Last-Modified:||2011-01-18 18:46:34 +0000 (Tue, 18 Jan 2011)|
|Author:||Georg Brandl <georg at python.org>|
This PEP describes a procedure for reviewing and improving standard library modules, especially those written in Python, making them ready for Python 3000. There can be different steps of refurbishing, each of which is described in a section below. Of course, not every step has to be performed for every module.
All modules marked as deprecated in 2.x versions should be removed for Python 3000. The same applies to modules which are seen as obsolete today, but are too widely used to be deprecated or removed. Python 3000 is the big occasion to get rid of them.
There will have to be a document listing all removed modules, together with information on possible substitutes or alternatives. This infor- mation will also have to be provided by the python3warn.py porting helper script mentioned in PEP XXX.
There are proposals for a "great stdlib renaming" introducing a hierarchic library namespace or a top-level package from which to import standard modules. That possibility aside, some modules' names are known to have been chosen unwisely, a mistake which could never be corrected in the 2.x series. Examples are names like "StringIO" or "Cookie". For Python 3000, there will be the possibility to give those modules less confusing and more conforming names.
Of course, each rename will have to be stated in the documentation of the respective module and perhaps in the global document of Step 1. Additionally, the python3warn.py script will recognize the old module names and notify the user accordingly.
If the name change is made in time for another release of the Python 2.x series, it is worth considering to introduce the new name in the 2.x branch to ease transition.
As most library modules written in Python have not been touched except for bug fixes, following the policy of never changing a running system, many of them may contain code that is not up to the newest language features and could be rewritten in a more concise, modern Python.
PyChecker should run cleanly over the library. With a carefully tuned configuration file, PyLint should also emit as few warnings as possible.
As long as these changes don't change the module's interface and behavior, no documentation updates are necessary.
Code coverage by unit tests varies greatly between modules. Each test suite should be checked for completeness, and the remaining classic tests should be converted to PyUnit (or whatever new shiny testing framework comes with Python 3000, perhaps py.test?).
It should also be verified that each publicly visible function has a meaningful docstring which ideally contains several doctests.
No documentation changes are necessary for enhancing test coverage.
This is a small and probably not very important step. There have been various attempts at providing author, version and similar metadata in modules (such as a "__version__" global). Those could be standardized and used throughout the library.
No documentation changes are necessary for this step, too.
Over the years, many bug reports have been filed which complained about bugs in standard library modules, but have subsequently been closed as "Won't fix" since a fix would have introduced a major incompatibility which was not acceptable in the Python 2.x series. In Python 3000, the fix can be applied if the interface per se is still acceptable.
Each slight behavioral change caused by such fixes must be mentioned in the documentation, perhaps in a "Changed in Version 3.0" paragraph.
The last and most disruptive change is the overhaul of a module's public interface. If a module's interface is to be changed, a justification should be made beforehand, or a PEP should be written.
The change must be fully documented as "New in Version 3.0", and the python3warn.py script must know about it.
This document has been placed in the public domain.