[Python-Dev] python-dev Summary for 2004-07-16 through 2004-07-31 [draft] (original) (raw)

Brett Cannon bac at OCF.Berkeley.EDU
Thu Aug 5 09:05:16 CEST 2004

OK, so I procrastinated from doing more work on my thesis and did the next summary. Plan to send this out some time over the weekend.

===================== Summary Announcements

Python 2.4a2 should be coming out any day now, so keep an eye out for it and test the heck out of it.

======== Summary

Makin' bdist_wininst purty and nicer for them non-English speaking folks

Thomas Heller asked if some better images for Distutil's bdist_wininst could be used. Some existing images were discussed, but no specific resolution was mentioned in the thread.

Walter Dörwald brought up the point that if this was all being worked on he wouldn't mind having Unicode support for things such as author name and such. Thomas and Walter worked on it and that has gone in.

Contributing threads:

Assigning to None a no-no, but still okay for booleans

Assigning to None is now a SyntaxError instead of a SyntaxWarning. Assignment to True and False, though, is still allowed and probably will not be restricted until Python 3.0 (which will probably have to wait until, if Guido's OSCON slides are to be believed, until he retires). The idea of restricting assignment, though, was brought up by Raymond Hettinger. Michael Hudson, though, pointed out it would be difficult and possibly require a special opcode to handle it.

Contributing threads:

'as' does not get to be treated with kiddie gloves

Philip Eby asked if 'as' was going to become a keyword. Guido said not now, but in Python 3.0 it will be.

Contributing threads:

Hopefully you didn't love FCNTL like a child...

... since if you did that kid was a lazy punk that did nothing but cause trouble for Windows users. That's why it got kicked out of the house to never come back.

Contributing threads:

How to get Python to compile with Microsoft's free compiler that should just come with the OS standard

  1. Download the free .NET compiler
  2. Download the Windows SDK (at http://www.microsoft.com/msdownload/platformsdk/sdkupdate/ with only IE)
  3. Download Garth's tools to generate a Makefile from the .sln files (at http://mail.python.org/pipermail/python-dev/2004-February/042595.html )
  4. Compile
  5. Realize you should be using an OS that doesn't make you go through this many hoops just to have a compiler for your platform

Contributing threads:

Tim Peters, whitespace Czar, will break your legs if you mess up the pristine whitespace in CVS

Tim Peters ran Tools/Scripts/reindent.py over the entire CVS tree and fixed it up. This means you had better not mess up and check in using anything but 4-space indents with no tabs!

Contributing threads:

Thread safety == won't blow up in your face

Paul Moore pointed out that the documentation for deques in the collections module stated they were thread-safe. It was clarified that the statement meant that internal state would not be corrupted if multiple threads accessed the same object; no guarantee that it isn't accessed in some order or anything. Basically the idea of thread-safety for C code is that it won't lead to the interpreter exploding, nothing more. And you don't even get that guarantee with Python code.

Contributing threads:

LC_NUMERIC PEP gets a pristine, new PEP number

Even though the code had already been put into the core, the PEP about LC_NUMERIC and being locale-independent never got a PEP number. Well, now it has one: PEP 331_.

.. _PEP 331: http://www.python.org/peps/pep-0331.html

Contributing threads:

Edward Loper becomes one of us rowdy developers

And he has already made his initial contribution by helping to rework doctest.

Contributing threads:

Any misbehaving modules during initialization from import now get the boot

Jim Fulton wanted a better way to detect when an import failed thanks to another module being directly imported (e.g., module A imports B which raised TypeError from some initialization code from importing; importing would leave A and B in a shoddy state in sys.modules along with raising TypeError in the import from A instead of raising ImportError). While the latter still occurs, modules are not left in sys.modules in a broken state from exceptions being raised during initialization thanks to Guido and Tim Peters.

There was a discussion on circular imports and how to handle those for proper rollback. Some suggestions were taking a snapshot of sys.modules and then restoring with that if something bad happens to putting in placeholder modules in sys.modules so as to not try to extra imports on modules in the process and to help track state. But this all gets sticky from side-effects that modules can do outside of themselves before they finish importing everyone else. If, for instance, module A, as a side-effect of importation, injected a custom version of len into module B. Now, if module A did this before doing all of its imports it could pull of the len injection but still fail from a bad import. That is not good. Basically the best solution is to not do that; there is a reason you should do all of your global imports as the first thing in a module.

Contributing threads:

Stuff about Unicode, state, and how to handle when a stream terminated early; all CJK to me

Walter Dörwald noticed that codecs.StreamReader.read() would read a few more bytes when it discovered an error. That's bad since there might not be more bytes and continuing once an error has been found is just not right. So he wanted to fix that problem.

Unfortunately he and MA Lemburg started to talk and I just couldn't follow everything about stateful and stateless decoders/encoders and the issues; I'm American so Unicode just doesn't fit in my brain well. So if you want to know what conclusions they reached you are going to read the thread on your own.

Contributing threads:

Use the docs to know what the public API is, people

Fernando Perez got bit by having something removed from the 'keyword' module that was not part of the documented API. Turns out that running 'help' on the module listed the function in question and so he just went ahead an used it. That's not the right way to go about finding the public API of stdlib code.

Always follow the documentation. If it is not listed there don't expect it to be there in the next version. While we try not to yank stuff out needlessly and will tweak things to be nice on occasion, we make no guarantees on stuff not listed in the API.

The next best thing is what is listed in all for a module. Since that is explicitly listed that usually can be considered part of the API.

And lastly, anything starting with an underscore is off limits in terms of promises of keeping it around.

Luckily Skip Montanaro applied a patch for pydoc to have it only list stuff in modules as specified by all, so it's a little safer out there. But the actual documentation is still king.

Contributing threads:

Linking against static builds on Windows

Thomas Heller pointed out that one couldn't dynamically link against a statically built Python interpreter DLL. He and Martin v. Löwis discussed various ways to try to get this to work, but it doesn't appear they came up with one.

Contributing threads:

More information about the Python-Dev mailing list