The Apache Ant API, a rarely used feature of the tool, provides reliable and portable ways to perform command-line operations from within Java.
The advent of a unified memory model in CUDA 6 has made GPU programming much easier by removing lots of implementation details that developers formerly had to manage manually.
A very fast, parallel, simple to implement sort makes efficient use of memory.
The lack of quality manuals and documentation for developer tools greatly diminishes our ability to work well in subtle but important ways.
The best books of the past 12 months.