Thu Mar 26 16:29:07 CET 2009
Some of today's intuitive flashes..
* pf object tree in filesystem
Instead of managing data objects internally, why not export them
as a filesystem? This would open some doors for serialization.
Since the whole idea of PF is to run in two regimes: startup
(compile code / build datastuctures) and run, this might not be
such a bad approach. The only question is: is this expensive?
Once disk objects are memory-mapped, do they behave as files or
as memory? And is a ramfs filesystem really fast enough?
* functions from types: all functions = type convertors
One of the early ideas in PF is to make all data-conversion
automatic. Thinking about this some more: a lot of programs
these days translate data from one form into another, and don't
really do much more (they don't "act"). Can this be
extrapolated to a point where types themselves define programs,
and functions between them can be derived automatically?
I ran into some google video presentation about code that does
this in a more mathematical setting, but can it be done for
(I.e. given a pair of images (which provides order) a binary
operator (-) and the "map" function represented as a type, the
combined operation can be derived.)