From 1904f0a2245f500aa85fba347b260620350efc78 Mon Sep 17 00:00:00 2001 From: Barney Gale Date: Fri, 15 Mar 2024 00:11:49 +0000 Subject: GH-113838: Add "Comparison to os.path" section to pathlib docs (#115926) --- Doc/library/pathlib.rst | 63 ++++++++++++++++++++++++++++++++++--------------- 1 file changed, 44 insertions(+), 19 deletions(-) diff --git a/Doc/library/pathlib.rst b/Doc/library/pathlib.rst index 9041f37..3ff2631 100644 --- a/Doc/library/pathlib.rst +++ b/Doc/library/pathlib.rst @@ -1664,7 +1664,7 @@ the pattern to match only directories. Comparison to the :mod:`glob` module -^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +------------------------------------ The patterns accepted and results generated by :meth:`Path.glob` and :meth:`Path.rglob` differ slightly from those by the :mod:`glob` module: @@ -1682,27 +1682,57 @@ The patterns accepted and results generated by :meth:`Path.glob` and 5. The values returned from pathlib's ``path.glob()`` and ``path.rglob()`` include the *path* as a prefix, unlike the results of ``glob.glob(root_dir=path)``. -6. ``bytes``-based paths and :ref:`paths relative to directory descriptors - ` are not supported by pathlib. -Correspondence to tools in the :mod:`os` module ------------------------------------------------ +Comparison to the :mod:`os` and :mod:`os.path` modules +------------------------------------------------------ -Below is a table mapping various :mod:`os` functions to their corresponding -:class:`PurePath`/:class:`Path` equivalent. +pathlib implements path operations using :class:`PurePath` and :class:`Path` +objects, and so it's said to be *object-oriented*. On the other hand, the +:mod:`os` and :mod:`os.path` modules supply functions that work with low-level +``str`` and ``bytes`` objects, which is a more *procedural* approach. Some +users consider the object-oriented style to be more readable. -.. note:: +Many functions in :mod:`os` and :mod:`os.path` support ``bytes`` paths and +:ref:`paths relative to directory descriptors `. These features aren't +available in pathlib. + +Python's ``str`` and ``bytes`` types, and portions of the :mod:`os` and +:mod:`os.path` modules, are written in C and are very speedy. pathlib is +written in pure Python and is often slower, but rarely slow enough to matter. + +pathlib's path normalization is slightly more opinionated and consistent than +:mod:`os.path`. For example, whereas :func:`os.path.abspath` eliminates +"``..``" segments from a path, which may change its meaning if symlinks are +involved, :meth:`Path.absolute` preserves these segments for greater safety. + +pathlib's path normalization may render it unsuitable for some applications: + +1. pathlib normalizes ``Path("my_folder/")`` to ``Path("my_folder")``, which + changes a path's meaning when supplied to various operating system APIs and + command-line utilities. Specifically, the absence of a trailing separator + may allow the path to be resolved as either a file or directory, rather + than a directory only. +2. pathlib normalizes ``Path("./my_program")`` to ``Path("my_program")``, + which changes a path's meaning when used as an executable search path, such + as in a shell or when spawning a child process. Specifically, the absence + of a separator in the path may force it to be looked up in :envvar:`PATH` + rather than the current directory. - Not all pairs of functions/methods below are equivalent. Some of them, - despite having some overlapping use-cases, have different semantics. They - include :func:`os.path.abspath` and :meth:`Path.absolute`, - :func:`os.path.relpath` and :meth:`PurePath.relative_to`. +As a consequence of these differences, pathlib is not a drop-in replacement +for :mod:`os.path`. + + +Corresponding tools +^^^^^^^^^^^^^^^^^^^ + +Below is a table mapping various :mod:`os` functions to their corresponding +:class:`PurePath`/:class:`Path` equivalent. ==================================== ============================== :mod:`os` and :mod:`os.path` :mod:`pathlib` ==================================== ============================== -:func:`os.path.abspath` :meth:`Path.absolute` [#]_ +:func:`os.path.abspath` :meth:`Path.absolute` :func:`os.path.realpath` :meth:`Path.resolve` :func:`os.chmod` :meth:`Path.chmod` :func:`os.mkdir` :meth:`Path.mkdir` @@ -1723,7 +1753,7 @@ Below is a table mapping various :mod:`os` functions to their corresponding :func:`os.link` :meth:`Path.hardlink_to` :func:`os.symlink` :meth:`Path.symlink_to` :func:`os.readlink` :meth:`Path.readlink` -:func:`os.path.relpath` :meth:`PurePath.relative_to` [#]_ +:func:`os.path.relpath` :meth:`PurePath.relative_to` :func:`os.stat` :meth:`Path.stat`, :meth:`Path.owner`, :meth:`Path.group` @@ -1735,8 +1765,3 @@ Below is a table mapping various :mod:`os` functions to their corresponding :func:`os.path.splitext` :attr:`PurePath.stem` and :attr:`PurePath.suffix` ==================================== ============================== - -.. rubric:: Footnotes - -.. [#] :func:`os.path.abspath` normalizes the resulting path, which may change its meaning in the presence of symlinks, while :meth:`Path.absolute` does not. -.. [#] :meth:`PurePath.relative_to` requires ``self`` to be the subpath of the argument, but :func:`os.path.relpath` does not. -- cgit v0.12