1DATAFLOW(1) User Contributed Perl Documentation DATAFLOW(1)
2
6 PDL::Dataflow -- description of the dataflow philosophy
7
9 pdl> $a = zeroes(10);
10 pdl> $b = $a->slice("2:4:2");
11 pdl> $b ++;
12 pdl> print $a;
13 [0 0 1 0 1 0 0 0 0 0]
14
16 Dataflow is very experimental. Many features of it are disabled for
17 2.0, particularly families for one-directional dataflow. If you wish to
18 use one-directional dataflow for something, please contact the author
19 first and we'll work out how to make it functional again.
20 Two-directional dataflow (which implements ->slice() etc.) is fully
22 functional, however. Just about any function which returns some subset
23 of the values in some piddle will make a binding so that
24 $a = some piddle
26 $b = $a->slice("some parts");
27 $b->set(3,3,10);
28 also changes the corresponding element in $a. $b has become effectively
30 a window to some sub-elements of $a. You can also define your own
31 routines that do different types of subsets. If you don't want $b to be
32 a window to $a, you must do
33 $b = $a->slice("some parts")->copy;
35 The copying turns off all dataflow between the two piddles.
37 The difficulties with one-directional dataflow are related to sequences
39 like
40 $b = $a + 1;
42 $b ++;
43 where there are several possible outcomes and the semantics get a
45 little murky.
46
48 Dataflow is new to PDL2.0. The basic philosophy behind dataflow is that
49 > $a = pdl 2,3,4;
51 > $b = $a * 2;
52 > print $b
53 [2 3 4]
54 > $a->set(0,5);
55 > print $b;
56 [10 3 4]
57 should work. It doesn't. It was considered that doing this might be too
59 confusing for novices and occasional users of the language. Therefore,
60 you need to explicitly turn on dataflow, so
61 > $a = pdl 2,3,4;
63 > $a->doflow();
64 > $b = $a * 2;
65 ...
66 produces the unexpected result. The rest of this documents explains
68 various features and details of the dataflow implementation.
69
71 When you calculate something like the above
72 > $a = pdl 2,3,4;
74 > $a->doflow();
75 > $b = $a * 2;
76 nothing will have been calculated at this point. Even the memory for
78 the contents of $b has not been allocated. Only the command
79 > print $b
81 will actually cause $b to be calculated. This is important to bear in
83 mind when doing performance measurements and benchmarks as well as when
84 tracking errors.
85 There is an explanation for this behaviour: it may save cycles but more
87 importantly, imagine the following:
88 > $a = pdl 2,3,4;
90 > $b = pdl 5,6,7;
91 > $c = $a + $b;
92 ...
93 > $a->resize(4);
94 > $b->resize(4);
95 > print $c;
96 Now, if $c were evaluated between the two resizes, an error condition
98 of incompatible sizes would occur.
99 What happens in the current version is that resizing $a raises a flag
101 in $c: "PDL_PARENTDIMSCHANGED" and $b just raises the same flag again.
102 When $c is next evaluated, the flags are checked and it is found that a
103 recalculation is needed.
104 Of course, lazy evaluation can sometimes make debugging more painful
106 because errors may occur somewhere where you'd not expect them. A
107 better stack trace for errors is in the works for PDL, probably so that
108 you can toggle a switch $PDL::traceevals and get a good trace of where
109 the error actually was.
110
112 This is one of the more intricate concepts of one-directional dataflow.
113 Consider the following code ($a and $b are pdls that have dataflow
114 enabled):
115 $c = $a + $b;
117 $e = $c + 1;
118 $d = $c->diagonal();
119 $d ++;
120 $f = $c + 1;
121 What should $e and $f contain now? What about when $a is changed and a
123 recalculation is triggered.
124 In order to make dataflow work like you'd expect, a rather strange
126 concept must be introduced: families. Let us make a diagram:
127 a b
129 \ /
130 c
131 /|
132 / |
133 e d
134 This is what PDL actually has in memory after the first three lines.
136 When $d is changed, we want $c to change but we don't want $e to change
137 because it already is on the graph. It may not be clear now why you
138 don't want it to change but if there were 40 lines of code between the
139 2nd and 4th lines, you would. So we need to make a copy of $c and $d:
140 a b
142 \ /
143 c' . . . c
144 /| |\
145 / | | \
146 e d' . . . d f
147 Notice that we primed the original c and d, because they do not
149 correspond to the objects in $c and $d any more. Also, notice the
150 dotted lines between the two objects: when $a is changed and this
151 diagram is re-evaluated, $c really does get the value of c' with the
152 diagonal incremented.
153 To generalize on the above, whenever a piddle is mutated i.e. when its
155 actual *value* is forcibly changed (not just the reference:
156 $d = $d + 1
158 would produce a completely different result ($c and $d would not be
160 bound any more whereas
161 $d .= $d + 1
163 would yield the same as $d++), a "family" consisting of all other
165 piddles joined to the mutated piddle by a two-way transformation is
166 created and all those are copied.
167 All slices or transformations that simply select a subset of the
169 original pdl are two-way. Matrix inverse should be. No arithmetic
170 operators are.
171
173 What you were told in the previous section is not quite true: the
174 behaviour described is not *always* what you want. Sometimes you would
175 probably like to have a data "source":
176 $a = pdl 2,3,4; $b = pdl 5,6,7;
178 $c = $a + $b;
179 line($c);
180 Now, if you know that $a is going to change and that you want its
182 children to change with it, you can declare it into a data source (XXX
183 unimplemented in current version):
184 $a->datasource(1);
186 After this, $a++ or $a .= something will not create a new family but
188 will alter $a and cut its relation with its previous parents. All its
189 children will follow its current value.
190 So if $c in the previous section had been declared as a source, $e and
192 $f would remain equal.
193
195 A dataflow mechanism would not be very useful without the ability to
196 bind events onto changed data. Therefore, we provide such a mechanism:
197 > $a = pdl 2,3,4
199 > $b = $a + 1;
200 > $c = $b * 2;
201 > $c->bind( sub { print "A now: $a, C now: $c\n" } )
202 > PDL::dowhenidle();
203 A now: [2,3,4], C now: [6 8 10]
204 > $a->set(0,1);
205 > $a->set(1,1);
206 > PDL::dowhenidle();
207 A now: [1,1,4], C now: [4 4 10]
208 Notice how the callbacks only get called during PDL::dowhenidle. An
210 easy way to interface this to Perl event loop mechanisms (such as Tk)
211 is being planned.
212 There are many kinds of uses for this feature: self-updating graphs,
214 for instance.
215 Blah blah blah XXX more explanation
217
219 Dataflow as such is a fairly limited addition on top of Perl. To get a
220 more refined addition, the internals of Perl need to be hacked a
221 little. A true implementation would enable flow of everything,
222 including
223 data
225 data size
226 datatype
227 operations
228 At the moment we only have the first two (hey, 50% in a couple of
230 months is not bad ;) but even this is useful by itself. However,
231 especially the last one is desirable since it would add the possibility
232 of flowing closures from place to place and would make many things more
233 flexible.
234 To get the rest working, the internals of dataflow probably need to be
236 changed to be a more general framework.
237 Additionally, it would be nice to be able to flow data in time, lucid-
239 like (so you could easily define all kinds of signal processing
240 things).
241
243 Copyright(C) 1997 Tuomas J. Lukka (lukka@fas.harvard.edu).
244 Redistribution in the same form is allowed provided that the copyright
245 notice stays intact but reprinting requires a permission from the
246 author.
247perl v5.30.0 2019-09-05 DATAFLOW(1)