#+STARTUP: content
#+STARTUP: hidestars
* Pycon Keynote Notes
** Guido Keynote
  - Merging unladen swallow into 3 within a year? Python 3
  - Guido also hates Python packaging
  - Thinks programming languages will disappear eventually.  Replaced
    by something more natural
  - AppEngine?
  - greenlet/eventlet?
  - doesn't think python should be used for functional programming,
    admires f languages from a distance
  - django > turbogears
  - uses emacs.... and vi, got a lot of boos for the first one
  - web code review tools: codereview.appspot.com
* Pycon Talk Notes
** Large Astronomical Data Reduction and Analysis Systems
   - LSST
   - Uses SimPy & ctypes
   - Large Scale simulations
   - Uses MPI for pipeline communications (mpi4py)
   - Astronomy, open data and open source.  Free Data
** Ian Bicking: Packaging and Deployment
 - No one uses Python 3, everyone sees it as the future
   but there's no practical movement towards using it
 - standard library is fundamental, very stable
 - think about dependency tree as layers
 - analogy of streets and traffic to language
 - showing complex dependency graphs (*) (woonerf)
 - sqlalchemy, extremely complex... lots of lateral dependencies
 - "The Cloud" -> 'toppcloud'  Ian's new tool
 - cloud is Testable
 - from questions:
   - monwhiskey instead of paste
   - toppcloud has config management (vs. chef, puppet)
** Python 2 -> 3
 - "Regrets" and "Warts"
 - new string model (PEPs 358, 3112, 3137, 3138)
 - exception changes
 - division becomes true division ( 1/2 = 0.5 instead of 0 )
 - floor division (my safe division?
 - new hex, octable, etc. literals
 - iterations are much more efficient (3.0 has a mem consumption theme)
 - xrange becomes range
 - new i/o class type
 - methods for tuples, new set and dict comprehensions (setcomps, dictcomps)
** Laziness
 - lazy data via generators & iterators
 - giving examples in scheme and haskell
 - memoizers as a form of laziness
 - haskell is completely lazy, evaluation only happens when necessary
 - itertools combinatoric iterator tools, very helpful and under used
 - itertools replaces older non-iterative tools like map, slice, etc.
 - not for infinite sets/lists (unlike haskell)
 - weakref, memoization
 - promises in a directed acyclic graph
** PyPy
   - Seems to be a very effective JIT for Python
   - operations like *args **kwargs are poor from a performance persp.
     + Significantly faster
     + I wonder if this is the same for CPython
** Revisioned Databases
   - From EVE
   - They use stackless python
   - Strategies
     + YAML
     + Server Authoritative
     + Settled on Revisioned Database
   - Multi-layered
     + DB Tables
     + BSD(Branch-Static-Data) layer that keeps changes/revision
       + CCP Revisioned Database (views on a per-row op)
       + User can only have one row out and allowed to CUD
   - Branching <-> Syncing
   - Doc developers can continuously work on versions of documentation
   - PyObj holds references to each table, polls recent revisions
     + Handles update tasklets
     + Merges into single query
   - BSDTable class managed BSDRow data
   - BSDRow handles editing and merging (bucketing)
   - Switched to sets from lists and got a huge speed improvement (interesting)
   - Caching/filtering using SQLite
   - Can't support 2 way (loading/written) transactions
   - Good for document editing
   - MSSQL 2k8 :( TransactSQL
** Dealing with unsightly data
   - ElementSoup?
   - xml.handler.ContentHandler for switching between diff formats
   - mapping between different protocols
   - this guy is british and hard to understand
   - get permission when converting prop data to open format
   - Reverse Engineering
     + Corner cases
     + Source data stability
     + provider notification of changes
     + Tools (docs, firebug or equiv)
   - DS -> [Connector|API] -> Consumer
   - data/type coercion
   - Strict out, forgiving in
   - Tests on API
   - Similar to issues I have had with the Yahoo Fantasy Sports Page
     though I used BeautifulSoup + Mechanize
   - Z39.50 protocol?