ICFP roundup, day #2

Here are the talks from day #2 that really stood out for me:

• The impact of higher-order state and control effects on local relational reasoning This was probably my favorite result presented at the conference, though not my favorite talk. The work studies two orthogonal language features -- higher-order state and control -- and characterizes reasoning principles valid in the presence or absence of these features. The reasoning principles, like the features themselves, are orthogonal. I believe in each case the model is fully abstract, but this is due more to the use of biorthogonality than the particulars of each feature. Still, the model seems quite insightful and I'm looking forward to reading the paper.
  
• Distance makes the types grow stronger Jason Reed gave a fun talk on "differential privacy" and how it can be achieved via a type system. The work is in the area of anonymity of data: how can you release enough data to perform useful statistical work without also leaking private information? A timely and surprisingly difficult problem. Differential privacy, if I understood it correctly, is a kind of metric-continuity: if you change a data set by a small amount, say by deleting a record, the information you can glean from the data set should likewise change by a small amount. The type system is meant to enforce this property.
  
• Fortifying macros Ryan Culpepper gave an excellent talk on his macro system, known as syntax-parse. The talk told a convincing story about both the power of macros, and the difficulty of making them robust. As is often the case, error checking and reporting code easily dwarfs and obscured the intended functionality of macros. The syntax-parse system allows you to write macros in a very direct style, but still provide robust error handling, most of which you get for free from the system. To make this work, syntactic forms are given a more explicit treatment: you can define a form like binding-list-without-repetitions, give it a user-friendly name and appropriate error handling, and then use it freely in macros. Such extra structure helps to keep macros at a high level of abstraction, and to give the system the right hooks to perform error handling on its own.