| ... | @@ -9,17 +9,15 @@ const autoHash = std.hash.autoHash; | ... | @@ -9,17 +9,15 @@ const autoHash = std.hash.autoHash; |
| 9 | const Wyhash = std.hash.Wyhash; | 9 | const Wyhash = std.hash.Wyhash; |
| 10 | const Allocator = mem.Allocator; | 10 | const Allocator = mem.Allocator; |
| 11 | const builtin = @import("builtin"); | 11 | const builtin = @import("builtin"); |
| 12 | | 12 | const hash_map = @This(); |
| 13 | const want_modification_safety = std.debug.runtime_safety; | | |
| 14 | const debug_u32 = if (want_modification_safety) u32 else void; | | |
| 15 | | 13 | |
| 16 | pub fn AutoHashMap(comptime K: type, comptime V: type) type { | 14 | pub fn AutoHashMap(comptime K: type, comptime V: type) type { |
| 17 | return HashMap(K, V, getAutoHashFn(K), getAutoEqlFn(K)); | 15 | return HashMap(K, V, getAutoHashFn(K), getAutoEqlFn(K), autoEqlIsCheap(K)); |
| 18 | } | 16 | } |
| 19 | | 17 | |
| 20 | /// Builtin hashmap for strings as keys. | 18 | /// Builtin hashmap for strings as keys. |
| 21 | pub fn StringHashMap(comptime V: type) type { | 19 | pub fn StringHashMap(comptime V: type) type { |
| 22 | return HashMap([]const u8, V, hashString, eqlString); | 20 | return HashMap([]const u8, V, hashString, eqlString, true); |
| 23 | } | 21 | } |
| 24 | | 22 | |
| 25 | pub fn eqlString(a: []const u8, b: []const u8) bool { | 23 | pub fn eqlString(a: []const u8, b: []const u8) bool { |
| ... | @@ -30,422 +28,846 @@ pub fn hashString(s: []const u8) u32 { | ... | @@ -30,422 +28,846 @@ pub fn hashString(s: []const u8) u32 { |
| 30 | return @truncate(u32, std.hash.Wyhash.hash(0, s)); | 28 | return @truncate(u32, std.hash.Wyhash.hash(0, s)); |
| 31 | } | 29 | } |
| 32 | | 30 | |
| 33 | pub fn HashMap(comptime K: type, comptime V: type, comptime hash: fn (key: K) u32, comptime eql: fn (a: K, b: K) bool) type { | 31 | /// Insertion order is preserved. |
| | 32 | /// Deletions perform a "swap removal" on the entries list. |
| | 33 | /// Modifying the hash map while iterating is allowed, however one must understand |
| | 34 | /// the (well defined) behavior when mixing insertions and deletions with iteration. |
| | 35 | /// For a hash map that can be initialized directly that does not store an Allocator |
| | 36 | /// field, see `HashMapUnmanaged`. |
| | 37 | /// When `store_hash` is `false`, this data structure is biased towards cheap `eql` |
| | 38 | /// functions. It does not store each item's hash in the table. Setting `store_hash` |
| | 39 | /// to `true` incurs slightly more memory cost by storing each key's hash in the table |
| | 40 | /// but only has to call `eql` for hash collisions. |
| | 41 | pub fn HashMap( |
| | 42 | comptime K: type, |
| | 43 | comptime V: type, |
| | 44 | comptime hash: fn (key: K) u32, |
| | 45 | comptime eql: fn (a: K, b: K) bool, |
| | 46 | comptime store_hash: bool, |
| | 47 | ) type { |
| 34 | return struct { | 48 | return struct { |
| 35 | entries: []Entry, | 49 | unmanaged: Unmanaged, |
| 36 | size: usize, | | |
| 37 | max_distance_from_start_index: usize, | | |
| 38 | allocator: *Allocator, | 50 | allocator: *Allocator, |
| 39 | | 51 | |
| 40 | /// This is used to detect bugs where a hashtable is edited while an iterator is running. | 52 | pub const Unmanaged = HashMapUnmanaged(K, V, hash, eql, store_hash); |
| 41 | modification_count: debug_u32, | 53 | pub const Entry = Unmanaged.Entry; |
| 42 | | 54 | pub const Hash = Unmanaged.Hash; |
| 43 | const Self = @This(); | 55 | pub const GetOrPutResult = Unmanaged.GetOrPutResult; |
| 44 | | | |
| 45 | /// A *KV is a mutable pointer into this HashMap's internal storage. | | |
| 46 | /// Modifying the key is undefined behavior. | | |
| 47 | /// Modifying the value is harmless. | | |
| 48 | /// *KV pointers become invalid whenever this HashMap is modified, | | |
| 49 | /// and then any access to the *KV is undefined behavior. | | |
| 50 | pub const KV = struct { | | |
| 51 | key: K, | | |
| 52 | value: V, | | |
| 53 | }; | | |
| 54 | | | |
| 55 | const Entry = struct { | | |
| 56 | used: bool, | | |
| 57 | distance_from_start_index: usize, | | |
| 58 | kv: KV, | | |
| 59 | }; | | |
| 60 | | | |
| 61 | pub const GetOrPutResult = struct { | | |
| 62 | kv: *KV, | | |
| 63 | found_existing: bool, | | |
| 64 | }; | | |
| 65 | | 56 | |
| | 57 | /// Deprecated. Iterate using `items`. |
| 66 | pub const Iterator = struct { | 58 | pub const Iterator = struct { |
| 67 | hm: *const Self, | 59 | hm: *const Self, |
| 68 | // how many items have we returned | 60 | /// Iterator through the entry array. |
| 69 | count: usize, | | |
| 70 | // iterator through the entry array | | |
| 71 | index: usize, | 61 | index: usize, |
| 72 | // used to detect concurrent modification | | |
| 73 | initial_modification_count: debug_u32, | | |
| 74 | | 62 | |
| 75 | pub fn next(it: *Iterator) ?*KV { | 63 | pub fn next(it: *Iterator) ?*Entry { |
| 76 | if (want_modification_safety) { | 64 | if (it.index >= it.hm.unmanaged.entries.items.len) return null; |
| 77 | assert(it.initial_modification_count == it.hm.modification_count); // concurrent modification | 65 | const result = &it.hm.unmanaged.entries.items[it.index]; |
| 78 | } | 66 | it.index += 1; |
| 79 | if (it.count >= it.hm.size) return null; | 67 | return result; |
| 80 | while (it.index < it.hm.entries.len) : (it.index += 1) { | | |
| 81 | const entry = &it.hm.entries[it.index]; | | |
| 82 | if (entry.used) { | | |
| 83 | it.index += 1; | | |
| 84 | it.count += 1; | | |
| 85 | return &entry.kv; | | |
| 86 | } | | |
| 87 | } | | |
| 88 | unreachable; // no next item | | |
| 89 | } | 68 | } |
| 90 | | 69 | |
| 91 | // Reset the iterator to the initial index | 70 | /// Reset the iterator to the initial index |
| 92 | pub fn reset(it: *Iterator) void { | 71 | pub fn reset(it: *Iterator) void { |
| 93 | it.count = 0; | | |
| 94 | it.index = 0; | 72 | it.index = 0; |
| 95 | // Resetting the modification count too | | |
| 96 | it.initial_modification_count = it.hm.modification_count; | | |
| 97 | } | 73 | } |
| 98 | }; | 74 | }; |
| 99 | | 75 | |
| | 76 | const Self = @This(); |
| | 77 | const Index = Unmanaged.Index; |
| | 78 | |
| 100 | pub fn init(allocator: *Allocator) Self { | 79 | pub fn init(allocator: *Allocator) Self { |
| 101 | return Self{ | 80 | return .{ |
| 102 | .entries = &[_]Entry{}, | 81 | .unmanaged = .{}, |
| 103 | .allocator = allocator, | 82 | .allocator = allocator, |
| 104 | .size = 0, | | |
| 105 | .max_distance_from_start_index = 0, | | |
| 106 | .modification_count = if (want_modification_safety) 0 else {}, | | |
| 107 | }; | 83 | }; |
| 108 | } | 84 | } |
| 109 | | 85 | |
| 110 | pub fn deinit(hm: Self) void { | 86 | pub fn deinit(self: *Self) void { |
| 111 | hm.allocator.free(hm.entries); | 87 | self.unmanaged.deinit(self.allocator); |
| | 88 | self.* = undefined; |
| 112 | } | 89 | } |
| 113 | | 90 | |
| 114 | pub fn clear(hm: *Self) void { | 91 | pub fn clearRetainingCapacity(self: *Self) void { |
| 115 | for (hm.entries) |*entry| { | 92 | return self.unmanaged.clearRetainingCapacity(); |
| 116 | entry.used = false; | | |
| 117 | } | | |
| 118 | hm.size = 0; | | |
| 119 | hm.max_distance_from_start_index = 0; | | |
| 120 | hm.incrementModificationCount(); | | |
| 121 | } | 93 | } |
| 122 | | 94 | |
| | 95 | pub fn clearAndFree(self: *Self, allocator: *Allocator) void { |
| | 96 | return self.unmanaged.clearAndFree(self.allocator); |
| | 97 | } |
| | 98 | |
| | 99 | /// Deprecated. Use `items().len`. |
| 123 | pub fn count(self: Self) usize { | 100 | pub fn count(self: Self) usize { |
| 124 | return self.size; | 101 | return self.items().len; |
| | 102 | } |
| | 103 | |
| | 104 | /// Deprecated. Iterate using `items`. |
| | 105 | pub fn iterator(self: *const Self) Iterator { |
| | 106 | return Iterator{ |
| | 107 | .hm = self, |
| | 108 | .index = 0, |
| | 109 | }; |
| 125 | } | 110 | } |
| 126 | | 111 | |
| 127 | /// If key exists this function cannot fail. | 112 | /// If key exists this function cannot fail. |
| 128 | /// If there is an existing item with `key`, then the result | 113 | /// If there is an existing item with `key`, then the result |
| 129 | /// kv pointer points to it, and found_existing is true. | 114 | /// `Entry` pointer points to it, and found_existing is true. |
| 130 | /// Otherwise, puts a new item with undefined value, and | 115 | /// Otherwise, puts a new item with undefined value, and |
| 131 | /// the kv pointer points to it. Caller should then initialize | 116 | /// the `Entry` pointer points to it. Caller should then initialize |
| 132 | /// the data. | 117 | /// the value (but not the key). |
| 133 | pub fn getOrPut(self: *Self, key: K) !GetOrPutResult { | 118 | pub fn getOrPut(self: *Self, key: K) !GetOrPutResult { |
| 134 | // TODO this implementation can be improved - we should only | 119 | return self.unmanaged.getOrPut(self.allocator, key); |
| 135 | // have to hash once and find the entry once. | 120 | } |
| 136 | if (self.get(key)) |kv| { | 121 | |
| 137 | return GetOrPutResult{ | 122 | /// If there is an existing item with `key`, then the result |
| 138 | .kv = kv, | 123 | /// `Entry` pointer points to it, and found_existing is true. |
| 139 | .found_existing = true, | 124 | /// Otherwise, puts a new item with undefined value, and |
| 140 | }; | 125 | /// the `Entry` pointer points to it. Caller should then initialize |
| 141 | } | 126 | /// the value (but not the key). |
| 142 | self.incrementModificationCount(); | 127 | /// If a new entry needs to be stored, this function asserts there |
| 143 | try self.autoCapacity(); | 128 | /// is enough capacity to store it. |
| 144 | const put_result = self.internalPut(key); | 129 | pub fn getOrPutAssumeCapacity(self: *Self, key: K) GetOrPutResult { |
| 145 | assert(put_result.old_kv == null); | 130 | return self.unmanaged.getOrPutAssumeCapacity(key); |
| 146 | return GetOrPutResult{ | 131 | } |
| 147 | .kv = &put_result.new_entry.kv, | 132 | |
| 148 | .found_existing = false, | 133 | pub fn getOrPutValue(self: *Self, key: K, value: V) !*Entry { |
| | 134 | return self.unmanaged.getOrPutValue(self.allocator, key, value); |
| | 135 | } |
| | 136 | |
| | 137 | /// Increases capacity, guaranteeing that insertions up until the |
| | 138 | /// `expected_count` will not cause an allocation, and therefore cannot fail. |
| | 139 | pub fn ensureCapacity(self: *Self, new_capacity: usize) !void { |
| | 140 | return self.unmanaged.ensureCapacity(self.allocator, new_capacity); |
| | 141 | } |
| | 142 | |
| | 143 | /// Returns the number of total elements which may be present before it is |
| | 144 | /// no longer guaranteed that no allocations will be performed. |
| | 145 | pub fn capacity(self: *Self) usize { |
| | 146 | return self.unmanaged.capacity(); |
| | 147 | } |
| | 148 | |
| | 149 | /// Clobbers any existing data. To detect if a put would clobber |
| | 150 | /// existing data, see `getOrPut`. |
| | 151 | pub fn put(self: *Self, key: K, value: V) !void { |
| | 152 | return self.unmanaged.put(self.allocator, key, value); |
| | 153 | } |
| | 154 | |
| | 155 | /// Inserts a key-value pair into the hash map, asserting that no previous |
| | 156 | /// entry with the same key is already present |
| | 157 | pub fn putNoClobber(self: *Self, key: K, value: V) !void { |
| | 158 | return self.unmanaged.putNoClobber(self.allocator, key, value); |
| | 159 | } |
| | 160 | |
| | 161 | /// Asserts there is enough capacity to store the new key-value pair. |
| | 162 | /// Clobbers any existing data. To detect if a put would clobber |
| | 163 | /// existing data, see `getOrPutAssumeCapacity`. |
| | 164 | pub fn putAssumeCapacity(self: *Self, key: K, value: V) void { |
| | 165 | return self.unmanaged.putAssumeCapacity(key, value); |
| | 166 | } |
| | 167 | |
| | 168 | /// Asserts there is enough capacity to store the new key-value pair. |
| | 169 | /// Asserts that it does not clobber any existing data. |
| | 170 | /// To detect if a put would clobber existing data, see `getOrPutAssumeCapacity`. |
| | 171 | pub fn putAssumeCapacityNoClobber(self: *Self, key: K, value: V) void { |
| | 172 | return self.unmanaged.putAssumeCapacityNoClobber(key, value); |
| | 173 | } |
| | 174 | |
| | 175 | /// Inserts a new `Entry` into the hash map, returning the previous one, if any. |
| | 176 | pub fn fetchPut(self: *Self, key: K, value: V) !?Entry { |
| | 177 | return self.unmanaged.fetchPut(self.allocator, key, value); |
| | 178 | } |
| | 179 | |
| | 180 | /// Inserts a new `Entry` into the hash map, returning the previous one, if any. |
| | 181 | /// If insertion happuns, asserts there is enough capacity without allocating. |
| | 182 | pub fn fetchPutAssumeCapacity(self: *Self, key: K, value: V) ?Entry { |
| | 183 | return self.unmanaged.fetchPutAssumeCapacity(key, value); |
| | 184 | } |
| | 185 | |
| | 186 | pub fn getEntry(self: Self, key: K) ?*Entry { |
| | 187 | return self.unmanaged.getEntry(key); |
| | 188 | } |
| | 189 | |
| | 190 | pub fn get(self: Self, key: K) ?V { |
| | 191 | return self.unmanaged.get(key); |
| | 192 | } |
| | 193 | |
| | 194 | pub fn contains(self: Self, key: K) bool { |
| | 195 | return self.unmanaged.contains(key); |
| | 196 | } |
| | 197 | |
| | 198 | /// If there is an `Entry` with a matching key, it is deleted from |
| | 199 | /// the hash map, and then returned from this function. |
| | 200 | pub fn remove(self: *Self, key: K) ?Entry { |
| | 201 | return self.unmanaged.remove(key); |
| | 202 | } |
| | 203 | |
| | 204 | /// Asserts there is an `Entry` with matching key, deletes it from the hash map, |
| | 205 | /// and discards it. |
| | 206 | pub fn removeAssertDiscard(self: *Self, key: K) void { |
| | 207 | return self.unmanaged.removeAssertDiscard(key); |
| | 208 | } |
| | 209 | |
| | 210 | pub fn items(self: Self) []Entry { |
| | 211 | return self.unmanaged.items(); |
| | 212 | } |
| | 213 | |
| | 214 | pub fn clone(self: Self) !Self { |
| | 215 | var other = try self.unmanaged.clone(self.allocator); |
| | 216 | return other.promote(self.allocator); |
| | 217 | } |
| | 218 | }; |
| | 219 | } |
| | 220 | |
| | 221 | /// General purpose hash table. |
| | 222 | /// Insertion order is preserved. |
| | 223 | /// Deletions perform a "swap removal" on the entries list. |
| | 224 | /// Modifying the hash map while iterating is allowed, however one must understand |
| | 225 | /// the (well defined) behavior when mixing insertions and deletions with iteration. |
| | 226 | /// This type does not store an Allocator field - the Allocator must be passed in |
| | 227 | /// with each function call that requires it. See `HashMap` for a type that stores |
| | 228 | /// an Allocator field for convenience. |
| | 229 | /// Can be initialized directly using the default field values. |
| | 230 | /// This type is designed to have low overhead for small numbers of entries. When |
| | 231 | /// `store_hash` is `false` and the number of entries in the map is less than 9, |
| | 232 | /// the overhead cost of using `HashMapUnmanaged` rather than `std.ArrayList` is |
| | 233 | /// only a single pointer-sized integer. |
| | 234 | /// When `store_hash` is `false`, this data structure is biased towards cheap `eql` |
| | 235 | /// functions. It does not store each item's hash in the table. Setting `store_hash` |
| | 236 | /// to `true` incurs slightly more memory cost by storing each key's hash in the table |
| | 237 | /// but guarantees only one call to `eql` per insertion/deletion. |
| | 238 | pub fn HashMapUnmanaged( |
| | 239 | comptime K: type, |
| | 240 | comptime V: type, |
| | 241 | comptime hash: fn (key: K) u32, |
| | 242 | comptime eql: fn (a: K, b: K) bool, |
| | 243 | comptime store_hash: bool, |
| | 244 | ) type { |
| | 245 | return struct { |
| | 246 | /// It is permitted to access this field directly. |
| | 247 | entries: std.ArrayListUnmanaged(Entry) = .{}, |
| | 248 | |
| | 249 | /// When entries length is less than `linear_scan_max`, this remains `null`. |
| | 250 | /// Once entries length grows big enough, this field is allocated. There is |
| | 251 | /// an IndexHeader followed by an array of Index(I) structs, where I is defined |
| | 252 | /// by how many total indexes there are. |
| | 253 | index_header: ?*IndexHeader = null, |
| | 254 | |
| | 255 | /// Modifying the key is illegal behavior. |
| | 256 | /// Modifying the value is allowed. |
| | 257 | /// Entry pointers become invalid whenever this HashMap is modified, |
| | 258 | /// unless `ensureCapacity` was previously used. |
| | 259 | pub const Entry = struct { |
| | 260 | /// This field is `void` if `store_hash` is `false`. |
| | 261 | hash: Hash, |
| | 262 | key: K, |
| | 263 | value: V, |
| | 264 | }; |
| | 265 | |
| | 266 | pub const Hash = if (store_hash) u32 else void; |
| | 267 | |
| | 268 | pub const GetOrPutResult = struct { |
| | 269 | entry: *Entry, |
| | 270 | found_existing: bool, |
| | 271 | }; |
| | 272 | |
| | 273 | pub const Managed = HashMap(K, V, hash, eql, store_hash); |
| | 274 | |
| | 275 | const Self = @This(); |
| | 276 | |
| | 277 | const linear_scan_max = 8; |
| | 278 | |
| | 279 | pub fn promote(self: Self, allocator: *Allocator) Managed { |
| | 280 | return .{ |
| | 281 | .unmanaged = self, |
| | 282 | .allocator = allocator, |
| 149 | }; | 283 | }; |
| 150 | } | 284 | } |
| 151 | | 285 | |
| 152 | pub fn getOrPutValue(self: *Self, key: K, value: V) !*KV { | 286 | pub fn deinit(self: *Self, allocator: *Allocator) void { |
| 153 | const res = try self.getOrPut(key); | 287 | self.entries.deinit(allocator); |
| 154 | if (!res.found_existing) | 288 | if (self.index_header) |header| { |
| 155 | res.kv.value = value; | 289 | header.free(allocator); |
| | 290 | } |
| | 291 | self.* = undefined; |
| | 292 | } |
| 156 | | 293 | |
| 157 | return res.kv; | 294 | pub fn clearRetainingCapacity(self: *Self) void { |
| | 295 | self.entries.items.len = 0; |
| | 296 | if (self.header) |header| { |
| | 297 | header.max_distance_from_start_index = 0; |
| | 298 | const indexes = header.indexes(u8); |
| | 299 | @memset(indexes.ptr, 0xff, indexes.len); |
| | 300 | } |
| 158 | } | 301 | } |
| 159 | | 302 | |
| 160 | fn optimizedCapacity(expected_count: usize) usize { | 303 | pub fn clearAndFree(self: *Self, allocator: *Allocator) void { |
| 161 | // ensure that the hash map will be at most 60% full if | 304 | self.entries.shrink(allocator, 0); |
| 162 | // expected_count items are put into it | 305 | if (self.header) |header| { |
| 163 | var optimized_capacity = expected_count * 5 / 3; | 306 | header.free(allocator); |
| 164 | // an overflow here would mean the amount of memory required would not | 307 | self.header = null; |
| 165 | // be representable in the address space | 308 | } |
| 166 | return math.ceilPowerOfTwo(usize, optimized_capacity) catch unreachable; | | |
| 167 | } | 309 | } |
| 168 | | 310 | |
| 169 | /// Increases capacity so that the hash map will be at most | 311 | /// If key exists this function cannot fail. |
| 170 | /// 60% full when expected_count items are put into it | 312 | /// If there is an existing item with `key`, then the result |
| 171 | pub fn ensureCapacity(self: *Self, expected_count: usize) !void { | 313 | /// `Entry` pointer points to it, and found_existing is true. |
| 172 | if (expected_count == 0) return; | 314 | /// Otherwise, puts a new item with undefined value, and |
| 173 | const optimized_capacity = optimizedCapacity(expected_count); | 315 | /// the `Entry` pointer points to it. Caller should then initialize |
| 174 | return self.ensureCapacityExact(optimized_capacity); | 316 | /// the value (but not the key). |
| | 317 | pub fn getOrPut(self: *Self, allocator: *Allocator, key: K) !GetOrPutResult { |
| | 318 | self.ensureCapacity(allocator, self.entries.items.len + 1) catch |err| { |
| | 319 | // "If key exists this function cannot fail." |
| | 320 | return GetOrPutResult{ |
| | 321 | .entry = self.getEntry(key) orelse return err, |
| | 322 | .found_existing = true, |
| | 323 | }; |
| | 324 | }; |
| | 325 | return self.getOrPutAssumeCapacity(key); |
| 175 | } | 326 | } |
| 176 | | 327 | |
| 177 | /// Sets the capacity to the new capacity if the new | 328 | /// If there is an existing item with `key`, then the result |
| 178 | /// capacity is greater than the current capacity. | 329 | /// `Entry` pointer points to it, and found_existing is true. |
| 179 | /// New capacity must be a power of two. | 330 | /// Otherwise, puts a new item with undefined value, and |
| 180 | fn ensureCapacityExact(self: *Self, new_capacity: usize) !void { | 331 | /// the `Entry` pointer points to it. Caller should then initialize |
| 181 | // capacity must always be a power of two to allow for modulo | 332 | /// the value (but not the key). |
| 182 | // optimization in the constrainIndex fn | 333 | /// If a new entry needs to be stored, this function asserts there |
| 183 | assert(math.isPowerOfTwo(new_capacity)); | 334 | /// is enough capacity to store it. |
| | 335 | pub fn getOrPutAssumeCapacity(self: *Self, key: K) GetOrPutResult { |
| | 336 | const header = self.index_header orelse { |
| | 337 | // Linear scan. |
| | 338 | const h = if (store_hash) hash(key) else {}; |
| | 339 | for (self.entries.items) |*item| { |
| | 340 | if (item.hash == h and eql(key, item.key)) { |
| | 341 | return GetOrPutResult{ |
| | 342 | .entry = item, |
| | 343 | .found_existing = true, |
| | 344 | }; |
| | 345 | } |
| | 346 | } |
| | 347 | const new_entry = self.entries.addOneAssumeCapacity(); |
| | 348 | new_entry.* = .{ |
| | 349 | .hash = if (store_hash) h else {}, |
| | 350 | .key = key, |
| | 351 | .value = undefined, |
| | 352 | }; |
| | 353 | return GetOrPutResult{ |
| | 354 | .entry = new_entry, |
| | 355 | .found_existing = false, |
| | 356 | }; |
| | 357 | }; |
| 184 | | 358 | |
| 185 | if (new_capacity <= self.entries.len) { | 359 | switch (header.capacityIndexType()) { |
| 186 | return; | 360 | .u8 => return self.getOrPutInternal(key, header, u8), |
| | 361 | .u16 => return self.getOrPutInternal(key, header, u16), |
| | 362 | .u32 => return self.getOrPutInternal(key, header, u32), |
| | 363 | .usize => return self.getOrPutInternal(key, header, usize), |
| 187 | } | 364 | } |
| | 365 | } |
| 188 | | 366 | |
| 189 | const old_entries = self.entries; | 367 | pub fn getOrPutValue(self: *Self, allocator: *Allocator, key: K, value: V) !*Entry { |
| 190 | try self.initCapacity(new_capacity); | 368 | const res = try self.getOrPut(allocator, key); |
| 191 | self.incrementModificationCount(); | 369 | if (!res.found_existing) |
| 192 | if (old_entries.len > 0) { | 370 | res.entry.value = value; |
| 193 | // dump all of the old elements into the new table | 371 | |
| 194 | for (old_entries) |*old_entry| { | 372 | return res.entry; |
| 195 | if (old_entry.used) { | 373 | } |
| 196 | self.internalPut(old_entry.kv.key).new_entry.kv.value = old_entry.kv.value; | 374 | |
| | 375 | /// Increases capacity, guaranteeing that insertions up until the |
| | 376 | /// `expected_count` will not cause an allocation, and therefore cannot fail. |
| | 377 | pub fn ensureCapacity(self: *Self, allocator: *Allocator, new_capacity: usize) !void { |
| | 378 | try self.entries.ensureCapacity(allocator, new_capacity); |
| | 379 | if (new_capacity <= linear_scan_max) return; |
| | 380 | |
| | 381 | // Resize if indexes would be more than 75% full. |
| | 382 | const needed_len = new_capacity * 4 / 3; |
| | 383 | if (self.index_header) |header| { |
| | 384 | if (needed_len > header.indexes_len) { |
| | 385 | var new_indexes_len = header.indexes_len; |
| | 386 | while (true) { |
| | 387 | new_indexes_len += new_indexes_len / 2 + 8; |
| | 388 | if (new_indexes_len >= needed_len) break; |
| 197 | } | 389 | } |
| | 390 | const new_header = try IndexHeader.alloc(allocator, new_indexes_len); |
| | 391 | self.insertAllEntriesIntoNewHeader(new_header); |
| | 392 | header.free(allocator); |
| | 393 | self.index_header = new_header; |
| 198 | } | 394 | } |
| 199 | self.allocator.free(old_entries); | 395 | } else { |
| | 396 | const header = try IndexHeader.alloc(allocator, needed_len); |
| | 397 | self.insertAllEntriesIntoNewHeader(header); |
| | 398 | self.index_header = header; |
| 200 | } | 399 | } |
| 201 | } | 400 | } |
| 202 | | 401 | |
| 203 | /// Returns the kv pair that was already there. | 402 | /// Returns the number of total elements which may be present before it is |
| 204 | pub fn put(self: *Self, key: K, value: V) !?KV { | 403 | /// no longer guaranteed that no allocations will be performed. |
| 205 | try self.autoCapacity(); | 404 | pub fn capacity(self: Self) usize { |
| 206 | return putAssumeCapacity(self, key, value); | 405 | const entry_cap = self.entries.capacity; |
| | 406 | const header = self.index_header orelse return math.min(linear_scan_max, entry_cap); |
| | 407 | const indexes_cap = (header.indexes_len + 1) * 3 / 4; |
| | 408 | return math.min(entry_cap, indexes_cap); |
| 207 | } | 409 | } |
| 208 | | 410 | |
| 209 | /// Calls put() and asserts that no kv pair is clobbered. | 411 | /// Clobbers any existing data. To detect if a put would clobber |
| 210 | pub fn putNoClobber(self: *Self, key: K, value: V) !void { | 412 | /// existing data, see `getOrPut`. |
| 211 | assert((try self.put(key, value)) == null); | 413 | pub fn put(self: *Self, allocator: *Allocator, key: K, value: V) !void { |
| | 414 | const result = try self.getOrPut(allocator, key); |
| | 415 | result.entry.value = value; |
| 212 | } | 416 | } |
| 213 | | 417 | |
| 214 | pub fn putAssumeCapacity(self: *Self, key: K, value: V) ?KV { | 418 | /// Inserts a key-value pair into the hash map, asserting that no previous |
| 215 | assert(self.count() < self.entries.len); | 419 | /// entry with the same key is already present |
| 216 | self.incrementModificationCount(); | 420 | pub fn putNoClobber(self: *Self, allocator: *Allocator, key: K, value: V) !void { |
| | 421 | const result = try self.getOrPut(allocator, key); |
| | 422 | assert(!result.found_existing); |
| | 423 | result.entry.value = value; |
| | 424 | } |
| 217 | | 425 | |
| 218 | const put_result = self.internalPut(key); | 426 | /// Asserts there is enough capacity to store the new key-value pair. |
| 219 | put_result.new_entry.kv.value = value; | 427 | /// Clobbers any existing data. To detect if a put would clobber |
| 220 | return put_result.old_kv; | 428 | /// existing data, see `getOrPutAssumeCapacity`. |
| | 429 | pub fn putAssumeCapacity(self: *Self, key: K, value: V) void { |
| | 430 | const result = self.getOrPutAssumeCapacity(key); |
| | 431 | result.entry.value = value; |
| 221 | } | 432 | } |
| 222 | | 433 | |
| | 434 | /// Asserts there is enough capacity to store the new key-value pair. |
| | 435 | /// Asserts that it does not clobber any existing data. |
| | 436 | /// To detect if a put would clobber existing data, see `getOrPutAssumeCapacity`. |
| 223 | pub fn putAssumeCapacityNoClobber(self: *Self, key: K, value: V) void { | 437 | pub fn putAssumeCapacityNoClobber(self: *Self, key: K, value: V) void { |
| 224 | assert(self.putAssumeCapacity(key, value) == null); | 438 | const result = self.getOrPutAssumeCapacity(key); |
| | 439 | assert(!result.found_existing); |
| | 440 | result.entry.value = value; |
| 225 | } | 441 | } |
| 226 | | 442 | |
| 227 | pub fn get(hm: *const Self, key: K) ?*KV { | 443 | /// Inserts a new `Entry` into the hash map, returning the previous one, if any. |
| 228 | if (hm.entries.len == 0) { | 444 | pub fn fetchPut(self: *Self, allocator: *Allocator, key: K, value: V) !?Entry { |
| | 445 | const gop = try self.getOrPut(allocator, key); |
| | 446 | var result: ?Entry = null; |
| | 447 | if (gop.found_existing) { |
| | 448 | result = gop.entry.*; |
| | 449 | } |
| | 450 | gop.entry.value = value; |
| | 451 | return result; |
| | 452 | } |
| | 453 | |
| | 454 | /// Inserts a new `Entry` into the hash map, returning the previous one, if any. |
| | 455 | /// If insertion happuns, asserts there is enough capacity without allocating. |
| | 456 | pub fn fetchPutAssumeCapacity(self: *Self, key: K, value: V) ?Entry { |
| | 457 | const gop = self.getOrPutAssumeCapacity(key); |
| | 458 | var result: ?Entry = null; |
| | 459 | if (gop.found_existing) { |
| | 460 | result = gop.entry.*; |
| | 461 | } |
| | 462 | gop.entry.value = value; |
| | 463 | return result; |
| | 464 | } |
| | 465 | |
| | 466 | pub fn getEntry(self: Self, key: K) ?*Entry { |
| | 467 | const header = self.index_header orelse { |
| | 468 | // Linear scan. |
| | 469 | const h = if (store_hash) hash(key) else {}; |
| | 470 | for (self.entries.items) |*item| { |
| | 471 | if (item.hash == h and eql(key, item.key)) { |
| | 472 | return item; |
| | 473 | } |
| | 474 | } |
| 229 | return null; | 475 | return null; |
| | 476 | }; |
| | 477 | |
| | 478 | switch (header.capacityIndexType()) { |
| | 479 | .u8 => return self.getInternal(key, header, u8), |
| | 480 | .u16 => return self.getInternal(key, header, u16), |
| | 481 | .u32 => return self.getInternal(key, header, u32), |
| | 482 | .usize => return self.getInternal(key, header, usize), |
| 230 | } | 483 | } |
| 231 | return hm.internalGet(key); | | |
| 232 | } | 484 | } |
| 233 | | 485 | |
| 234 | pub fn getValue(hm: *const Self, key: K) ?V { | 486 | pub fn get(self: Self, key: K) ?V { |
| 235 | return if (hm.get(key)) |kv| kv.value else null; | 487 | return if (self.getEntry(key)) |entry| entry.value else null; |
| 236 | } | 488 | } |
| 237 | | 489 | |
| 238 | pub fn contains(hm: *const Self, key: K) bool { | 490 | pub fn contains(self: Self, key: K) bool { |
| 239 | return hm.get(key) != null; | 491 | return self.getEntry(key) != null; |
| 240 | } | 492 | } |
| 241 | | 493 | |
| 242 | /// Returns any kv pair that was removed. | 494 | /// If there is an `Entry` with a matching key, it is deleted from |
| 243 | pub fn remove(hm: *Self, key: K) ?KV { | 495 | /// the hash map, and then returned from this function. |
| 244 | if (hm.entries.len == 0) return null; | 496 | pub fn remove(self: *Self, key: K) ?Entry { |
| 245 | hm.incrementModificationCount(); | 497 | const header = self.index_header orelse { |
| 246 | const start_index = hm.keyToIndex(key); | 498 | // Linear scan. |
| 247 | { | 499 | const h = if (store_hash) hash(key) else {}; |
| 248 | var roll_over: usize = 0; | 500 | for (self.entries.items) |item, i| { |
| 249 | while (roll_over <= hm.max_distance_from_start_index) : (roll_over += 1) { | 501 | if (item.hash == h and eql(key, item.key)) { |
| 250 | const index = hm.constrainIndex(start_index + roll_over); | 502 | return self.entries.swapRemove(i); |
| 251 | var entry = &hm.entries[index]; | | |
| 252 | | | |
| 253 | if (!entry.used) return null; | | |
| 254 | | | |
| 255 | if (!eql(entry.kv.key, key)) continue; | | |
| 256 | | | |
| 257 | const removed_kv = entry.kv; | | |
| 258 | while (roll_over < hm.entries.len) : (roll_over += 1) { | | |
| 259 | const next_index = hm.constrainIndex(start_index + roll_over + 1); | | |
| 260 | const next_entry = &hm.entries[next_index]; | | |
| 261 | if (!next_entry.used or next_entry.distance_from_start_index == 0) { | | |
| 262 | entry.used = false; | | |
| 263 | hm.size -= 1; | | |
| 264 | return removed_kv; | | |
| 265 | } | | |
| 266 | entry.* = next_entry.*; | | |
| 267 | entry.distance_from_start_index -= 1; | | |
| 268 | entry = next_entry; | | |
| 269 | } | 503 | } |
| 270 | unreachable; // shifting everything in the table | | |
| 271 | } | 504 | } |
| | 505 | return null; |
| | 506 | }; |
| | 507 | switch (header.capacityIndexType()) { |
| | 508 | .u8 => return self.removeInternal(key, header, u8), |
| | 509 | .u16 => return self.removeInternal(key, header, u16), |
| | 510 | .u32 => return self.removeInternal(key, header, u32), |
| | 511 | .usize => return self.removeInternal(key, header, usize), |
| 272 | } | 512 | } |
| 273 | return null; | | |
| 274 | } | 513 | } |
| 275 | | 514 | |
| 276 | /// Calls remove(), asserts that a kv pair is removed, and discards it. | 515 | /// Asserts there is an `Entry` with matching key, deletes it from the hash map, |
| 277 | pub fn removeAssertDiscard(hm: *Self, key: K) void { | 516 | /// and discards it. |
| 278 | assert(hm.remove(key) != null); | 517 | pub fn removeAssertDiscard(self: *Self, key: K) void { |
| | 518 | assert(self.remove(key) != null); |
| 279 | } | 519 | } |
| 280 | | 520 | |
| 281 | pub fn iterator(hm: *const Self) Iterator { | 521 | pub fn items(self: Self) []Entry { |
| 282 | return Iterator{ | 522 | return self.entries.items; |
| 283 | .hm = hm, | | |
| 284 | .count = 0, | | |
| 285 | .index = 0, | | |
| 286 | .initial_modification_count = hm.modification_count, | | |
| 287 | }; | | |
| 288 | } | 523 | } |
| 289 | | 524 | |
| 290 | pub fn clone(self: Self) !Self { | 525 | pub fn clone(self: Self, allocator: *Allocator) !Self { |
| 291 | var other = Self.init(self.allocator); | 526 | // TODO this can be made more efficient by directly allocating |
| 292 | try other.initCapacity(self.entries.len); | 527 | // the memory slices and memcpying the elements. |
| 293 | var it = self.iterator(); | 528 | var other = Self.init(); |
| 294 | while (it.next()) |entry| { | 529 | try other.initCapacity(allocator, self.entries.len); |
| 295 | try other.putNoClobber(entry.key, entry.value); | 530 | for (self.entries.items) |entry| { |
| | 531 | other.putAssumeCapacityNoClobber(entry.key, entry.value); |
| 296 | } | 532 | } |
| 297 | return other; | 533 | return other; |
| 298 | } | 534 | } |
| 299 | | 535 | |
| 300 | fn autoCapacity(self: *Self) !void { | 536 | fn removeInternal(self: *Self, key: K, header: *IndexHeader, comptime I: type) ?Entry { |
| 301 | if (self.entries.len == 0) { | 537 | const indexes = header.indexes(I); |
| 302 | return self.ensureCapacityExact(16); | 538 | const h = hash(key); |
| 303 | } | 539 | const start_index = header.hashToIndex(h); |
| 304 | // if we get too full (60%), double the capacity | 540 | var roll_over: usize = 0; |
| 305 | if (self.size * 5 >= self.entries.len * 3) { | 541 | while (roll_over <= header.max_distance_from_start_index) : (roll_over += 1) { |
| 306 | return self.ensureCapacityExact(self.entries.len * 2); | 542 | const index_index = (start_index + roll_over) % header.indexes_len; |
| 307 | } | 543 | var index = &indexes[index_index]; |
| 308 | } | 544 | if (index.isEmpty()) |
| | 545 | return null; |
| | 546 | |
| | 547 | const entry = &self.entries.items[index.entry_index]; |
| | 548 | |
| | 549 | const hash_match = if (store_hash) h == entry.hash else true; |
| | 550 | if (!hash_match or !eql(key, entry.key)) |
| | 551 | continue; |
| 309 | | 552 | |
| 310 | fn initCapacity(hm: *Self, capacity: usize) !void { | 553 | const removed_entry = self.entries.swapRemove(index.entry_index); |
| 311 | hm.entries = try hm.allocator.alloc(Entry, capacity); | 554 | if (self.entries.items.len > 0 and self.entries.items.len != index.entry_index) { |
| 312 | hm.size = 0; | 555 | // Because of the swap remove, now we need to update the index that was |
| 313 | hm.max_distance_from_start_index = 0; | 556 | // pointing to the last entry and is now pointing to this removed item slot. |
| 314 | for (hm.entries) |*entry| { | 557 | self.updateEntryIndex(header, self.entries.items.len, index.entry_index, I, indexes); |
| 315 | entry.used = false; | 558 | } |
| | 559 | |
| | 560 | // Now we have to shift over the following indexes. |
| | 561 | roll_over += 1; |
| | 562 | while (roll_over < header.indexes_len) : (roll_over += 1) { |
| | 563 | const next_index_index = (start_index + roll_over) % header.indexes_len; |
| | 564 | const next_index = &indexes[next_index_index]; |
| | 565 | if (next_index.isEmpty() or next_index.distance_from_start_index == 0) { |
| | 566 | index.setEmpty(); |
| | 567 | return removed_entry; |
| | 568 | } |
| | 569 | index.* = next_index.*; |
| | 570 | index.distance_from_start_index -= 1; |
| | 571 | index = next_index; |
| | 572 | } |
| | 573 | unreachable; |
| 316 | } | 574 | } |
| | 575 | return null; |
| 317 | } | 576 | } |
| 318 | | 577 | |
| 319 | fn incrementModificationCount(hm: *Self) void { | 578 | fn updateEntryIndex( |
| 320 | if (want_modification_safety) { | 579 | self: *Self, |
| 321 | hm.modification_count +%= 1; | 580 | header: *IndexHeader, |
| | 581 | old_entry_index: usize, |
| | 582 | new_entry_index: usize, |
| | 583 | comptime I: type, |
| | 584 | indexes: []Index(I), |
| | 585 | ) void { |
| | 586 | const h = if (store_hash) self.entries.items[new_entry_index].hash else hash(self.entries.items[new_entry_index].key); |
| | 587 | const start_index = header.hashToIndex(h); |
| | 588 | var roll_over: usize = 0; |
| | 589 | while (roll_over <= header.max_distance_from_start_index) : (roll_over += 1) { |
| | 590 | const index_index = (start_index + roll_over) % header.indexes_len; |
| | 591 | const index = &indexes[index_index]; |
| | 592 | if (index.entry_index == old_entry_index) { |
| | 593 | index.entry_index = @intCast(I, new_entry_index); |
| | 594 | return; |
| | 595 | } |
| 322 | } | 596 | } |
| | 597 | unreachable; |
| 323 | } | 598 | } |
| 324 | | 599 | |
| 325 | const InternalPutResult = struct { | 600 | /// Must ensureCapacity before calling this. |
| 326 | new_entry: *Entry, | 601 | fn getOrPutInternal(self: *Self, key: K, header: *IndexHeader, comptime I: type) GetOrPutResult { |
| 327 | old_kv: ?KV, | 602 | const indexes = header.indexes(I); |
| 328 | }; | 603 | const h = hash(key); |
| 329 | | 604 | const start_index = header.hashToIndex(h); |
| 330 | /// Returns a pointer to the new entry. | | |
| 331 | /// Asserts that there is enough space for the new item. | | |
| 332 | fn internalPut(self: *Self, orig_key: K) InternalPutResult { | | |
| 333 | var key = orig_key; | | |
| 334 | var value: V = undefined; | | |
| 335 | const start_index = self.keyToIndex(key); | | |
| 336 | var roll_over: usize = 0; | 605 | var roll_over: usize = 0; |
| 337 | var distance_from_start_index: usize = 0; | 606 | var distance_from_start_index: usize = 0; |
| 338 | var got_result_entry = false; | 607 | while (roll_over <= header.indexes_len) : ({ |
| 339 | var result = InternalPutResult{ | | |
| 340 | .new_entry = undefined, | | |
| 341 | .old_kv = null, | | |
| 342 | }; | | |
| 343 | while (roll_over < self.entries.len) : ({ | | |
| 344 | roll_over += 1; | 608 | roll_over += 1; |
| 345 | distance_from_start_index += 1; | 609 | distance_from_start_index += 1; |
| 346 | }) { | 610 | }) { |
| 347 | const index = self.constrainIndex(start_index + roll_over); | 611 | const index_index = (start_index + roll_over) % header.indexes_len; |
| 348 | const entry = &self.entries[index]; | 612 | const index = indexes[index_index]; |
| 349 | | 613 | if (index.isEmpty()) { |
| 350 | if (entry.used and !eql(entry.kv.key, key)) { | 614 | indexes[index_index] = .{ |
| 351 | if (entry.distance_from_start_index < distance_from_start_index) { | 615 | .distance_from_start_index = @intCast(I, distance_from_start_index), |
| 352 | // robin hood to the rescue | 616 | .entry_index = @intCast(I, self.entries.items.len), |
| 353 | const tmp = entry.*; | 617 | }; |
| 354 | self.max_distance_from_start_index = math.max(self.max_distance_from_start_index, distance_from_start_index); | 618 | header.maybeBumpMax(distance_from_start_index); |
| 355 | if (!got_result_entry) { | 619 | const new_entry = self.entries.addOneAssumeCapacity(); |
| 356 | got_result_entry = true; | 620 | new_entry.* = .{ |
| 357 | result.new_entry = entry; | 621 | .hash = if (store_hash) h else {}, |
| | 622 | .key = key, |
| | 623 | .value = undefined, |
| | 624 | }; |
| | 625 | return .{ |
| | 626 | .found_existing = false, |
| | 627 | .entry = new_entry, |
| | 628 | }; |
| | 629 | } |
| | 630 | |
| | 631 | // This pointer survives the following append because we call |
| | 632 | // entries.ensureCapacity before getOrPutInternal. |
| | 633 | const entry = &self.entries.items[index.entry_index]; |
| | 634 | const hash_match = if (store_hash) h == entry.hash else true; |
| | 635 | if (hash_match and eql(key, entry.key)) { |
| | 636 | return .{ |
| | 637 | .found_existing = true, |
| | 638 | .entry = entry, |
| | 639 | }; |
| | 640 | } |
| | 641 | if (index.distance_from_start_index < distance_from_start_index) { |
| | 642 | // In this case, we did not find the item. We will put a new entry. |
| | 643 | // However, we will use this index for the new entry, and move |
| | 644 | // the previous index down the line, to keep the max_distance_from_start_index |
| | 645 | // as small as possible. |
| | 646 | indexes[index_index] = .{ |
| | 647 | .distance_from_start_index = @intCast(I, distance_from_start_index), |
| | 648 | .entry_index = @intCast(I, self.entries.items.len), |
| | 649 | }; |
| | 650 | header.maybeBumpMax(distance_from_start_index); |
| | 651 | const new_entry = self.entries.addOneAssumeCapacity(); |
| | 652 | new_entry.* = .{ |
| | 653 | .hash = if (store_hash) h else {}, |
| | 654 | .key = key, |
| | 655 | .value = undefined, |
| | 656 | }; |
| | 657 | |
| | 658 | distance_from_start_index = index.distance_from_start_index; |
| | 659 | var prev_entry_index = index.entry_index; |
| | 660 | |
| | 661 | // Find somewhere to put the index we replaced by shifting |
| | 662 | // following indexes backwards. |
| | 663 | roll_over += 1; |
| | 664 | distance_from_start_index += 1; |
| | 665 | while (roll_over < header.indexes_len) : ({ |
| | 666 | roll_over += 1; |
| | 667 | distance_from_start_index += 1; |
| | 668 | }) { |
| | 669 | const next_index_index = (start_index + roll_over) % header.indexes_len; |
| | 670 | const next_index = indexes[next_index_index]; |
| | 671 | if (next_index.isEmpty()) { |
| | 672 | header.maybeBumpMax(distance_from_start_index); |
| | 673 | indexes[next_index_index] = .{ |
| | 674 | .entry_index = prev_entry_index, |
| | 675 | .distance_from_start_index = @intCast(I, distance_from_start_index), |
| | 676 | }; |
| | 677 | return .{ |
| | 678 | .found_existing = false, |
| | 679 | .entry = new_entry, |
| | 680 | }; |
| | 681 | } |
| | 682 | if (next_index.distance_from_start_index < distance_from_start_index) { |
| | 683 | header.maybeBumpMax(distance_from_start_index); |
| | 684 | indexes[next_index_index] = .{ |
| | 685 | .entry_index = prev_entry_index, |
| | 686 | .distance_from_start_index = @intCast(I, distance_from_start_index), |
| | 687 | }; |
| | 688 | distance_from_start_index = next_index.distance_from_start_index; |
| | 689 | prev_entry_index = next_index.entry_index; |
| 358 | } | 690 | } |
| 359 | entry.* = Entry{ | | |
| 360 | .used = true, | | |
| 361 | .distance_from_start_index = distance_from_start_index, | | |
| 362 | .kv = KV{ | | |
| 363 | .key = key, | | |
| 364 | .value = value, | | |
| 365 | }, | | |
| 366 | }; | | |
| 367 | key = tmp.kv.key; | | |
| 368 | value = tmp.kv.value; | | |
| 369 | distance_from_start_index = tmp.distance_from_start_index; | | |
| 370 | } | 691 | } |
| 371 | continue; | 692 | unreachable; |
| 372 | } | 693 | } |
| | 694 | } |
| | 695 | unreachable; |
| | 696 | } |
| 373 | | 697 | |
| 374 | if (entry.used) { | 698 | fn getInternal(self: Self, key: K, header: *IndexHeader, comptime I: type) ?*Entry { |
| 375 | result.old_kv = entry.kv; | 699 | const indexes = header.indexes(I); |
| 376 | } else { | 700 | const h = hash(key); |
| 377 | // adding an entry. otherwise overwriting old value with | 701 | const start_index = header.hashToIndex(h); |
| 378 | // same key | 702 | var roll_over: usize = 0; |
| 379 | self.size += 1; | 703 | while (roll_over <= header.max_distance_from_start_index) : (roll_over += 1) { |
| 380 | } | 704 | const index_index = (start_index + roll_over) % header.indexes_len; |
| | 705 | const index = indexes[index_index]; |
| | 706 | if (index.isEmpty()) |
| | 707 | return null; |
| | 708 | |
| | 709 | const entry = &self.entries.items[index.entry_index]; |
| | 710 | const hash_match = if (store_hash) h == entry.hash else true; |
| | 711 | if (hash_match and eql(key, entry.key)) |
| | 712 | return entry; |
| | 713 | } |
| | 714 | return null; |
| | 715 | } |
| 381 | | 716 | |
| 382 | self.max_distance_from_start_index = math.max(distance_from_start_index, self.max_distance_from_start_index); | 717 | fn insertAllEntriesIntoNewHeader(self: *Self, header: *IndexHeader) void { |
| 383 | if (!got_result_entry) { | 718 | switch (header.capacityIndexType()) { |
| 384 | result.new_entry = entry; | 719 | .u8 => return self.insertAllEntriesIntoNewHeaderGeneric(header, u8), |
| 385 | } | 720 | .u16 => return self.insertAllEntriesIntoNewHeaderGeneric(header, u16), |
| 386 | entry.* = Entry{ | 721 | .u32 => return self.insertAllEntriesIntoNewHeaderGeneric(header, u32), |
| 387 | .used = true, | 722 | .usize => return self.insertAllEntriesIntoNewHeaderGeneric(header, usize), |
| 388 | .distance_from_start_index = distance_from_start_index, | | |
| 389 | .kv = KV{ | | |
| 390 | .key = key, | | |
| 391 | .value = value, | | |
| 392 | }, | | |
| 393 | }; | | |
| 394 | return result; | | |
| 395 | } | 723 | } |
| 396 | unreachable; // put into a full map | | |
| 397 | } | 724 | } |
| 398 | | 725 | |
| 399 | fn internalGet(hm: Self, key: K) ?*KV { | 726 | fn insertAllEntriesIntoNewHeaderGeneric(self: *Self, header: *IndexHeader, comptime I: type) void { |
| 400 | const start_index = hm.keyToIndex(key); | 727 | const indexes = header.indexes(I); |
| 401 | { | 728 | entry_loop: for (self.entries.items) |entry, i| { |
| | 729 | const h = if (store_hash) entry.hash else hash(entry.key); |
| | 730 | const start_index = header.hashToIndex(h); |
| | 731 | var entry_index = i; |
| 402 | var roll_over: usize = 0; | 732 | var roll_over: usize = 0; |
| 403 | while (roll_over <= hm.max_distance_from_start_index) : (roll_over += 1) { | 733 | var distance_from_start_index: usize = 0; |
| 404 | const index = hm.constrainIndex(start_index + roll_over); | 734 | while (roll_over < header.indexes_len) : ({ |
| 405 | const entry = &hm.entries[index]; | 735 | roll_over += 1; |
| 406 | | 736 | distance_from_start_index += 1; |
| 407 | if (!entry.used) return null; | 737 | }) { |
| 408 | if (eql(entry.kv.key, key)) return &entry.kv; | 738 | const index_index = (start_index + roll_over) % header.indexes_len; |
| | 739 | const next_index = indexes[index_index]; |
| | 740 | if (next_index.isEmpty()) { |
| | 741 | header.maybeBumpMax(distance_from_start_index); |
| | 742 | indexes[index_index] = .{ |
| | 743 | .distance_from_start_index = @intCast(I, distance_from_start_index), |
| | 744 | .entry_index = @intCast(I, entry_index), |
| | 745 | }; |
| | 746 | continue :entry_loop; |
| | 747 | } |
| | 748 | if (next_index.distance_from_start_index < distance_from_start_index) { |
| | 749 | header.maybeBumpMax(distance_from_start_index); |
| | 750 | indexes[index_index] = .{ |
| | 751 | .distance_from_start_index = @intCast(I, distance_from_start_index), |
| | 752 | .entry_index = @intCast(I, entry_index), |
| | 753 | }; |
| | 754 | distance_from_start_index = next_index.distance_from_start_index; |
| | 755 | entry_index = next_index.entry_index; |
| | 756 | } |
| 409 | } | 757 | } |
| | 758 | unreachable; |
| 410 | } | 759 | } |
| 411 | return null; | | |
| 412 | } | 760 | } |
| | 761 | }; |
| | 762 | } |
| | 763 | |
| | 764 | const CapacityIndexType = enum { u8, u16, u32, usize }; |
| 413 | | 765 | |
| 414 | fn keyToIndex(hm: Self, key: K) usize { | 766 | fn capacityIndexType(indexes_len: usize) CapacityIndexType { |
| 415 | return hm.constrainIndex(@as(usize, hash(key))); | 767 | if (indexes_len < math.maxInt(u8)) |
| | 768 | return .u8; |
| | 769 | if (indexes_len < math.maxInt(u16)) |
| | 770 | return .u16; |
| | 771 | if (indexes_len < math.maxInt(u32)) |
| | 772 | return .u32; |
| | 773 | return .usize; |
| | 774 | } |
| | 775 | |
| | 776 | fn capacityIndexSize(indexes_len: usize) usize { |
| | 777 | switch (capacityIndexType(indexes_len)) { |
| | 778 | .u8 => return @sizeOf(Index(u8)), |
| | 779 | .u16 => return @sizeOf(Index(u16)), |
| | 780 | .u32 => return @sizeOf(Index(u32)), |
| | 781 | .usize => return @sizeOf(Index(usize)), |
| | 782 | } |
| | 783 | } |
| | 784 | |
| | 785 | fn Index(comptime I: type) type { |
| | 786 | return extern struct { |
| | 787 | entry_index: I, |
| | 788 | distance_from_start_index: I, |
| | 789 | |
| | 790 | const Self = @This(); |
| | 791 | |
| | 792 | fn isEmpty(idx: Self) bool { |
| | 793 | return idx.entry_index == math.maxInt(I); |
| 416 | } | 794 | } |
| 417 | | 795 | |
| 418 | fn constrainIndex(hm: Self, i: usize) usize { | 796 | fn setEmpty(idx: *Self) void { |
| 419 | // this is an optimization for modulo of power of two integers; | 797 | idx.entry_index = math.maxInt(I); |
| 420 | // it requires hm.entries.len to always be a power of two | | |
| 421 | return i & (hm.entries.len - 1); | | |
| 422 | } | 798 | } |
| 423 | }; | 799 | }; |
| 424 | } | 800 | } |
| 425 | | 801 | |
| | 802 | /// This struct is trailed by an array of `Index(I)`, where `I` |
| | 803 | /// and the array length are determined by `indexes_len`. |
| | 804 | const IndexHeader = struct { |
| | 805 | max_distance_from_start_index: usize, |
| | 806 | indexes_len: usize, |
| | 807 | |
| | 808 | fn hashToIndex(header: IndexHeader, h: u32) usize { |
| | 809 | return @as(usize, h) % header.indexes_len; |
| | 810 | } |
| | 811 | |
| | 812 | fn indexes(header: *IndexHeader, comptime I: type) []Index(I) { |
| | 813 | const start = @ptrCast([*]Index(I), @ptrCast([*]u8, header) + @sizeOf(IndexHeader)); |
| | 814 | return start[0..header.indexes_len]; |
| | 815 | } |
| | 816 | |
| | 817 | fn capacityIndexType(header: IndexHeader) CapacityIndexType { |
| | 818 | return hash_map.capacityIndexType(header.indexes_len); |
| | 819 | } |
| | 820 | |
| | 821 | fn maybeBumpMax(header: *IndexHeader, distance_from_start_index: usize) void { |
| | 822 | if (distance_from_start_index > header.max_distance_from_start_index) { |
| | 823 | header.max_distance_from_start_index = distance_from_start_index; |
| | 824 | } |
| | 825 | } |
| | 826 | |
| | 827 | fn alloc(allocator: *Allocator, len: usize) !*IndexHeader { |
| | 828 | const index_size = hash_map.capacityIndexSize(len); |
| | 829 | const nbytes = @sizeOf(IndexHeader) + index_size * len; |
| | 830 | const bytes = try allocator.allocAdvanced(u8, @alignOf(IndexHeader), nbytes, .exact); |
| | 831 | @memset(bytes.ptr + @sizeOf(IndexHeader), 0xff, bytes.len - @sizeOf(IndexHeader)); |
| | 832 | const result = @ptrCast(*IndexHeader, bytes.ptr); |
| | 833 | result.* = .{ |
| | 834 | .max_distance_from_start_index = 0, |
| | 835 | .indexes_len = len, |
| | 836 | }; |
| | 837 | return result; |
| | 838 | } |
| | 839 | |
| | 840 | fn free(header: *IndexHeader, allocator: *Allocator) void { |
| | 841 | const index_size = hash_map.capacityIndexSize(header.indexes_len); |
| | 842 | const ptr = @ptrCast([*]u8, header); |
| | 843 | const slice = ptr[0 .. @sizeOf(IndexHeader) + header.indexes_len * index_size]; |
| | 844 | allocator.free(slice); |
| | 845 | } |
| | 846 | }; |
| | 847 | |
| 426 | test "basic hash map usage" { | 848 | test "basic hash map usage" { |
| 427 | var map = AutoHashMap(i32, i32).init(std.testing.allocator); | 849 | var map = AutoHashMap(i32, i32).init(std.testing.allocator); |
| 428 | defer map.deinit(); | 850 | defer map.deinit(); |
| 429 | | 851 | |
| 430 | testing.expect((try map.put(1, 11)) == null); | 852 | testing.expect((try map.fetchPut(1, 11)) == null); |
| 431 | testing.expect((try map.put(2, 22)) == null); | 853 | testing.expect((try map.fetchPut(2, 22)) == null); |
| 432 | testing.expect((try map.put(3, 33)) == null); | 854 | testing.expect((try map.fetchPut(3, 33)) == null); |
| 433 | testing.expect((try map.put(4, 44)) == null); | 855 | testing.expect((try map.fetchPut(4, 44)) == null); |
| 434 | | 856 | |
| 435 | try map.putNoClobber(5, 55); | 857 | try map.putNoClobber(5, 55); |
| 436 | testing.expect((try map.put(5, 66)).?.value == 55); | 858 | testing.expect((try map.fetchPut(5, 66)).?.value == 55); |
| 437 | testing.expect((try map.put(5, 55)).?.value == 66); | 859 | testing.expect((try map.fetchPut(5, 55)).?.value == 66); |
| 438 | | 860 | |
| 439 | const gop1 = try map.getOrPut(5); | 861 | const gop1 = try map.getOrPut(5); |
| 440 | testing.expect(gop1.found_existing == true); | 862 | testing.expect(gop1.found_existing == true); |
| 441 | testing.expect(gop1.kv.value == 55); | 863 | testing.expect(gop1.entry.value == 55); |
| 442 | gop1.kv.value = 77; | 864 | gop1.entry.value = 77; |
| 443 | testing.expect(map.get(5).?.value == 77); | 865 | testing.expect(map.getEntry(5).?.value == 77); |
| 444 | | 866 | |
| 445 | const gop2 = try map.getOrPut(99); | 867 | const gop2 = try map.getOrPut(99); |
| 446 | testing.expect(gop2.found_existing == false); | 868 | testing.expect(gop2.found_existing == false); |
| 447 | gop2.kv.value = 42; | 869 | gop2.entry.value = 42; |
| 448 | testing.expect(map.get(99).?.value == 42); | 870 | testing.expect(map.getEntry(99).?.value == 42); |
| 449 | | 871 | |
| 450 | const gop3 = try map.getOrPutValue(5, 5); | 872 | const gop3 = try map.getOrPutValue(5, 5); |
| 451 | testing.expect(gop3.value == 77); | 873 | testing.expect(gop3.value == 77); |
| ... | @@ -454,15 +876,15 @@ test "basic hash map usage" { | ... | @@ -454,15 +876,15 @@ test "basic hash map usage" { |
| 454 | testing.expect(gop4.value == 41); | 876 | testing.expect(gop4.value == 41); |
| 455 | | 877 | |
| 456 | testing.expect(map.contains(2)); | 878 | testing.expect(map.contains(2)); |
| 457 | testing.expect(map.get(2).?.value == 22); | 879 | testing.expect(map.getEntry(2).?.value == 22); |
| 458 | testing.expect(map.getValue(2).? == 22); | 880 | testing.expect(map.get(2).? == 22); |
| 459 | | 881 | |
| 460 | const rmv1 = map.remove(2); | 882 | const rmv1 = map.remove(2); |
| 461 | testing.expect(rmv1.?.key == 2); | 883 | testing.expect(rmv1.?.key == 2); |
| 462 | testing.expect(rmv1.?.value == 22); | 884 | testing.expect(rmv1.?.value == 22); |
| 463 | testing.expect(map.remove(2) == null); | 885 | testing.expect(map.remove(2) == null); |
| | 886 | testing.expect(map.getEntry(2) == null); |
| 464 | testing.expect(map.get(2) == null); | 887 | testing.expect(map.get(2) == null); |
| 465 | testing.expect(map.getValue(2) == null); | | |
| 466 | | 888 | |
| 467 | map.removeAssertDiscard(3); | 889 | map.removeAssertDiscard(3); |
| 468 | } | 890 | } |
| ... | @@ -498,8 +920,8 @@ test "iterator hash map" { | ... | @@ -498,8 +920,8 @@ test "iterator hash map" { |
| 498 | it.reset(); | 920 | it.reset(); |
| 499 | | 921 | |
| 500 | var count: usize = 0; | 922 | var count: usize = 0; |
| 501 | while (it.next()) |kv| : (count += 1) { | 923 | while (it.next()) |entry| : (count += 1) { |
| 502 | buffer[@intCast(usize, kv.key)] = kv.value; | 924 | buffer[@intCast(usize, entry.key)] = entry.value; |
| 503 | } | 925 | } |
| 504 | testing.expect(count == 3); | 926 | testing.expect(count == 3); |
| 505 | testing.expect(it.next() == null); | 927 | testing.expect(it.next() == null); |
| ... | @@ -510,8 +932,8 @@ test "iterator hash map" { | ... | @@ -510,8 +932,8 @@ test "iterator hash map" { |
| 510 | | 932 | |
| 511 | it.reset(); | 933 | it.reset(); |
| 512 | count = 0; | 934 | count = 0; |
| 513 | while (it.next()) |kv| { | 935 | while (it.next()) |entry| { |
| 514 | buffer[@intCast(usize, kv.key)] = kv.value; | 936 | buffer[@intCast(usize, entry.key)] = entry.value; |
| 515 | count += 1; | 937 | count += 1; |
| 516 | if (count >= 2) break; | 938 | if (count >= 2) break; |
| 517 | } | 939 | } |
| ... | @@ -531,14 +953,14 @@ test "ensure capacity" { | ... | @@ -531,14 +953,14 @@ test "ensure capacity" { |
| 531 | defer map.deinit(); | 953 | defer map.deinit(); |
| 532 | | 954 | |
| 533 | try map.ensureCapacity(20); | 955 | try map.ensureCapacity(20); |
| 534 | const initialCapacity = map.entries.len; | 956 | const initial_capacity = map.capacity(); |
| 535 | testing.expect(initialCapacity >= 20); | 957 | testing.expect(initial_capacity >= 20); |
| 536 | var i: i32 = 0; | 958 | var i: i32 = 0; |
| 537 | while (i < 20) : (i += 1) { | 959 | while (i < 20) : (i += 1) { |
| 538 | testing.expect(map.putAssumeCapacity(i, i + 10) == null); | 960 | testing.expect(map.fetchPutAssumeCapacity(i, i + 10) == null); |
| 539 | } | 961 | } |
| 540 | // shouldn't resize from putAssumeCapacity | 962 | // shouldn't resize from putAssumeCapacity |
| 541 | testing.expect(initialCapacity == map.entries.len); | 963 | testing.expect(initial_capacity == map.capacity()); |
| 542 | } | 964 | } |
| 543 | | 965 | |
| 544 | pub fn getHashPtrAddrFn(comptime K: type) (fn (K) u32) { | 966 | pub fn getHashPtrAddrFn(comptime K: type) (fn (K) u32) { |
| ... | @@ -575,6 +997,24 @@ pub fn getAutoEqlFn(comptime K: type) (fn (K, K) bool) { | ... | @@ -575,6 +997,24 @@ pub fn getAutoEqlFn(comptime K: type) (fn (K, K) bool) { |
| 575 | }.eql; | 997 | }.eql; |
| 576 | } | 998 | } |
| 577 | | 999 | |
| | 1000 | pub fn autoEqlIsCheap(comptime K: type) bool { |
| | 1001 | return switch (@typeInfo(K)) { |
| | 1002 | .Bool, |
| | 1003 | .Int, |
| | 1004 | .Float, |
| | 1005 | .Pointer, |
| | 1006 | .ComptimeFloat, |
| | 1007 | .ComptimeInt, |
| | 1008 | .Enum, |
| | 1009 | .Fn, |
| | 1010 | .ErrorSet, |
| | 1011 | .AnyFrame, |
| | 1012 | .EnumLiteral, |
| | 1013 | => true, |
| | 1014 | else => false, |
| | 1015 | }; |
| | 1016 | } |
| | 1017 | |
| 578 | pub fn getAutoHashStratFn(comptime K: type, comptime strategy: std.hash.Strategy) (fn (K) u32) { | 1018 | pub fn getAutoHashStratFn(comptime K: type, comptime strategy: std.hash.Strategy) (fn (K) u32) { |
| 579 | return struct { | 1019 | return struct { |
| 580 | fn hash(key: K) u32 { | 1020 | fn hash(key: K) u32 { |