Title: Feature Requests
Author: Jeremy Hylton <email@example.com>
This PEP contains a list of feature requests that may be
considered for future versions of Python. Large feature requests
should not be included here, but should be described in separate
PEPs; however a large feature request that doesn't have its own
PEP can be listed here until its own PEP is created. See
PEP 0 for details.
This PEP was created to allow us to close bug reports that are really
feature requests. Marked as Open, they distract from the list of real
bugs (which should ideally be less than a page). Marked as Closed, they
tend to be forgotten. The procedure now is: if a bug report is really
a feature request, add the feature request to this PEP; mark the bug as
"feature request", "later", and "closed"; and add a comment to the bug
saying that this is the case (mentioning the PEP explicitly). It is
also acceptable to move large feature requests directly from the bugs
database to a separate PEP.
This PEP should really be separated into four different categories
(categories due to Laura Creighton):
1. BDFL rejects as a bad idea. Don't come back with it.
2. BDFL will put in if somebody writes the code. (Or at any rate,
BDFL will say 'change this and I will put it in' if you show up
(possibly divided into:
2a) BDFL would really like to see some code!
2b) BDFL is never going to be enthusiastic about this, but
will work it in when it's easy.
3. If you show up with code, BDFL will make a pronouncement. It
might be ICK.
4. This is too vague. This is rejected, but only on the grounds
of vagueness. If you like this enhancement, make a new PEP.
Core Language / Builtins
- The parser should handle more deeply nested parse trees.
The following will fail -- eval("["*50 + "]"*50) -- because the
parser has a hard-coded limit on stack size. This limit should
be raised or removed. Removal would be hard because the
current compiler can overflow the C stack if the nesting is too
- Non-accidental IEEE-754 support (Infs, NaNs, settable traps, etc).
- Windows: Trying to create (or even access) files with certain magic
names can hang or crash Windows systems. This is really a bug in the
OSes, but some apps try to shield users from it. When it happens,
the symptoms are very confusing.
Hang using files named prn.txt, etc
- eval and free variables: It might be useful if there was a way
to pass bindings for free variables to eval when a code object
with free variables is passed.
- The urllib module should support proxies which require
authentication. See SourceForge bug #210619 for information:
- os.rename() should be modified to handle EXDEV errors on
platforms that don't allow rename() to operate across filesystem
boundaries by copying the file over and removing the original.
Linux is one system that requires this treatment.
- signal handling doesn't always work as expected. E.g. if
sys.stdin.readline() is interrupted by a (returning) signal
handler, it returns "". It would be better to make it raise an
exception (corresponding to EINTR) or to restart. But these
changes would have to applied to all places that can do blocking
interruptable I/O. So it's a big project.
- Extend Windows utime to accept directory paths.
- Extend copy.py to module & function types.
- Better checking for bad input to marshal.load*().
- rfc822.py should be more lenient than the spec in the types of
address fields it parses. Specifically, an invalid address of
the form "From: Amazon.com <firstname.lastname@example.org>" should
be parsed correctly.
- cgi.py's FieldStorage class should be more conservative with
memory in the face of large binary file uploads.
There are two issues here: first, because
read_lines_to_outerboundary() uses readline() it is possible
that a large amount of data will be read into memory for a
binary file upload. This should probably look at the
Content-Type header of the section and do a chunked read if it's
a binary type.
The second issue was related to the self.lines attribute, which
was removed in revision 1.56 of cgi.py (see also):
- urllib should support proxy definitions that contain just the
host and port
- urlparse should be updated to comply with RFC 2396, which
defines optional parameters for each segment of the path.
- The exceptions raised by pickle and cPickle are currently
different; these should be unified (probably the exceptions
should be defined in a helper module that's imported by both).
[No bug report; I just thought of this.]
- More standard library routines should support Unicode. For
example, urllib.quote() could convert Unicode strings to UTF-8
and then do the usual %HH conversion. But this is not the only
- There should be a way to say that you don't mind if str() or
__str__() return a Unicode string object. Or a different
function -- ustr() has been proposed. Or something...
- Killing a thread from another thread. Or maybe sending a
signal. Or maybe raising an asynchronous exception.
- The debugger (pdb) should understand packages.
- Jim Fulton suggested the following:
I wonder if it would be a good idea to have a new kind of
temporary file that stored data in memory unless:
- The data exceeds some size, or
- Somebody asks for a fileno.
Then the cgi module (and other apps) could use this thing in a
- Jim Fulton pointed out that binascii's b2a_base64() function
has situations where it makes sense not to append a newline,
or to append something else than a newline.
- add an optional argument giving the delimiter string to be
appended, defaulting to "\n"
- possibly special-case None as the delimiter string to avoid
adding the pad bytes too???
- pydoc should be integrated with the HTML docs, or at least
be able to link to them.
- Distutils should deduce dependencies for .c and .h files.
- asynchat is buggy in the face of multithreading.
- It would be nice if the higher level modules (httplib, smtplib,
nntplib, etc.) had options for setting socket timeouts.
- The curses library is missing two important calls: newterm() and
- It would be nice if the built-in SSL socket type could be used
for non-blocking SSL I/O. Currently packages such as Twisted
which implement async servers using SSL have to require third-party
packages such as pyopenssl.
- reST as a standard library module
- The import lock could use some redesign.
- A nicer API to open text files, replacing the ugly (in some
people's eyes) "U" mode flag. There's a proposal out there to
have a new built-in type textfile(filename, mode, encoding).
(Shouldn't it have a bufsize argument too?)
- Support new widgets and/or parameters for Tkinter
- For a class defined inside another class, the __name__ should be
"outer.inner", and pickling should work. (GvR is no longer certain
this is easy or even right.)
- Decide on a clearer deprecation policy (especially for modules)
and act on it.
- Provide alternatives for common uses of the types module;
Skip Montanaro has posted a proto-PEP for this idea:
- Use pending deprecation for the types and string modules. This
requires providing alternatives for the parts that aren't
covered yet (e.g. string.whitespace and types.TracebackType).
It seems we can't get consensus on this.
- Lazily tracking tuples?
- Make 'as' a keyword. It has been a pseudo-keyword long enough.
(It's deprecated in 2.5, and will become a keyword in 2.6.)
C API wishes
- Add C API functions to help Windows users who are building
embedded applications where the FILE * structure does not match
the FILE * the interpreter was compiled with.
See this bug report for a specific suggestion that will allow a
Borland C++ builder application to interact with a python.dll
build with MSVC.
- Python could use a GUI builder.
Building and Installing
- Modules/makesetup should make sure the 'config.c' file it
generates from the various Setup files, is valid C. It currently
accepts module names with characters that are not allowable in
Python or C identifiers.
- Building from source should not attempt to overwrite the
Include/graminit.h and Parser/graminit.c files, at least for
people downloading a source release rather than working from
Subversion or snapshots. Some people find this a problem in unusual
- The configure script has probably grown a bit crufty with age and may
not track autoconf's more recent features very well. It should be
looked at and possibly cleaned up.
- Make Python compliant to the FHS (the Filesystem Hierarchy Standard)