Differences between revisions 8 and 9
Revision 8 as of 2005-08-02 11:02:11
Size: 1264
Editor: dsl-082-082-051-041
Comment:
Revision 9 as of 2005-11-24 20:36:41
Size: 4458
Editor: Z-pc1-959-S1
Comment:
Deletions are marked like this. Additions are marked like this.
Line 3: Line 3:
== Implementation(s) == == Introduction ==
Line 5: Line 5:
Reinhold has modified Jason Orendorff's original path.py to fit discussions in python-dev and comp.lang.python and has placed it in Python CVS under nondist/sandbox/path. The Python Standard Library currently has many modules that allow
the developer to interact with the filesystem. Currently this
set of modules is:
    * os
    * os.path
    * fnmatch
    * glob
    * shutil
    * stat
Line 7: Line 15:
Other implementations to look at include the original, path.py and some things that others have mentioned on c.l.p. and python-dev. This PEP proposes that a new class or module be added to the
Python Standard Library that will make all filesystem operations
available from one place in a consistent way.

== Reference Implementation ==
Reinhold has modified Jason Orendorff's original path.py to fit
discussions in python-dev and comp.lang.python and has placed it in
Python under nondist/sandbox/path: http://svn.python.org/projects/sandbox/trunk/path/path.py

Other implementations to look at include the original, path.py and
some things that others have mentioned on c.l.p. and python-dev.

Jason Orendorff's original path.py can be found here: http://www.jorendorff.com/articles/python/path/

== Motivation ==

The motivation for a single standard class or module to handle
filesystem operations is threefold:
    * Reduce the number of modules a user must import and refer to
    to do common operations

    * Aesthetics: code written with the reference implementation is more
    concise, easier to read, and also easier to write

    * Consistency: All filesystem operations should be accessed from
    within the same module.
Line 11: Line 44:
1. A Path should be a drop-in replacement for a str or unicode as much as possible. 1. A Path should be a drop-in replacement for a str or unicode as much
as possible.
Line 13: Line 47:
2. Properties are the interface for actual attributes of a Path. For example, a Path's basename will be the same regardless of data on the local filesystem. Accessing a Path property will never result in an IOError. 2. Properties are the interface for actual attributes of a Path. For
example, a Path's basename will be the same regardless of data on the
local filesystem. Accessing a Path property will never result in an
IOError.
Line 15: Line 52:
3. Methods are the interface for attributes of whatever the Path represents, for example, the last-modified-time or contents of a file. Calling a Path method may result in an IOError if the method accesses the actual filesystem and finds a problem (e.g. the Path refers to a nonexistent file). 3. Methods are the interface for attributes of whatever the Path
represents, for example, the last-modified-time or contents of a file.
Calling a Path method may result in an IOError if the method accesses
the actual filesystem and finds a problem (e.g. the Path refers to a
nonexistent file).
Line 17: Line 58:
4. [Done in CVS] Should be subclassable so third parties can offer richer subclasses. Use {{{self.__class__()}}} instead of {{{Path()}}} in method bodies. Alternate constructors like {{{Path.cwd()}}} should be a class methods rather than static methods. (MikeOrr) 4. [Done in CVS] Should be subclassable so third parties can offer
richer subclasses. Use {{{self.__class__()}}} instead of {{{Path()}}}
in method bodies. Alternate constructors like {{{Path.cwd()}}} should
be a class methods rather than static methods. (MikeOrr)

== Backwards Compatibility ==

If this PEP is accepted, then several of the existing standard
modules will become redundant, which violates the ["TOOWTDI"]
principal. The following modules will become redundant:
    * os.path
    * shutils
    * fnmatch
    * glob
    * stat
    * parts of os (like mkdir)

It is proposed that Python 2.5 will mark redundant modules as
deprecated and issue a warning when they are imported. The
functionality that these modules offer should be moved into the
path module.

Python 2.6 will remove the redundant modules from the standard
library.

== Open Issues ==
    
    * What to call this module / class, and where to put it?

    * API issues with reference implementation:
        * Remove duplicate functionality:
            * .joinpath / .joinwith

        * Property / method consistency:
            * .parent -> .dirname (and get rid of .dirname())
            * .name -> .filename (and get rid of .basename())
            * .namebase -> .filebase
            * .atime/.mtime/.ctime -> .atime()/.mtime()/.ctime()
            * .size -> .filesize()

        * .splitall() -> .parts()
        * bytes() -> get_bytes()
        * write_bytes() -> set_bytes() / append_bytes()
        * text() -> get_text()
        * write_text() -> set_text() / append_text()
        * lines() -> get_lines()
        * write_lines() -> set_lines() / append_lines()

        * .joinpath(*args) -> .subpath(*args)?
            * This needs consensus, I prefer .joinpath()

        * .listdir() -> .subpaths() OR .listpaths()

        * drop .getcwd()?
            * default constructor could assign absolute cwd, or "."?

        * Should .mtime()/etc. return datetime objects?
            * I vote no, timestamps are fine

== Discussions ==

    * http://sourceforge.net/tracker/index.php?func=detail&aid=1226256&group_id=5470&atid=355470
    * http://thread.gmane.org/gmane.comp.python.devel/69403
    * http://mail.python.org/pipermail/python-dev/2005-June/054438.html
    * http://groups.google.ca/group/comp.lang.python/msg/822a302350658a01

Path class Pre-Pre-PEP

Introduction

The Python Standard Library currently has many modules that allow the developer to interact with the filesystem. Currently this set of modules is:

  • os
  • os.path
  • fnmatch
  • glob
  • shutil
  • stat

This PEP proposes that a new class or module be added to the Python Standard Library that will make all filesystem operations available from one place in a consistent way.

Reference Implementation

Reinhold has modified Jason Orendorff's original path.py to fit discussions in python-dev and comp.lang.python and has placed it in Python under nondist/sandbox/path: http://svn.python.org/projects/sandbox/trunk/path/path.py

Other implementations to look at include the original, path.py and some things that others have mentioned on c.l.p. and python-dev.

Jason Orendorff's original path.py can be found here: http://www.jorendorff.com/articles/python/path/

Motivation

The motivation for a single standard class or module to handle filesystem operations is threefold:

  • Reduce the number of modules a user must import and refer to to do common operations
  • Aesthetics: code written with the reference implementation is more concise, easier to read, and also easier to write
  • Consistency: All filesystem operations should be accessed from within the same module.

Design Principles

1. A Path should be a drop-in replacement for a str or unicode as much as possible.

2. Properties are the interface for actual attributes of a Path. For example, a Path's basename will be the same regardless of data on the local filesystem. Accessing a Path property will never result in an IOError.

3. Methods are the interface for attributes of whatever the Path represents, for example, the last-modified-time or contents of a file. Calling a Path method may result in an IOError if the method accesses the actual filesystem and finds a problem (e.g. the Path refers to a nonexistent file).

4. [Done in CVS] Should be subclassable so third parties can offer richer subclasses. Use self.__class__() instead of Path() in method bodies. Alternate constructors like Path.cwd() should be a class methods rather than static methods. (MikeOrr)

Backwards Compatibility

If this PEP is accepted, then several of the existing standard modules will become redundant, which violates the ["TOOWTDI"] principal. The following modules will become redundant:

  • os.path
  • shutils
  • fnmatch
  • glob
  • stat
  • parts of os (like mkdir)

It is proposed that Python 2.5 will mark redundant modules as deprecated and issue a warning when they are imported. The functionality that these modules offer should be moved into the path module.

Python 2.6 will remove the redundant modules from the standard library.

Open Issues

  • What to call this module / class, and where to put it?
  • API issues with reference implementation:
    • Remove duplicate functionality:
      • .joinpath / .joinwith
    • Property / method consistency:
      • .parent -> .dirname (and get rid of .dirname())

      • .name -> .filename (and get rid of .basename())

      • .namebase -> .filebase

      • .atime/.mtime/.ctime -> .atime()/.mtime()/.ctime()

      • .size -> .filesize()

    • .splitall() -> .parts()

    • bytes() -> get_bytes()

    • write_bytes() -> set_bytes() / append_bytes()

    • text() -> get_text()

    • write_text() -> set_text() / append_text()

    • lines() -> get_lines()

    • write_lines() -> set_lines() / append_lines()

    • .joinpath(*args) -> .subpath(*args)?

      • This needs consensus, I prefer .joinpath()
    • .listdir() -> .subpaths() OR .listpaths()

    • drop .getcwd()?
      • default constructor could assign absolute cwd, or "."?
    • Should .mtime()/etc. return datetime objects?
      • I vote no, timestamps are fine

Discussions

PathClass (last edited 2009-09-13 06:07:15 by 216)

Unable to edit the page? See the FrontPage for instructions.