Multi-Core OO: Part 2

However, although most of us can "see" any number of written sentences at the same time, few of us could "read" more than one sentence at the same time. Even the most experienced stock market traders would probably be challenged by three or more simultaneous telephone calls; our linguistic skills (textual and verbal) appear to be more or less sequential.

This should not be surprising because linguistic meaning is heavily dependent on word order, and so "watching" a football match is a very different experience from listening to it on the radio, or reading about it in a paper. Films can usually tell stories in less time than books because apart from anything else, the information bandwidth is higher (but of course that does not necessarily make it a more satisfying aesthetic experience!).

When Is Visual Programming Applicable?

Few would disagree that the best way to express an FFT or Matrix inversion algorithm is almost certainly textual; VHDL is now used by electronics engineers for the domains in which it is appropriate. In the same way that the electronics industry now adopts a hybrid visual/textual approach, software engineering can also benefit from the same specializations.

Few people would disagree that GUIs are best developed using graphical drag-and-drop metaphors, class designers are also making increased use of visual semantics, but equally few people would try to use pictures to describe a recursive quick-sort algorithm. TPL, TBB, and Blueprint are clearly not mutually exclusive.

Conventional text programs derive much of their "meaning" through their statement ordering, whereas an electronic circuit diagram derives equivalent meaning through its connectivity (which can branch and merge). In the specific cases where statement order doesn't matter, then there is typically potential for concurrent execution.

Most text languages are primarily concerned with algorithmic logic (usually branching on particular data values) and don't have an intuitive way of expressing concurrency in general, particularly the more complex irregular dependencies like "f3 can be executed when f1 and f2 complete, but f1 can't be executed while f0 is executing".

Whilst Parallel-For, Map-Reduce, and other existing mechanisms can address "regular" parallelism in an effective "bottom-up" manner, the "futures" approach to irregular problems that involve more than half a dozen concurrent threads of execution can be difficult to conjure with.