[notes,feldspar
tom@zwizwa.be**20130524223309
 Ignore-this: a8195a0356cab1bdd1bffe441ade99f1
] hunk ./meta.txt 29646
+Entry: What is a stream?
+Date: Fri May 24 15:31:45 EDT 2013
+
+That is the real question.  The problem with the array implementation
+is that it is hard to interpret in terms of imperative code.
+
+So is there a way to keep more true to the definition?
+
+The main problem seems to be situated when moving from a pure
+operation on a stream, to a non-pure stateful operation on the next
+element.
+
+Somehow, something is lost.
+
+Who owns that context, the state streams?
+
+
+
+
+Entry: FeldSpar
+Date: Fri May 24 15:39:19 EDT 2013
+
+From [1]:
+
+    The current version of Feldspar deals only with pure data
+    processing; although, we have initiated work to extend the
+    language to encompass control.
+
+It looks like they avoided feedback state.
+
+Basic Ideas:
+  - Symbolic arrays: map f (Indexed l ixf) = Indexed l (f . ixf)
+
+
+Maybe the idea of symbolic arrays can be reused.  Currently I rely on
+type inference, but to abstract the indexing procedure in the form of
+curried functions is a possibility.
+
+Basically, given an opaque node `s' that is passed to a map function
+will return a new opaque node s.  The map doesn't need to be forced
+until the type is known.  Let's give this a try.
+
+[1] http://journals.cambridge.org/action/displayJournal?jid=JFP
+
+
+Entry: Grid
+Date: Fri May 24 15:50:01 EDT 2013
+
+What is the basic idea?  DSP is about computations on a grid.  There
+are different ways of constructing grids.  What we're interested in is
+local context: given a point on a grid, there is a way to get to a
+neighbouring point.
+
+One limitiation is important: the time dimension is special.  If the
+time dimension is causal (computations never refer to the future) it
+is possible to re-arrange the computations in such a way that they
+take the form of a state space model.
+
+The key point is to be able to separate the description of grid
+computations and their dependencies, from an actual implementation as
+loops.
+
+It's probably time to take a look again at APL and J.
+
+
+
+Entry: Roadmap
+Date: Fri May 24 18:19:33 EDT 2013
+
+- Generalize reduce to foldmap (accumulation + output streams)
+
+- Allow for compile-time vectors?  Maybe not necessary (just use
+  constant arrays?)
+
+
+
+
+
+
+
hunk ./rai/doc/rai.html 37
+<li> Go over <a href="">FeldSpar</a> and its goals and inspirations.
+This <a href="http://www.cse.chalmers.se/~ms/MemoCode.pdf">paper</a> has a nice list of references to Spiral, Pan, Obsidian, 
+
+