Insight VR

James Savage at MacWorld took a look at Battle for Vesta today. It came away with 3 out of 5 mice in what I think was a very fair and accurate review. As positives he cites intuitive controls, “buttery smooth” graphics, and solid space combat. As cons he lists pretty much what I myself think are the problems: need more variety in enemies, upgradable ship, and multiplayer.

I can say right now that head to head multiplayer isn’t happening, but all the rest is coming eventually. I do wish he had used a more compelling screenshot, but the game is so fast that it is really hard to generate your own while playing.

Jeff LaMarche

GameKit is a provided framework that only works over bluetooth. Some hardware doesn’t have support for this functionality.

Handles peer discovery and connection.

Three modes: Server, Client, and Peer. Peer is the friendliest to use.

Important to set delegates to nil prior to releasing the associated object.

Very easy to send data to all peers. You can send to specific peers as well.

Data always arrives intact so one send results in one receive.

Bluetooth is a very easy way to have multiplayer.

Online play is more difficult.

Apple doesn’t want you to use threads for networking. Use the UIApplication event loop instead.

NSStream doesn’t have a concept of discrete data. So sends don’t equal receives.

Use Bonjour to be able to find peers.

http://iphonedevbook.com/ has code for this networking stuff.

Drew McCormack wrote Sumo Master as a hobby project and to learn Core Animation.

Pretty simple top down 2d game.

Iterative design process. Accelerometer control didn’t work out. Went with touch.

Wrinklypea.com does design work, has some really nice looking UI and character design work.

For many things you can let Apple handle the animation for you. Views, buttons, help system, etc. Use modal transition styles to control how view controllers come in and out of view.

You can apply CA animations to UI elements in order to spiff up your UI.

Trick for squashing sumos: move the origin from the center of the sumo to the rear and then scale in along the y.

Performance tips: reduce the number of layers, merge layers that you can, make layers smaller, opaque layers are faster, cache CATransform3D objects.

http://bit.ly/sumophysics for info on the physics of the game.

Mark Aufflick: Notifications.

http://mark.aufflick.com/talks/apns

Limited to 256 bytes. Register device with Apple and then pass token back to your server. Then your server can use the token to send a message to Apple which will be passed to the device.

The notification itself is a binary format. 256 byte limit is payload only. Payload is json formatted.

If a user removes the app you get a notification.

There is a Perl library available that sends notifications.

Separate Sandbox(test) and Production connections for push notifications.

http://github.com/aufflick/p5-net-apns-persistent

Justin Williams from Second Gear Software

Sold off his iPhone business and is focusing on Mac products.

Hard to find a marketing niche in the app store.

Had to find a buyer for his app. Complained enough on twitter that someone approached him asking if he was serious and wanted to sell it. Sold it for 2 to 3 years worth of revenue for the app.

Once you agree upon a price you might want to get lawyers involved. Then sign papers, get money.

Transferring ownership of the application is likely impossible. You have to contact World Wide Developer Relations. They email you a list of four questions. Then they don’t respond. After 90 days got unofficial notification that unless you are selling the whole company they won’t do the transfer.

To ease the pain of the transfer the new owner offered the apps for free for a week, hoping that old users would figure it out and update to the new app.

A workaround might be to create an LLC for each app in case you want to sell it. Very annoying. Apple is not talkative on this topic.

Rod Strougo of Prop.gr – Cocos2D

Wraps a bunch of more complicated technologies for making games.

Two physics engines. Box2D is C++. Squirrel is C.

Box2D is tuned for 1 meter objects, so scale your objects to that.

Cocos2d is at http://www.cocos2d-iphone.org which comes with Box2d, but you can get Box2d from Google Code as well.

Looks like a cool set of libs to quickly build 2d games.

Marcus Zarra

ZSynch – desktop and iPhone data synchronization

CoreData is the best way to persist data on the iPhone.

Bare minimum for using ZSynch is two calls: set up a delegate and request a synch.

Eventually will synch to the cloud. You’ll be able to run your own server or pay for central hosting.

BSD license.

Jeff LaMarche

http://iphonedevelopment.blogspot.com

A bit of OpenGL history.

OpenGL ES removes parts that were redundant or killed performance. This includes direct mode, which was great for learing the API. Now you have to jump into the deep end.

Static inline functions for small bits of logic that are called frequently.

Avoid allocating memory a lot. Reuse memory between loop runs when possible.

Use the “f” versions of math.h functions so sqrtf() instead of sqrt() which expects and returns doubles.

Make universal apps to enable Thumb for Armv7 and disable it for Armv6.

Default is counter-clockwise winding.

Nitin Madnani

Python is well suited to NLP due to nicode support, C/C++ extensibility, etc.

NLTK comes with its own corpora, lots of tools, and WordNet integration. Has its own O’Reilly book.

Dumbo is Python bindings for Hadoop Streaming. Hadoop Streaming lets you use any executable or script for mappers and reducers.

Word association example is trivially parallelized using Hadoop on EC2.

Henrique Bastos – @henriquebastos

Small Acts are only the essential things.

Python 2008 was hosted in Rio. They had 400 people over 3 days. Henrique sent out an email asking for more information. The python population in Brazil is young, majority under 30, tend to be students, use it for web development (mostly Django) and consider Python essential to the work they do.

http://pythoncampus.org/ does tutorials and talks on university campuses in Brazil. Did 4 sessions last year. Will do one a month this year.

Rio Pythonistas run dojos weekly. Team coding in front of an audience with a projector showing their work.

Horaextra (Overtime) is a weekly social hour for technology enthusiasts.

Python enthusiasts attend conferences for other languages to meet other developers and build community.

They gave a Python t-shirt to President Lula, video of this became a hit at conferences and on YouTube.

SmallActsManifesto.org lists the principles used to build the community in Brazil.

Raymond Hettinger

Deque is like a list. Pronounced “deck” and stands for double ended queue. Can append() and pop() at both ends, can be indexed but not efficiently, no insert. Basically efficient on the ends. In collections.deque().

Timsort uses partially sorted lists to sort in O(n) time.

Random.sample() picks between two algorithms depending on how you are going to use it because 1,000 choose 900 is very different from 1,000,000 choose 10.

OrderedDicts usually have O(n) performance for deletion. By using a doubly linked list to store items in order and a dict for lookup you get O(1) for all operations. New code is in Python 3.

Python has native support for sets of sets. This enables easy translation of English description of problems involving sets to Python code in a few lines. Applicable to translating an NFA to a DFA.

David Beazley

http://www.dabeaz.com/

Simple CPU bound processes run more slowly when threaded. 2x slower in example. Threads actually run faster if you disable a processor.

Rationale for the GIL is that it makes the interpreter massively simpler. So rather than parallel computing what you get in a python process is cooperative multi-tasking. So IO bound operations allow other things to do work while they are waiting for data.

Thread running code is pretty simple. The big question is how something so simple has such diabolical behavior in what seem like simple programs.

CPU bound threads might reacquire the GIL immediately after release.

On multi-processor machines the sleeping threads try each 100 ticks to acquire the GIL and often fail. But they do switch context more often than the single processor case. So you get lots of useless attempts to run on one CPU while the other does work, plus lots of thrashing.

Thrashing also occurs on IO bound systems as well.

GIL code hasn’t changed for many years. All available versions of Python have essentially the same code in place for the GIL.

New GIL has been implemented that doesn’t rely on signals. There is a 5ms timeout period. This causes latency to event handling. Also the thread that requested the timeout might not be the one that gets executed.

New GIL also has diabolical behaviors as well. Possible solutions already exist in the area of operating systems. Reward threads that give up the GIL before their timeslice is over with higher priority. Penalize threads that use their entire timeslice with a lower priority.