=== Perry Greenfield - pyfits and pyraf ===

--- pyfits ---

>>> import pyfits
>>> pyfits.info('pix.fits')				# get info
>>> im = pyfits.getdata('py.fits)		# load

>>> import numdisplay
>>> numdisplay.display(im)				# view image with, e.g., ds9

>>> hdr = pyfits.getheader('pix.fits')	# FITS headers have keywords and values; load to memory
>>> print hdr['date']					# use as a dictionary
>>> hdr['date'] = '4th of July'
>>> hdr.update('newkey','value')		# but do this to add key

# lots more stuff in adding 'cards', writing data to file...

>>> tab = pyfits.getdata('alpha.fits')	# can also work with tables (FITS BinTableHDU rather than ImageHDU)
>>> tab.names							# fields (i.e., columns)
>>> tab.shape							# record will show as one element
>>> tab[0]								# first row
>>> flux = tab.field('flux')			# create numpy array as view into data
>>> wave[0] = 0							# can use view to change table

# there's also an object-oriented interface to files
>>> ff = pyfits.open('pix.fits')		# open file object
>>> sum = 0
>>> for hdu in ff[1:]:					# look over images in this file, skip header
		sum = sum + hdu.data
		hdu.data = None					# delete from memory to save space
>>> ff.close()

# can create new FITS tables starting from numpy arrays


=== David Cournapeau - numscons ===

- for users:
 	- customization, ease of use
- for developers:
	- dependency handling
	- fine-grained control of build
	- better configuration (libraries, platform)
	- new features (ctypes)
	- easier to understand
- numscons
	- 0.9.1 on pypi
	- builds numpy and scipy all libraries
	- supports libraries
	- can change CFLAGS, CC on the fly... numscons will recompile objects
	- can build in parallel
	- extends scons to build Python extensions and fortran
- scons
	- is python
	- has basic configuration system
	- each target can be built differently
	- is extensible
	- extensible, has good manual, good community
- scons basics
	- Builder's
		- SharedLibrary("foo.c")
		- StaticLibrary("foo.c")
		- Program("foo.c")
	- each object is customizable
	- dependency handling through DAG
		- detected automatically by scanning source files
	- Environment's
		- global object to keep configurations
		- builders attach to environments
	- checking for library dependencies
		- config = env.Configure()
		- config.CheckLib(...)
		- config.Finish()
	- should know about pkgconfig
- scons tools
	- e.g., a compiler
	- a python module with two public methods
	- sets up Environment values
	- complicated unfortunately...
- as a build system or numpy, scipy
	- uses scons for low level build
	- simple, clear separation between core and customization
	- packages, subpackages, 
- numscons basics
	- NumpyEnvironment extends NumpyEnvironment
	- need setup.py (with add_sconscript), SConstruct, SConscript (do the real work):
		- env.DistutilsPythonExtension("hello", source = [...])
		- env.NumpyPythonExtension
		- env.NumpyCtypesExtension
	- NumpyConfigure()
		- can generate a config.h header
	- Can use .in processor as in autoconf
		- env['SUBST_DICT] = {'@FOO1@': 'foo'}
		- env.SubstInFile('header.h','header.h.in')
		- for instance, use for f77 compiler mangling


=== Robert Bradshaw - Writing optimized extensions with Cython ===

- cython
	- almost a Python compiler
	- extended Python language for fast extensions, interfacing C/C++ libraries
	- not a dedicated wrapper like C
- a Nov 2006 fork of Pyrex
- efficient code
	- conditions and loops run 2-8x faster than Python
	- thanks to optional type declarations (Cython generates C rather than Python)
	- reads .pyx file and produces .c -> compiled into .so
- cython basics
	- cdef keyword (in the future, use decorators?)
	- cdef type var
	- cdef return-type function
- external functions
	- cdef extern from "math.h":
	  	double sin(double)
		double cos(double)
	- either link the library, or include the source code in the extension
	- put declarations in .pxd files; some already in Cython/Includes
- memory management
	- full in Python, not in C
	- Cython has both
- pointers
	- use int* for list of ints (malloced?)
		- cdef int* mylist = <int*>malloc(sizeof(int) * n)
		- access elements with [...]
	- but Python list is pretty fast
- numpy integration (Dag Seljebotn)
	- new interface works for any object with buffer protocol
	- cimport numpy
	- cdef numpy.ndarray[int,dim=1] arr = x
	- use @cython.boundscheck(False) from "cimport cython" before function to speed it up
	- make it faster also by using unsigned indices (perhaps with <unsigned int> casts)
- extension classes
	- can have c attributes and methods
	- but not dictionaries, multiple inheritance
	- need cpdef to call methods from python
- cython can wrap C++ somewhat
	- fool it by defining a structure
	- with special functions for new and delete