X-Git-Url: http://nitlanguage.org

diff --git a/lib/more_collections.nit b/lib/more_collections.nit
index 004cc24..dd68f12 100644
--- a/lib/more_collections.nit
+++ b/lib/more_collections.nit
@@ -13,7 +13,10 @@
 # limitations under the License.
 
 # Highly specific, but useful, collections-related classes.
-module more_collections
+module more_collections is serialize
+
+import serialization
+import poset
 
 # Simple way to store an `HashMap[K, Array[V]]`
 #
@@ -33,7 +36,20 @@ class MultiHashMap[K, V]
 	super HashMap[K, Array[V]]
 
 	# Add `v` to the array associated with `k`.
+	#
 	# If there is no array associated, then create it.
+	#
+	# For the inverse operation, see `remove_one`.
+	#
+	# ```
+	# var m = new MultiHashMap[String, Char]
+	# m.add_one("four", 'i')
+	# m.add_one("four", 'i')
+	# m.add_one("four", 'i')
+	# m.add_one("four", 'i')
+	# assert m.has_key("four")
+	# assert m["four"] == ['i', 'i', 'i', 'i']
+	# ```
 	fun add_one(k: K, v: V)
 	do
 		var x = self.get_or_null(k)
@@ -49,6 +65,73 @@ class MultiHashMap[K, V]
 		self[key] = res
 		return res
 	end
+
+	# Remove an occurrence of `v` from the array associated with `k`.
+	#
+	# If the associated array does not contain `v`, do nothing. If the
+	# associated array only contain one element and this element is `v`, remove
+	# the key `k`.
+	#
+	# In a nutshell, does the inverse operation of `add_one`.
+	#
+	# ```
+	# var m = new MultiHashMap[String, Char]
+	# m["four"]        =  ['4', 'i', 'i', 'i', 'i']
+	# m.remove_one("four", 'i')
+	# assert m["four"] == ['4', 'i', 'i', 'i']
+	#
+	# m = new MultiHashMap[String, Char]
+	# m.add_one("one", '1')
+	# m.remove_one("one", '?')
+	# assert m["one"] == ['1']
+	# m.remove_one("one", '1')
+	# assert not m.has_key("one")
+	# assert m["one"] == new Array[Char]
+	#
+	# m = new MultiHashMap[String, Char]
+	# m.add_one("one", '1')
+	# m.remove_one("two", '2')
+	# assert not m.has_key("two")
+	# assert m["one"] == ['1']
+	# assert m["two"] == new Array[Char]
+	# ```
+	fun remove_one(k: K, v: V)
+	do
+		var x = get_or_null(k)
+		if x != null then
+			x.remove(v)
+			if x.is_empty then keys.remove(k)
+		end
+	end
+
+	# Search the values in `pe.greaters` from the most smaller elements that have a value.
+	#
+	# Elements without values are ignored.
+	#
+	# Basically, values defined in nearest greater elements of `pe` are inherited.
+	#
+	# ~~~
+	# var pos = new POSet[String]
+	# pos.add_chain(["E", "D", "C", "B", "A"])
+	# pos.add_chain(["D", "X", "B"])
+	#
+	# var map = new MultiHashMap[String, String]
+	# map["A"].append(["a", "1"])
+	# map["C"].append(["c", "2"])
+	# map["X"].append(["x", "2"])
+	# map["E"].add "e"
+	#
+	# assert map.lookup_joined_values(pos["B"]).has_exactly(["a", "1"])
+	# assert map.lookup_joined_values(pos["C"]).has_exactly(["c", "2"])
+	# assert map.lookup_joined_values(pos["D"]).has_exactly(["c", "x", "2"])
+	# ~~~
+	fun lookup_joined_values(pe: POSetElement[K]): Set[V]
+	do
+		var res = new Set[V]
+		for k in pe.poset.select_smallest(filter_keys(pe.greaters)) do res.add_all self[k]
+		return res
+
+	end
 end
 
 # Simple way to store an `HashMap[K1, HashMap[K2, V]]`
@@ -61,6 +144,7 @@ end
 # assert hm2[2, "not-two"] == null
 # ~~~~
 class HashMap2[K1, K2, V]
+
 	private var level1 = new HashMap[K1, HashMap[K2, V]]
 
 	# Return the value associated to the keys `k1` and `k2`.
@@ -93,6 +177,16 @@ class HashMap2[K1, K2, V]
 		if level2 == null then return
 		level2.keys.remove(k2)
 	end
+
+	# Is there a value at `k1, k2`?
+	fun has(k1: K1, k2: K2): Bool
+	do
+		if not level1.keys.has(k1) then return false
+		return level1[k1].keys.has(k2)
+	end
+
+	# Remove all items
+	fun clear do level1.clear
 end
 
 # Simple way to store an `HashMap[K1, HashMap[K2, HashMap[K3, V]]]`
@@ -105,6 +199,7 @@ end
 # assert hm3[2, "not-two", 22] == null
 # ~~~~
 class HashMap3[K1, K2, K3, V]
+
 	private var level1 = new HashMap[K1, HashMap2[K2, K3, V]]
 
 	# Return the value associated to the keys `k1`, `k2`, and `k3`.
@@ -137,6 +232,71 @@ class HashMap3[K1, K2, K3, V]
 		if level2 == null then return
 		level2.remove_at(k2, k3)
 	end
+
+	# Is there a value at `k1, k2, k3`?
+	fun has(k1: K1, k2: K2, k3: K3): Bool
+	do
+		if not level1.keys.has(k1) then return false
+		return level1[k1].has(k2, k3)
+	end
+
+	# Remove all items
+	fun clear do level1.clear
+end
+
+# Simple way to store an `HashMap[K1, HashMap[K2, HashMap[K3, HashMap[K4, V]]]]`
+#
+# ~~~~
+# var hm4 = new HashMap4[Int, String, Int, String, Float]
+# hm4[1, "one", 11, "un"] = 1.0
+# hm4[2, "two", 22, "deux"] = 2.0
+# assert hm4[1, "one", 11, "un"] == 1.0
+# assert hm4[2, "not-two", 22, "deux"] == null
+# ~~~~
+class HashMap4[K1, K2, K3, K4, V]
+
+	private var level1 = new HashMap[K1, HashMap3[K2, K3, K4, V]]
+
+	# Return the value associated to the keys `k1`, `k2`, `k3` and `k4`.
+	# Return `null` if no such a value.
+	fun [](k1: K1, k2: K2, k3: K3, k4: K4): nullable V
+	do
+		var level1 = self.level1
+		var level2 = level1.get_or_null(k1)
+		if level2 == null then return null
+		return level2[k2, k3, k4]
+	end
+
+	# Set `v` the value associated to the keys `k1`, `k2`, `k3` and `k4`.
+	fun []=(k1: K1, k2: K2, k3: K3, k4: K4, v: V)
+	do
+		var level1 = self.level1
+		var level2 = level1.get_or_null(k1)
+		if level2 == null then
+			level2 = new HashMap3[K2, K3, K4, V]
+			level1[k1] = level2
+		end
+		level2[k2, k3, k4] = v
+	end
+
+	# Remove the item at `k1`, `k2`, `k3` and `k4`
+	fun remove_at(k1: K1, k2: K2, k3: K3, k4: K4)
+	do
+		var level1 = self.level1
+		var level2 = level1.get_or_null(k1)
+		if level2 == null then return
+		level2.remove_at(k2, k3, k4)
+	end
+
+	# Is there a value at `k1, k2, k3, k4`?
+	fun has(k1: K1, k2: K2, k3: K3, k4: K4): Bool
+	do
+		if not level1.keys.has(k1) then return false
+		return level1[k1].has(k2, k3, k4)
+	end
+
+	# Remove all items
+	fun clear do level1.clear
 end
 
 # A map with a default value.
@@ -187,3 +347,373 @@ class DefaultMap[K, V]
 
 	redef fun provide_default_value(key) do return default
 end
+
+# An unrolled linked list
+#
+# A sequence implemented as a linked list of arrays.
+#
+# This data structure is similar to the `List` but it can benefit from
+# better cache performance, lower data overhead for the nodes metadata and
+# it spares the GC to allocate many small nodes.
+class UnrolledList[E]
+	super Sequence[E]
+
+	# Desired capacity for each nodes
+	#
+	# By default at `32`, it can be increased for very large lists.
+	#
+	# It can be modified anytime, but newly created nodes may still be assigned
+	# the same capacity of old nodes when created by `insert`.
+	var nodes_length = 32 is writable
+
+	private var head_node: UnrolledNode[E] = new UnrolledNode[E](nodes_length)
+
+	private var tail_node: UnrolledNode[E] = head_node
+
+	redef var length = 0
+
+	redef fun clear
+	do
+		head_node = new UnrolledNode[E](nodes_length)
+		tail_node = head_node
+		length = 0
+	end
+
+	# Out parameter of `node_at`
+	private var index_within_node = 0
+
+	private fun node_at(index: Int): UnrolledNode[E]
+	do
+		assert index >= 0 and index < length
+
+		var node = head_node
+		while index >= node.length do
+			index -= node.length
+			node = node.next.as(not null)
+		end
+
+		index_within_node = index
+		return node
+	end
+
+	private fun insert_node(node: UnrolledNode[E], prev, next: nullable UnrolledNode[E])
+	do
+		if prev != null then
+			prev.next = node
+		else head_node = node
+
+		if next != null then
+			next.prev = node
+		else tail_node = node
+
+		node.next = next
+		node.prev = prev
+	end
+
+	redef fun [](index)
+	do
+		var node = node_at(index)
+		index = index_within_node + node.head_index
+		return node.items[index]
+	end
+
+	redef fun []=(index, value)
+	do
+		var node = node_at(index)
+		index = index_within_node + node.head_index
+		node.items[index] = value
+	end
+
+	redef fun push(item)
+	do
+		var node = tail_node
+		if not node.full then
+			if node.tail_index < node.capacity then
+				# There's room at the tail
+				node.tail_index += 1
+			else
+				# Move everything over by `d`
+				assert node.head_index > 0
+				var d = node.head_index / 2 + 1
+				node.move_head(node.length, d)
+				for i in d.times do node.items[node.tail_index - i] = null
+				node.head_index -= d
+				node.tail_index += -d+1
+			end
+			node.items[node.tail_index-1] = item
+		else
+			# New node!
+			node = new UnrolledNode[E](nodes_length)
+			insert_node(node, tail_node, null)
+			node.tail_index = 1
+			node.items[0] = item
+		end
+		length += 1
+	end
+
+	redef fun unshift(item)
+	do
+		var node = head_node
+		if not node.full then
+			if node.head_index > 0 then
+				# There's room at the head
+				node.head_index -= 1
+			else
+				# Move everything over by `d`
+				assert node.tail_index < node.capacity
+				var d = (node.capacity-node.tail_index) / 2 + 1
+				node.move_tail(0, d)
+				for i in d.times do node.items[node.head_index+i] = null
+				node.head_index += d-1
+				node.tail_index += d
+			end
+			node.items[node.head_index] = item
+		else
+			# New node!
+			node = new UnrolledNode[E](nodes_length)
+			insert_node(node, null, head_node)
+			node.head_index = node.capacity-1
+			node.tail_index = node.capacity
+			node.items[node.capacity-1] = item
+		end
+		length += 1
+	end
+
+	redef fun pop
+	do
+		assert not_empty
+
+		var node = tail_node
+		while node.length == 0 do
+			# Delete empty node
+			var nullable_node = node.prev
+			assert is_not_empty: nullable_node != null
+			node = nullable_node
+			node.next = null
+			self.tail_node = node
+		end
+
+		var item = node.items[node.tail_index-1]
+		node.tail_index -= 1
+		length -= 1
+		return item
+	end
+
+	redef fun shift
+	do
+		assert not_empty
+
+		var node = head_node
+		while node.length == 0 do
+			# Delete empty node
+			var nullable_node = node.next
+			assert is_not_empty: nullable_node != null
+			node = nullable_node
+			node.prev = null
+			self.head_node = node
+		end
+
+		var item = node.items[node.head_index]
+		node.head_index += 1
+		length -= 1
+		return item
+	end
+
+	redef fun insert(item, index)
+	do
+		if index == length then
+			push item
+			return
+		end
+
+		var node = node_at(index)
+		index = index_within_node
+		if node.full then
+			# Move half to a new node
+			var new_node = new UnrolledNode[E](nodes_length.max(node.capacity))
+
+			# Plug in the new node
+			var next_node = node.next
+			insert_node(new_node, node, next_node)
+
+			# Move items at and after `index` to the new node
+			var to_displace = node.length-index
+			var offset = (new_node.capacity-to_displace)/2
+			for i in [0..to_displace[ do
+				new_node.items[offset+i] = node.items[index+i]
+				node.items[index+i] = null
+			end
+			new_node.head_index = offset
+			new_node.tail_index = offset + to_displace
+			node.tail_index -= to_displace
+
+			# Store `item`
+			if index > node.capacity / 2 then
+				new_node.items[offset-1] = item
+				new_node.head_index -= 1
+			else
+				node.items[node.head_index+index] = item
+				node.tail_index += 1
+			end
+		else
+			if node.tail_index < node.capacity then
+				# Move items towards the tail
+				node.move_tail(index, 1)
+				node.tail_index += 1
+				node.items[node.head_index + index] = item
+			else
+				# Move items towards the head
+				node.move_head(index, 1)
+				node.items[node.head_index + index-1] = item
+				node.head_index -= 1
+			end
+		end
+		length += 1
+	end
+
+	redef fun remove_at(index)
+	do
+		var node = node_at(index)
+		index = index_within_node + node.head_index
+
+		# Shift left all the elements after `index`
+		for i in [index+1 .. node.tail_index[ do
+			node.items[i-1] = node.items[i]
+		end
+		node.tail_index -= 1
+		node.items[node.tail_index] = null
+
+		length -= 1
+
+		var next_node = node.next
+		var prev_node = node.prev
+		if node.is_empty and (next_node != null or prev_node != null) then
+			# Empty and non-head or tail node, delete
+			if next_node != null then
+				next_node.prev = node.prev
+			else tail_node = prev_node.as(not null)
+
+			if prev_node != null then
+				prev_node.next = node.next
+			else head_node = next_node.as(not null)
+		end
+	end
+
+	redef fun iterator do return new UnrolledIterator[E](self)
+end
+
+# Node composing an `UnrolledList`
+#
+# Stores the elements in the `items` array. The elements in the `items` array
+# begin at `head_index` and end right before `tail_index`. The data is contiguous,
+# but there can be empty cells at the beginning and the end of the array.
+private class UnrolledNode[E]
+
+	var prev: nullable UnrolledNode[E] = null
+
+	var next: nullable UnrolledNode[E] = null
+
+	# Desired length of `items`
+	var capacity: Int
+
+	# `Array` of items in this node, filled with `null`
+	var items = new Array[nullable E].filled_with(null, capacity) is lazy
+
+	# Index of the first element in `items`
+	var head_index = 0
+
+	# Index after the last element in `items`
+	var tail_index = 0
+
+	fun length: Int do return tail_index - head_index
+
+	fun full: Bool do return length == capacity
+
+	fun is_empty: Bool do return tail_index == head_index
+
+	# Move towards the head all elements before `index` of `displace` cells
+	fun move_tail(index, displace: Int)
+	do
+		for i in [tail_index-1..head_index+index].step(-1) do
+			items[i+displace] = items[i]
+		end
+	end
+
+	# Move towards the tail all elements at and after `index` of `displace` cells
+	fun move_head(index, displace: Int)
+	do
+		for i in [head_index..head_index+index[ do
+			items[i-displace] = items[i]
+		end
+	end
+end
+
+private class UnrolledIterator[E]
+	super IndexedIterator[E]
+
+	var list: UnrolledList[E]
+
+	var node: nullable UnrolledNode[E] = list.head_node is lazy
+
+	# Index of the current `item`
+	redef var index = 0
+
+	# Index within the current `node`
+	var index_in_node: Int = node.head_index is lazy
+
+	redef fun item do return node.items[index_in_node]
+
+	redef fun is_ok do return index < list.length
+
+	redef fun next
+	do
+		index += 1
+		index_in_node += 1
+
+		if index_in_node >= node.tail_index then
+			node = node.next
+			if node != null then index_in_node = node.head_index
+		end
+	end
+end
+
+# Keep track of the best elements according to a distance value.
+#
+# ~~~
+# var bests = new BestDistance[String](5)
+# bests.update(10, "Too big")
+# assert bests.best_items.is_empty
+# bests.update(5, "Just fine")
+# bests.update(5, "Another one")
+# assert bests.best_items.has_exactly(["Just fine", "Another one"])
+# bests.update(2, "A better one")
+# bests.update(4, "Not good enough")
+# assert bests.best_distance == 2
+# assert bests.best_items.has_exactly(["A better one"])
+# ~~~
+class BestDistance[E]
+	# Current smallest distance
+	var best_distance: Int is writable
+
+	# Known elements with the smallest distance
+	var best_items = new Set[E] is writable
+
+	# Register a `candidate` with a `distance`
+	#
+	# * To high, it is ignored.
+	# * Equal to the current best, it is added
+	# * Better that them, is is the new best element
+	#
+	# Return `true` if the candidate is kept (alone or with other)
+	# returns `false` if the candidate is ignored.
+	fun update(distance: Int, candidate: E): Bool
+	do
+		if distance > best_distance then return false
+		if distance < best_distance then
+			best_distance = distance
+			best_items.clear
+		end
+		best_items.add candidate
+		return true
+	end
+end