|
| 1 | +--- |
| 2 | +sidebar_position: 7 |
| 3 | +--- |
| 4 | + |
| 5 | +# Regular expressions |
| 6 | + |
| 7 | +Lenses, prisms, and basic traversals have an important limitation: the only |
| 8 | +go _one level deep_ in the data structure. The functions in the |
| 9 | +`arrow.optics.regex` package remove that limitation, providing a great |
| 10 | +foundation for querying and modifying hierarchical data, such as trees |
| 11 | +or JSON documents. |
| 12 | + |
| 13 | +<!--- TEST_NAME OpticsRegex --> |
| 14 | + |
| 15 | +## Repetition |
| 16 | + |
| 17 | +<!--- INCLUDE .* |
| 18 | +import arrow.optics.* |
| 19 | +import arrow.optics.dsl.* |
| 20 | +import arrow.optics.regex.* |
| 21 | +import arrow.optics.regex.dsl.* |
| 22 | +import io.kotest.matchers.shouldBe |
| 23 | +--> |
| 24 | + |
| 25 | +We shall use different forms of trees in the examples below. |
| 26 | +The first variation only has data in the leaves. |
| 27 | + |
| 28 | +```kotlin |
| 29 | +@optics sealed interface BinaryTree1<out A> { |
| 30 | + companion object |
| 31 | +} |
| 32 | + |
| 33 | +@optics data class Node1<out A>( |
| 34 | + val children: List<BinaryTree1<A>> |
| 35 | +) : BinaryTree1<A> { |
| 36 | + constructor(vararg children: BinaryTree1<A>) : this(children.toList()) |
| 37 | + |
| 38 | + companion object |
| 39 | +} |
| 40 | + |
| 41 | +@optics data class Leaf1<out A>( |
| 42 | + val value: A |
| 43 | +) : BinaryTree1<A> { |
| 44 | + companion object |
| 45 | +} |
| 46 | +``` |
| 47 | + |
| 48 | +Suppose we want to increment all the numbers in a binary tree of integers. |
| 49 | +The code below attempts to do that, but fails because it only traverses the |
| 50 | +children of nodes in one level -- this is why only the last `Leaf1` is |
| 51 | +modified after the call. |
| 52 | + |
| 53 | +```kotlin |
| 54 | +val exampleTree1 = Node1(Node1(Leaf1(1), Leaf1(2)), Leaf1(3)) |
| 55 | + |
| 56 | +fun example() { |
| 57 | + val path = BinaryTree1.node1<Int>().children().every.leaf1().value() |
| 58 | + |
| 59 | + val modifiedTree1 = path.modify(exampleTree1) { it + 1 } |
| 60 | + modifiedTree1 shouldBe Node1(Node1(Leaf1(1), Leaf1(2)), Leaf1(4)) |
| 61 | +} |
| 62 | +``` |
| 63 | + |
| 64 | +<!--- KNIT example-optics-regex-01.kt --> |
| 65 | +<!--- TEST assert --> |
| 66 | + |
| 67 | +<!--- INCLUDE .* |
| 68 | +@optics sealed interface BinaryTree1<out A> { |
| 69 | + companion object |
| 70 | +} |
| 71 | +
|
| 72 | +@optics data class Node1<out A>( |
| 73 | + val children: List<BinaryTree1<A>> |
| 74 | +) : BinaryTree1<A> { |
| 75 | + constructor(vararg children: BinaryTree1<A>) : this(children.toList()) |
| 76 | +
|
| 77 | + companion object |
| 78 | +} |
| 79 | +
|
| 80 | +@optics data class Leaf1<out A>( |
| 81 | + val value: A |
| 82 | +) : BinaryTree1<A> { |
| 83 | + companion object |
| 84 | +} |
| 85 | +
|
| 86 | +val exampleTree1 = Node1(Node1(Leaf1(1), Leaf1(2)), Leaf1(3)) |
| 87 | +--> |
| 88 | + |
| 89 | +So how should be look at this problem? First, we know that we'll always |
| 90 | +end by traversing a final `Leaf1` and the `value` there. In the middle, |
| 91 | +we may need to go down the children once or more. In fact, _zero_ |
| 92 | +times should also be considered, as the binary tree could be just a |
| 93 | +single leaf. We can express this idea by wrapping the first segment |
| 94 | +of the previous path with `zeroOrMore`. |
| 95 | + |
| 96 | +```kotlin |
| 97 | +fun example() { |
| 98 | + val path = zeroOrMore(BinaryTree1.node1<Int>().children().every).leaf1().value() |
| 99 | + |
| 100 | + val modifiedTree1 = path.modify(exampleTree1) { it + 1 } |
| 101 | + modifiedTree1 shouldBe Node1(Node1(Leaf1(2), Leaf1(3)), Leaf1(4)) |
| 102 | +} |
| 103 | +``` |
| 104 | + |
| 105 | +<!--- KNIT example-optics-regex-02.kt --> |
| 106 | +<!--- TEST assert --> |
| 107 | + |
| 108 | +The functions `zeroOrMore` and `onceOrMore` provide _repetition_ of |
| 109 | +a single lens, prims, or traversal, that is applied recursively. |
| 110 | +These functions are available on every scenario in which you can |
| 111 | +construct an optics going from a type to itself -- in our example, |
| 112 | +`node1<Int>().children().every` focus from `BinaryTree1` into |
| 113 | +`BinaryTree1`. |
| 114 | + |
| 115 | +## Combination |
| 116 | + |
| 117 | +Let's now consider another variation of binary trees, in which now |
| 118 | +at every step (leaf or node) we find a value. |
| 119 | + |
| 120 | +```kotlin |
| 121 | +@optics sealed interface BinaryTree2<out A> { |
| 122 | + companion object |
| 123 | +} |
| 124 | + |
| 125 | +@optics data class Node2<out A>( |
| 126 | + val innerValue: A, |
| 127 | + val children: List<BinaryTree2<A>> |
| 128 | +) : BinaryTree2<A> { |
| 129 | + constructor(value: A, vararg children: BinaryTree2<A>) : this(value, children.toList()) |
| 130 | + |
| 131 | + companion object |
| 132 | +} |
| 133 | + |
| 134 | +@optics data class Leaf2<out A>( |
| 135 | + val value: A |
| 136 | +) : BinaryTree2<A> { |
| 137 | + companion object |
| 138 | +} |
| 139 | +``` |
| 140 | + |
| 141 | +If we construct a path similar to the previous one, we shall only |
| 142 | +focus on those values in leaves, as we can see in the example below. |
| 143 | + |
| 144 | +```kotlin |
| 145 | +val exampleTree2 = Node2(1, Node2(2, Leaf2(3), Leaf2(4)), Leaf2(5)) |
| 146 | + |
| 147 | +fun example() { |
| 148 | + val path = zeroOrMore(BinaryTree2.node2<Int>().children().every).leaf2().value() |
| 149 | + |
| 150 | + val modifiedTree2 = path.modify(exampleTree2) { it + 1 } |
| 151 | + modifiedTree2 shouldBe Node2(1, Node2(2, Leaf2(4), Leaf2(5)), Leaf2(6)) |
| 152 | +} |
| 153 | +``` |
| 154 | + |
| 155 | +<!--- KNIT example-optics-regex-03.kt --> |
| 156 | +<!--- TEST assert --> |
| 157 | + |
| 158 | +<!--- INCLUDE .* |
| 159 | +@optics sealed interface BinaryTree2<out A> { |
| 160 | + companion object |
| 161 | +} |
| 162 | +
|
| 163 | +@optics data class Node2<out A>( |
| 164 | + val innerValue: A, |
| 165 | + val children: List<BinaryTree2<A>> |
| 166 | +) : BinaryTree2<A> { |
| 167 | + constructor(value: A, vararg children: BinaryTree2<A>) : this(value, children.toList()) |
| 168 | +
|
| 169 | + companion object |
| 170 | +} |
| 171 | +
|
| 172 | +@optics data class Leaf2<out A>( |
| 173 | + val value: A |
| 174 | +) : BinaryTree2<A> { |
| 175 | + companion object |
| 176 | +} |
| 177 | +
|
| 178 | +val exampleTree2 = Node2(1, Node2(2, Leaf2(3), Leaf2(4)), Leaf2(5)) |
| 179 | +--> |
| 180 | + |
| 181 | +The solution in this case is to _combine_ two different traversals into a single one. |
| 182 | +In the code below we build `nodeValues`, that focuses on values found in nodes, |
| 183 | +and `leafValues`, that focuses on those values in the leaves. Then we combine them |
| 184 | +using the `and` infix function from the library. |
| 185 | + |
| 186 | +```kotlin |
| 187 | +fun example() { |
| 188 | + val nodeValues = zeroOrMore(BinaryTree2.node2<Int>().children().every).node2().innerValue() |
| 189 | + val leafValues = zeroOrMore(BinaryTree2.node2<Int>().children().every).leaf2().value() |
| 190 | + val path = nodeValues and leafValues |
| 191 | + |
| 192 | + val modifiedTree2 = path.modify(exampleTree2) { it + 1 } |
| 193 | + modifiedTree2 shouldBe Node2(2, Node2(3, Leaf2(4), Leaf2(5)), Leaf2(6)) |
| 194 | +} |
| 195 | +``` |
| 196 | + |
| 197 | +<!--- KNIT example-optics-regex-04.kt --> |
| 198 | +<!--- TEST assert --> |
| 199 | + |
| 200 | +It is also possible to remove some of the duplication in the code above. |
| 201 | +In particular, we can separate the "digging" in the binary tree (the part where |
| 202 | +we use `zeroOrMore`) from obtaining the value by either looking at the one |
| 203 | +on a node or at the one on a leaf. |
| 204 | + |
| 205 | +```kotlin |
| 206 | +fun example() { |
| 207 | + val pathToValue = BinaryTree2.node2<Int>().innerValue() and BinaryTree2.leaf2<Int>().value() |
| 208 | + val path = zeroOrMore(BinaryTree2.node2<Int>().children().every) compose pathToValue |
| 209 | + |
| 210 | + val modifiedTree2 = path.modify(exampleTree2) { it + 1 } |
| 211 | + modifiedTree2 shouldBe Node2(2, Node2(3, Leaf2(4), Leaf2(5)), Leaf2(6)) |
| 212 | +} |
| 213 | +``` |
| 214 | + |
| 215 | +<!--- KNIT example-optics-regex-05.kt --> |
| 216 | +<!--- TEST assert --> |
| 217 | + |
| 218 | +Unfortunately, with this refactor we no longer can use the chained syntax |
| 219 | +using `.` all the time. We need to resort to `compose` instead. |
| 220 | + |
| 221 | +:::info Regular expressions? |
| 222 | + |
| 223 | +You might be wondering why we call `zeroOrMore`, `onceOrMore`, and `and` the |
| 224 | +_regular expression_ functions. To understand this, we need to view _paths_ |
| 225 | +to data as "strings" in which each "letter" represent one single optic. |
| 226 | + |
| 227 | +For example, if we use `n` for `node1`, `c` for `children`, `e` for `every`, |
| 228 | +`l` for `leaf1`, and `v` for `value`, the "strings" we want to use to access |
| 229 | +all the values are those of the form: |
| 230 | + |
| 231 | +``` |
| 232 | +lv |
| 233 | +ncelv |
| 234 | +ncencelv |
| 235 | +ncencencelv |
| 236 | +... |
| 237 | +``` |
| 238 | + |
| 239 | +From that point of view, the regular expression that matches all the possible |
| 240 | +paths we want to take is `(nce)*lv`, where `*` is the regular expression |
| 241 | +operator that matches that string zero or more times. |
| 242 | + |
| 243 | +Similarly, `onceOrMore` corresponds to `+`. The `and` functions corresponds |
| 244 | +to `+`, since it allows choosing between several matches, that in turn |
| 245 | +correspond to different possible paths to focus on the data. |
| 246 | + |
| 247 | +::: |
0 commit comments