Donnerstag, 20. August 2009

Microsoft gives MSDN Premium subscription to PSF members

This almost went through unnoticed. Steve Holden (Python Software Foundation) has worked out a fantastic deal with Sam Ramji and Tom Hanrahan (both leading Open Source guys at Microsoft). Microsoft has given fourteen MSDN Premium subscriptions to Python core developers and PSF members. I'm one of the lucky few. [1], [2]

The premium subscription includes licenses and downloads of almost every Microsoft product from MS-DOS 6.22 to Windows 7, all versions of Visual Studio and many more stuff. This is very useful for Python core developers. Every developer with a subscription can finally set up multiple virtual boxes with 32 and 64bit versions of XP, Vista and Windows 7 to test and debug issues. 64bit versions of Windows were hard and costly to come by.

I'll keep my Ubuntu boxes for daily work and I'll still be skeptical about Microsoft's open source politics. However I'm glad that their paradigm towards Open Source is changing into the right direction. Python (more precisely IronPython) is going to become more important to Microsoft. I'll put my subscription into good use.

Thanks to Sam, Tom and Steve!

[1] http://mail.python.org/pipermail/python-dev/2009-July/090704.html
[2] http://mail.python.org/pipermail/python-dev/2009-August/091020.html

Donnerstag, 13. August 2009

How to add a new module search path

Once in a while Python users are asking how to add some directories to sys.path permanently. Usually a solution like the PYTHONPATH env variable are suggested to the op. Other solutions require root privileges or modify the search path for all users. PEP 370 adds another way that is more clean and easy to use. It doesn't require root privileges and it doesn't suffer from other issues. PYTHONPATH causes trouble for multiple Python versions. C extensions only work for one version of Python, most Python modules won't work on Python 2 and 3.

My preferred way adds additional search pathes just for one version of Python and just for me. It uses a .pth file as explained in the site module manual. .pth files only work in site-packages directories, either the global or the user specific directories.

The Python way

$ python2.6
>>> import os
>>> import site
>>> site.USER_SITE

'/home/heimes/.local/lib/python2.6/site-packages'

Create the directory if it doesn't exist yet

>>> if not os.path.isdir(site.USER_SITE):
... os.makedirs(site.USER_SITE)
...

mypath.pth is going to contain my list of addition search path

>>> mypth = os.path.join(site.USER_SITE, "mypath.pth")

>>> path_to_add = ["/home/heimes/modules", "/home/heimes/other_modules"]

Add a list of search paths line by line, also make sure we end with an empty line

>>> with open(mypth, "a") as f:

... f.write("\n".join(path_to_add))
... f.write("\n")
...
>>>

The bash way

$ python2.6 -m site --user-site
/home/heimes/.local/lib/python2.6/site-packages
$ mkdir -p $(python2.6 -m site --user-site)
$ echo "/home/heimes/more_modules" >> $(python2.6 -m site --user-site)/mypath.pth

Let's check if it works

check the pth file

$ cat $(python2.6 -m site --user-site)/mypath.pth

/home/heimes/modules
/home/heimes/other_modules
/home/heimes/more_modules

Let's see if the modules are in the new search path ... they aren't because the directories don't exist yet.

$ python2.6 -m site
sys.path = [
'/home/heimes',
'/usr/lib/python2.6',
'/usr/lib/python2.6/plat-linux2',
'/usr/lib/python2.6/lib-tk',
'/usr/lib/python2.6/lib-old',
'/usr/lib/python2.6/lib-dynload',
'/home/heimes/.local/lib/python2.6/site-packages',
'/usr/lib/python2.6/dist-packages',
'/usr/lib/python2.6/dist-packages/PIL',
'/usr/lib/python2.6/dist-packages/gst-0.10',
'/var/lib/python-support/python2.6',
'/usr/lib/python2.6/dist-packages/gtk-2.0',
'/var/lib/python-support/python2.6/gtk-2.0',
'/var/lib/python-support/python2.6/pyinotify',
'/usr/lib/python2.6/dist-packages/wx-2.6-gtk2-unicode',
'/usr/local/lib/python2.6/dist-packages',
]
USER_BASE: '/home/heimes/.local' (exists)
USER_SITE: '/home/heimes/.local/lib/python2.6/site-packages' (exists)
ENABLE_USER_SITE: True

create one example directory

$ mkdir /home/heimes/modules

$ python2.6 -m site
sys.path = [
'/home/heimes',
'/usr/lib/python2.6',
'/usr/lib/python2.6/plat-linux2',
'/usr/lib/python2.6/lib-tk',
'/usr/lib/python2.6/lib-old',
'/usr/lib/python2.6/lib-dynload',
'/home/heimes/.local/lib/python2.6/site-packages',
'/home/heimes/modules',
'/usr/lib/python2.6/dist-packages',
'/usr/lib/python2.6/dist-packages/PIL',
'/usr/lib/python2.6/dist-packages/gst-0.10',
'/var/lib/python-support/python2.6',
'/usr/lib/python2.6/dist-packages/gtk-2.0',
'/var/lib/python-support/python2.6/gtk-2.0',
'/var/lib/python-support/python2.6/pyinotify',
'/usr/lib/python2.6/dist-packages/wx-2.6-gtk2-unicode',
'/usr/local/lib/python2.6/dist-packages',
]
USER_BASE: '/home/heimes/.local' (exists)
USER_SITE: '/home/heimes/.local/lib/python2.6/site-packages' (exists)
ENABLE_USER_SITE: True

Easy, isnt' it?

Mittwoch, 12. August 2009

libxml2 crash on 64bit Ubuntu

I've spent the last couple of hours debugging a really strange segfault. Our application stack had a reproduceable crash in libxml2 -- but only with self compiled versions of libxml2. Ubuntu's 2.6.32 worked like a charm, my self compiled 2.6.32 didn't. The very same version works on several other Debian, Redhat and SuSE boxes, 32 and 64bit, too. WTF!?

The crash always occured in xmlIO.c:__xmlParserInputBufferCreateFilename() with xmlGzfileOpen() as open handler. After several gdb debugging sessions and several recompiles I noticed a suspicious message in the make output:

xmlIO.c: In function 'xmlGzfileOpen_real':
xmlIO.c:1132: warning: implicit declaration of function 'gzopen64'
xmlIO.c:1132: warning: nested extern declaration of 'gzopen64'
xmlIO.c:1132: warning: assignment makes pointer from integer without a cast
xmlIO.c: In function 'xmlGzfileOpenW':
xmlIO.c:1200: warning: assignment makes pointer from integer without a cast

The message only occured during my own compiles but not during "apt-get source -b libxml2" . Apparently Ubuntu has patched the sources to fix the issue. The changelog contains yet another hint:

* libxml.h: define _LARGEFILE64_SOURCE to properly get gzopen64 defines in zlib.h. Closes: #439843. Thanks Dann Frazier.

That's the solution to my problem! CFLAGS="-D_LARGEFILE64_SOURCE" ./configure and both the compiler warning and the crash is gone.