3. ๏ Dynamic programming languages are
powerful, productive
Thursday, August 18, 11 2
4. ๏ Dynamic programming languages are
powerful, productive
๏ Errors from dynamic typing are not fun
Thursday, August 18, 11 2
5. ๏ Dynamic programming languages are
powerful, productive
๏ Errors from dynamic typing are not fun
๏ Too much type information is “trapped
inside”
Thursday, August 18, 11 2
6. ๏ Dynamic programming languages are
powerful, productive
๏ Errors from dynamic typing are not fun
๏ Too much type information is “trapped
inside”
๏ We can do better without resorting to
static types
Thursday, August 18, 11 2
7. closed versus open
(cond
(map? x) ...
(vector? x) ...
(list? x) ...
:else ...)
Thursday, August 18, 11 3
9. we want to enumerate what is
allowed
duck-typing doesn’t let us specify a
specific set of things which are
allowed
Thursday, August 18, 11 5
10. we want to enumerate what is
allowed
duck-typing doesn’t let us specify a
specific set of things which are
allowed
๏ Types are often too coarse a granularity
for the kind of dispatch we would like to
specify
Thursday, August 18, 11 5
11. we want to enumerate what is
allowed
duck-typing doesn’t let us specify a
specific set of things which are
allowed
๏ Types are often too coarse a granularity
for the kind of dispatch we would like to
specify
๏ Duck-typing can be a source of pain.
Thursday, August 18, 11 5
12. we want to enumerate what is
allowed
duck-typing doesn’t let us specify a
specific set of things which are
allowed
๏ Types are often too coarse a granularity
for the kind of dispatch we would like to
specify
๏ Duck-typing can be a source of pain.
๏ Pre and post conditions are also
“trapped inside” functions
Thursday, August 18, 11 5
13. we often reach into a
data structure to pull it
part ...
but isn’t this just like
what we do w/ cond +
map? vector?
(cond
(= (first s) 1) ...
(= (first s) 2) ...
(= (second s) :foo) ...
...)
Thursday, August 18, 11 6
14. destructuring does make
it more convenient... but
something is still missing
(let [[x & r] s]
(cond
(= x 1) ...
(= x 2) ...
(= (second s) :foo) ...
...))
Thursday, August 18, 11 7
15. this where we are today
with match...
we’ll talk about this but
note that this is very
much a chocolate fudge
machine infused cond
(match [x]
[[1 & r]] ...
[[2 & r]] ...
[[_ :foo & r]] ...
...)
Thursday, August 18, 11 8
16. but this is where we’d
like to be
(extend-pred foo [[1 & r]] ...)
(extend-pred foo [[2 & r]] ...)
(extend-pred foo [[_ :foo & r]] ...)
Thursday, August 18, 11 9
18. Goals
๏ As fast or faster than destructuring for
matches with few cases
Thursday, August 18, 11 10
19. Goals
๏ As fast or faster than destructuring for
matches with few cases
๏ Pattern matching should follow
destructuring syntax when possible
Thursday, August 18, 11 10
20. Goals
๏ As fast or faster than destructuring for
matches with few cases
๏ Pattern matching should follow
destructuring syntax when possible
๏ Extensible (!)
Thursday, August 18, 11 10
21. Goals
๏ As fast or faster than destructuring for
matches with few cases
๏ Pattern matching should follow
destructuring syntax when possible
๏ Extensible (!)
๏ Testbed for predicate dispatch
Thursday, August 18, 11 10
24. ๏ Popular feature among functional
programming languages - Standard ML,
Erlang, Haskell, OCaml, Scala
Thursday, August 18, 11 12
25. ๏ Popular feature among functional
programming languages - Standard ML,
Erlang, Haskell, OCaml, Scala
๏ Literature on efficient pattern matching
in the ML language family is extensive
Thursday, August 18, 11 12
26. ๏ Popular feature among functional
programming languages - Standard ML,
Erlang, Haskell, OCaml, Scala
๏ Literature on efficient pattern matching
in the ML language family is extensive
๏ Decisions trees and backtracking
automata popular approaches
Thursday, August 18, 11 12
33. ๏ Compiling Pattern Matching to Good
Decision Trees
Thursday, August 18, 11 16
34. ๏ Compiling Pattern Matching to Good
Decision Trees
๏ Simple compilation algorithm
Thursday, August 18, 11 16
35. ๏ Compiling Pattern Matching to Good
Decision Trees
๏ Simple compilation algorithm
๏ Big idea is choosing which column to
test based on the notion of “necessity”
from lazy pattern matching
Thursday, August 18, 11 16
36. How it works in match
Thursday, August 18, 11 17
37. false and true are literals, _ is a
wildcard pattern, it will match
anything
(match [x y z]
[_ false true] 1
[false true _ ] 2
[_ _ false] 3
[_ _ true] 4
:else 5)
Thursday, August 18, 11 18
38. x y z
[_ f t] 1
[f t _] 2
[_ _ f] 3
[_ _ t] 4
[_ _ _] 5
Thursday, August 18, 11 19
39. top down evaluation
order, we don’t need to
test anything below a
wildcard
x y z
[_ f t] 1
[f t _] 2
[_ _ f] 3
[_ _ t] 4
[_ _ _] 5
Thursday, August 18, 11 20
40. y column has the largest
useful (non-wildcard)
patterns
y
[_ f t] 1
[f t _] 2
[_ _ f] 3
[_ _ t] 4
[_ _ _] 5
Thursday, August 18, 11 21
41. swap y column to the
front, now we need to
specialize
y x z
[f _ t] 1
[t f _] 2
[_ _ f] 3
[_ _ t] 4
[_ _ _] 5
Thursday, August 18, 11 22
44. we remove the rows that
x z
don’t match the value for
y. we drop the y column.
here are the next
2
pattern matrices for the
t wo values of y
[f _]
t [_
[_
f]
t]
3
4
[_ _] 5
x z
[_ t] 1
f [_
[_
f]
t]
3
4
[_ _] 5
Thursday, August 18, 11 25
46. we can take that
process and produce a
nested conditional.
(cond note that the order
(= y false) (cond testing matches
what we saw in
(= z false) (let [] 3) previous slides
(= z true) (let [] 1)
:else (throw (java.lang.Exception. "Found
FailNode")))
(= y true) (cond
(= x false) (let [] 2)
:else (cond
(= z false) 3
(= z true) 4
:else (throw
(java.lang.Exception.
"Found FailNode"))))
:else (cond
(= z false) (let [] 3)
(= z true) (let [] 4)
:else (throw (java.lang.Exception. "Found FailNode"))))
Thursday, August 18, 11 27
48. ๏ a pattern matrix is composed of pattern
rows
Thursday, August 18, 11 28
49. ๏ a pattern matrix is composed of pattern
rows
๏ pattern rows contain all the specified
patterns in addition to an action
Thursday, August 18, 11 28
50. ๏ a pattern matrix is composed of pattern
rows
๏ pattern rows contain all the specified
patterns in addition to an action
๏ pattern rows must all be of the same size
(equal number of patterns)
Thursday, August 18, 11 28
52. ๏ Patterns in match are implemented as
deftypes
Thursday, August 18, 11 29
53. ๏ Patterns in match are implemented as
deftypes
๏ The key protocol for a pattern to extend
is ISpecializeMatrix which defines a
single protocol fn - specialize-matrix
Thursday, August 18, 11 29
54. ๏ Patterns in match are implemented as
deftypes
๏ The key protocol for a pattern to extend
is ISpecializeMatrix which defines a
single protocol fn - specialize-matrix
๏ The pattern produces the new matrix
after specialization. This may involve
introducing new occurrences.
Thursday, August 18, 11 29
55. what are occurrences?
y x z
[f _ t]
[t f _]
[_ _ f]
[_ _ t]
Thursday, August 18, 11 30
56. consider this seq pattern
match
(match [x]
[[1 & r] 1
[[2 & r] 2
[[3 & r] 3)
Thursday, August 18, 11 31
57. x
[[1 & r]]
[[2 & r]]
[[3 & r]]
Thursday, August 18, 11 32
58. when the matrix is specialized by
SeqPattern we get this new pattern
matrix. We have the occurrence that
represents the head of the list and
the tail of the list.
xh xt
[1 r]
[2 r]
[3 r]
Thursday, August 18, 11 33
62. ๏ We end up with a tree of switch nodes.
Each switch nodes represents a
occurrence/binding and a multiway test
which points to other switch nodes, leaf
nodes, or fail nodes.
Thursday, August 18, 11 36
63. ๏ We end up with a tree of switch nodes.
Each switch nodes represents a
occurrence/binding and a multiway test
which points to other switch nodes, leaf
nodes, or fail nodes.
๏ This is the decision tree.
Thursday, August 18, 11 36
65. Seq pattern matching
note that the length of the
seq patterns do no matter!
(match [x]
[[1] :a0
[[1 2]] :a1
[[1 2 nil nil nil]] :a2
:else :a3)
Thursday, August 18, 11 38
66. note the rest pattern syntax
support just like destructuring.
notice that we can introduce
bindings anywhere we would
use a wildcard
(match [x]
[[1]] :a0
[[_ 2 & [a & b]]] [:a1 a b]
:else :a2)
Thursday, August 18, 11 39
68. we can restrict that only
maps with the exact keys will
match
(match [x]
[{_ :a 2 :b :only [:a :b]}] :a0
[{1 :a c :c}] :a1
[{3 :c d :d 4 :e}] :a2
:else :a3)
Thursday, August 18, 11 41
69. or patterns!
(match [x y z]
[[1 (3 | 4) 3]] :a0
[[1 (2 | 3) 3]] :a1
:else :a2)
Thursday, August 18, 11 42
71. sometimes you want to
match a specific
portion and bind that
to a local name
(match [v]
[[3 1]] :a0
[[([1 a] :as b)]] [:a1 a b]
:else :a2)
Thursday, August 18, 11 44
72. you can pattern match
Java classes!
(extend-type java.util.Date
IMatchLookup
(val-at* [this k not-found]
(case k
:year (.getYear this)
:month (.getMonth this)
:date (.getDate this)
:hours (.getHours this)
:minutes (.getMinutes this)
not-found)))
Thursday, August 18, 11 45
73. (let [d (java.util.Date. 2010 10 1 12 30)]
(match [d]
[{2009 :year a :month}] [:a0 a]
[{(2010 | 2011) :year b :month}] [:a1 b]))
Thursday, August 18, 11 46
75. problems
๏ Pattern matching is closed
Thursday, August 18, 11 47
76. problems
๏ Pattern matching is closed
๏ We want the matching to happen at the
level of our most powerful abstraction -
functions
Thursday, August 18, 11 47
77. problems
๏ Pattern matching is closed
๏ We want the matching to happen at the
level of our most powerful abstraction -
functions
๏ Without open dispatch, all we have is a
chocolate fudge machine powered cond
Thursday, August 18, 11 47
80. problems
๏ Hard coded dispatch function
Thursday, August 18, 11 49
81. problems
๏ Hard coded dispatch function
๏ In order to specify complex matches we
have to construct a collection
Thursday, August 18, 11 49
82. problems
๏ Hard coded dispatch function
๏ In order to specify complex matches we
have to construct a collection
๏ For complex matches, performance is less
than we would like
Thursday, August 18, 11 49
84. future directions
๏ Vector patterns: for any data type that
supports random access and fast
“slicing” - PersistentVector, primitive
arrays, buffers, etc.
Thursday, August 18, 11 50
85. future directions
๏ Vector patterns: for any data type that
supports random access and fast
“slicing” - PersistentVector, primitive
arrays, buffers, etc.
๏ Predicate Dispatch (huh?)
Thursday, August 18, 11 50
92. Goals
๏ move the matching to level of function, as with
multimethods
Thursday, August 18, 11 53
93. Goals
๏ move the matching to level of function, as with
multimethods
๏ this change is in conflict with the semantics of
pattern matching - pattern matching is ordered
Thursday, August 18, 11 53
94. Goals
๏ move the matching to level of function, as with
multimethods
๏ this change is in conflict with the semantics of
pattern matching - pattern matching is ordered
๏ We need some way to know where to put new
pattern rows in the matrix
Thursday, August 18, 11 53
96. A Sketch
๏ Use core.logic to order the pattern rows
Thursday, August 18, 11 54
97. A Sketch
๏ Use core.logic to order the pattern rows
๏ A high performance in-memory DAG
representation of the decision tree
Thursday, August 18, 11 54
98. A Sketch
๏ Use core.logic to order the pattern rows
๏ A high performance in-memory DAG
representation of the decision tree
๏ Perhaps we can go the route of deftype-
inline patterns get best performance
Thursday, August 18, 11 54
100. Challenges
๏ How much of the pattern matching
syntax can we bring over?
Thursday, August 18, 11 55
101. Challenges
๏ How much of the pattern matching
syntax can we bring over?
๏ How close can we get to the performance
of static code?
Thursday, August 18, 11 55
102. Challenges
๏ How much of the pattern matching
syntax can we bring over?
๏ How close can we get to the performance
of static code?
๏ Can we limit the scope of changes to
namespaces?
Thursday, August 18, 11 55