High-level views on low-level representations

Top-3 contributions of the paper:

• Provide "bit representation" in Functional Languages: typed encoding of binary data, with concatenation and pattern-matching
• Provide "bitdata": translates binary representations into almost-algebraic data-types
• All these things come without loosing type inference!

Most glaring problems:

• "Junk" is badly dealt with, IMHO: in a Total functional programming mindset, I would rather ask "fromBit" to be called in a Maybe monad or something like that. I also understand there design choice, though.
• The interaction with the Garbage Collector is not discussed. However, I suspect that the GC expects the data to be packed in a certain, uniform way. It's not the case here. For instance, how do we "mark" a 32 bits binary data on a 32 bits architecture?
• The comparison/interaction with IDL's might deserve a longer discussion: especially, I would have loved to hear about the potential interactions between IDL and a pure functional language (basically, doesn't the IDL define the combinators of a (Driver) monad?)

What could be improved:

The future works mentioned at the end are interesting. I'll stay here, holding my breath until it's done.

What are the implications for the Research community:

Extremely general design that solves an important problem for the adoption of Functional Programming languages in low-level System design.