| 1 | // gc_mylib.h - corresponds to mycpp/mylib.py
|
| 2 |
|
| 3 | #ifndef MYCPP_GC_MYLIB_H
|
| 4 | #define MYCPP_GC_MYLIB_H
|
| 5 |
|
| 6 | #include "mycpp/gc_alloc.h" // gHeap
|
| 7 | #include "mycpp/gc_dict.h" // for dict_erase()
|
| 8 | #include "mycpp/gc_mops.h"
|
| 9 | #include "mycpp/gc_tuple.h"
|
| 10 |
|
| 11 | template <class K, class V>
|
| 12 | class Dict;
|
| 13 |
|
| 14 | namespace mylib {
|
| 15 |
|
| 16 | void InitCppOnly();
|
| 17 |
|
| 18 | // Wrappers around our C++ APIs
|
| 19 |
|
| 20 | inline void MaybeCollect() {
|
| 21 | gHeap.MaybeCollect();
|
| 22 | }
|
| 23 |
|
| 24 | inline void PrintGcStats() {
|
| 25 | gHeap.PrintShortStats(); // print to stderr
|
| 26 | }
|
| 27 |
|
| 28 | void print_stderr(BigStr* s);
|
| 29 |
|
| 30 | inline int ByteAt(BigStr* s, int i) {
|
| 31 | DCHECK(0 <= i);
|
| 32 | DCHECK(i <= len(s));
|
| 33 |
|
| 34 | return static_cast<unsigned char>(s->data_[i]);
|
| 35 | }
|
| 36 |
|
| 37 | inline int ByteEquals(int byte, BigStr* ch) {
|
| 38 | DCHECK(0 <= byte);
|
| 39 | DCHECK(byte < 256);
|
| 40 |
|
| 41 | DCHECK(len(ch) == 1);
|
| 42 |
|
| 43 | return byte == static_cast<unsigned char>(ch->data_[0]);
|
| 44 | }
|
| 45 |
|
| 46 | inline int ByteInSet(int byte, BigStr* byte_set) {
|
| 47 | DCHECK(0 <= byte);
|
| 48 | DCHECK(byte < 256);
|
| 49 |
|
| 50 | int n = len(byte_set);
|
| 51 | for (int i = 0; i < n; ++i) {
|
| 52 | int b = static_cast<unsigned char>(byte_set->data_[i]);
|
| 53 | if (byte == b) {
|
| 54 | return true;
|
| 55 | }
|
| 56 | }
|
| 57 | return false;
|
| 58 | }
|
| 59 |
|
| 60 | BigStr* JoinBytes(List<int>* byte_list);
|
| 61 |
|
| 62 | void BigIntSort(List<mops::BigInt>* keys);
|
| 63 |
|
| 64 | // const int kStdout = 1;
|
| 65 | // const int kStderr = 2;
|
| 66 |
|
| 67 | // void writeln(BigStr* s, int fd = kStdout);
|
| 68 |
|
| 69 | Tuple2<BigStr*, BigStr*> split_once(BigStr* s, BigStr* delim);
|
| 70 |
|
| 71 | template <typename K, typename V>
|
| 72 | void dict_erase(Dict<K, V>* haystack, K needle) {
|
| 73 | DCHECK(haystack->obj_header().heap_tag != HeapTag::Global);
|
| 74 |
|
| 75 | int pos = haystack->hash_and_probe(needle);
|
| 76 | if (pos == kTooSmall) {
|
| 77 | return;
|
| 78 | }
|
| 79 | DCHECK(pos >= 0);
|
| 80 | int kv_index = haystack->index_->items_[pos];
|
| 81 | if (kv_index < 0) {
|
| 82 | return;
|
| 83 | }
|
| 84 |
|
| 85 | int last_kv_index = haystack->len_ - 1;
|
| 86 | DCHECK(kv_index <= last_kv_index);
|
| 87 |
|
| 88 | // Swap the target entry with the most recently inserted one before removing
|
| 89 | // it. This has two benefits.
|
| 90 | // (1) It keeps the entry arrays compact. All valid entries occupy a
|
| 91 | // contiguous region in memory.
|
| 92 | // (2) It prevents holes in the entry arrays. This makes iterating over
|
| 93 | // entries (e.g. in keys() or DictIter()) trivial and doesn't require
|
| 94 | // any extra validity state (like a bitset of unusable slots). This is
|
| 95 | // important because keys and values wont't always be pointers, so we
|
| 96 | // can't rely on NULL checks for validity. We also can't wrap the slab
|
| 97 | // entry types in some other type without modifying the garbage
|
| 98 | // collector to trace through unmanaged types (or paying the extra
|
| 99 | // allocations for the outer type).
|
| 100 | if (kv_index != last_kv_index) {
|
| 101 | K last_key = haystack->keys_->items_[last_kv_index];
|
| 102 | V last_val = haystack->values_->items_[last_kv_index];
|
| 103 | int last_pos = haystack->hash_and_probe(last_key);
|
| 104 | DCHECK(last_pos != kNotFound);
|
| 105 | haystack->keys_->items_[kv_index] = last_key;
|
| 106 | haystack->values_->items_[kv_index] = last_val;
|
| 107 | haystack->index_->items_[last_pos] = kv_index;
|
| 108 | }
|
| 109 |
|
| 110 | // Zero out for GC. These could be nullptr or 0
|
| 111 | haystack->keys_->items_[last_kv_index] = 0;
|
| 112 | haystack->values_->items_[last_kv_index] = 0;
|
| 113 | haystack->index_->items_[pos] = kDeletedEntry;
|
| 114 | haystack->len_--;
|
| 115 | DCHECK(haystack->len_ < haystack->capacity_);
|
| 116 | }
|
| 117 |
|
| 118 | inline BigStr* hex_lower(int i) {
|
| 119 | // Note: Could also use OverAllocatedStr, but most strings are small?
|
| 120 | char buf[kIntBufSize];
|
| 121 | int len = snprintf(buf, kIntBufSize, "%x", i);
|
| 122 | return ::StrFromC(buf, len);
|
| 123 | }
|
| 124 |
|
| 125 | // Abstract type: Union of LineReader and Writer
|
| 126 | class File {
|
| 127 | public:
|
| 128 | File() {
|
| 129 | }
|
| 130 | // Writer
|
| 131 | virtual void write(BigStr* s) = 0;
|
| 132 | virtual void flush() = 0;
|
| 133 |
|
| 134 | // Reader
|
| 135 | virtual BigStr* readline() = 0;
|
| 136 |
|
| 137 | // Both
|
| 138 | virtual bool isatty() = 0;
|
| 139 | virtual void close() = 0;
|
| 140 |
|
| 141 | static constexpr ObjHeader obj_header() {
|
| 142 | return ObjHeader::ClassFixed(field_mask(), sizeof(File));
|
| 143 | }
|
| 144 |
|
| 145 | static constexpr uint32_t field_mask() {
|
| 146 | return kZeroMask;
|
| 147 | }
|
| 148 | };
|
| 149 |
|
| 150 | // Wrap a FILE* for read and write
|
| 151 | class CFile : public File {
|
| 152 | public:
|
| 153 | explicit CFile(FILE* f) : File(), f_(f) {
|
| 154 | }
|
| 155 | // Writer
|
| 156 | void write(BigStr* s) override;
|
| 157 | void flush() override;
|
| 158 |
|
| 159 | // Reader
|
| 160 | BigStr* readline() override;
|
| 161 |
|
| 162 | // Both
|
| 163 | bool isatty() override;
|
| 164 | void close() override;
|
| 165 |
|
| 166 | static constexpr ObjHeader obj_header() {
|
| 167 | return ObjHeader::ClassFixed(field_mask(), sizeof(CFile));
|
| 168 | }
|
| 169 |
|
| 170 | static constexpr uint32_t field_mask() {
|
| 171 | // not mutating field_mask because FILE* isn't a GC object
|
| 172 | return File::field_mask();
|
| 173 | }
|
| 174 |
|
| 175 | private:
|
| 176 | FILE* f_;
|
| 177 |
|
| 178 | DISALLOW_COPY_AND_ASSIGN(CFile)
|
| 179 | };
|
| 180 |
|
| 181 | // Abstract File we can only read from.
|
| 182 | // TODO: can we get rid of DCHECK() and reinterpret_cast?
|
| 183 | class LineReader : public File {
|
| 184 | public:
|
| 185 | LineReader() : File() {
|
| 186 | }
|
| 187 | void write(BigStr* s) override {
|
| 188 | CHECK(false); // should not happen
|
| 189 | }
|
| 190 | void flush() override {
|
| 191 | CHECK(false); // should not happen
|
| 192 | }
|
| 193 |
|
| 194 | static constexpr ObjHeader obj_header() {
|
| 195 | return ObjHeader::ClassFixed(field_mask(), sizeof(LineReader));
|
| 196 | }
|
| 197 |
|
| 198 | static constexpr uint32_t field_mask() {
|
| 199 | return kZeroMask;
|
| 200 | }
|
| 201 | };
|
| 202 |
|
| 203 | class BufLineReader : public LineReader {
|
| 204 | public:
|
| 205 | explicit BufLineReader(BigStr* s) : LineReader(), s_(s), pos_(0) {
|
| 206 | }
|
| 207 | virtual BigStr* readline();
|
| 208 | virtual bool isatty() {
|
| 209 | return false;
|
| 210 | }
|
| 211 | virtual void close() {
|
| 212 | }
|
| 213 |
|
| 214 | BigStr* s_;
|
| 215 | int pos_;
|
| 216 |
|
| 217 | static constexpr ObjHeader obj_header() {
|
| 218 | return ObjHeader::ClassFixed(field_mask(), sizeof(LineReader));
|
| 219 | }
|
| 220 |
|
| 221 | static constexpr uint32_t field_mask() {
|
| 222 | return LineReader::field_mask() | maskbit(offsetof(BufLineReader, s_));
|
| 223 | }
|
| 224 |
|
| 225 | DISALLOW_COPY_AND_ASSIGN(BufLineReader)
|
| 226 | };
|
| 227 |
|
| 228 | extern LineReader* gStdin;
|
| 229 |
|
| 230 | inline LineReader* Stdin() {
|
| 231 | if (gStdin == nullptr) {
|
| 232 | gStdin = reinterpret_cast<LineReader*>(Alloc<CFile>(stdin));
|
| 233 | }
|
| 234 | return gStdin;
|
| 235 | }
|
| 236 |
|
| 237 | LineReader* open(BigStr* path);
|
| 238 |
|
| 239 | // Abstract File we can only write to.
|
| 240 | // TODO: can we get rid of DCHECK() and reinterpret_cast?
|
| 241 | class Writer : public File {
|
| 242 | public:
|
| 243 | Writer() : File() {
|
| 244 | }
|
| 245 | BigStr* readline() override {
|
| 246 | CHECK(false); // should not happen
|
| 247 | }
|
| 248 |
|
| 249 | static constexpr ObjHeader obj_header() {
|
| 250 | return ObjHeader::ClassFixed(field_mask(), sizeof(Writer));
|
| 251 | }
|
| 252 |
|
| 253 | static constexpr uint32_t field_mask() {
|
| 254 | return kZeroMask;
|
| 255 | }
|
| 256 | };
|
| 257 |
|
| 258 | class MutableStr;
|
| 259 |
|
| 260 | class BufWriter : public Writer {
|
| 261 | public:
|
| 262 | BufWriter() : Writer(), str_(nullptr), len_(0) {
|
| 263 | }
|
| 264 | void write(BigStr* s) override;
|
| 265 | void write_spaces(int n);
|
| 266 | void clear() { // Reuse this instance
|
| 267 | str_ = nullptr;
|
| 268 | len_ = 0;
|
| 269 | is_valid_ = true;
|
| 270 | }
|
| 271 | void close() override {
|
| 272 | }
|
| 273 | void flush() override {
|
| 274 | }
|
| 275 | bool isatty() override {
|
| 276 | return false;
|
| 277 | }
|
| 278 | BigStr* getvalue(); // part of cStringIO API
|
| 279 |
|
| 280 | //
|
| 281 | // Low Level API for C++ usage only
|
| 282 | //
|
| 283 |
|
| 284 | // Convenient API that avoids BigStr*
|
| 285 | void WriteConst(const char* c_string);
|
| 286 |
|
| 287 | // Potentially resizes the buffer.
|
| 288 | void EnsureMoreSpace(int n);
|
| 289 | // After EnsureMoreSpace(42), you can write 42 more bytes safely.
|
| 290 | //
|
| 291 | // Note that if you call EnsureMoreSpace(42), write 5 byte, and then
|
| 292 | // EnsureMoreSpace(42) again, the amount of additional space reserved is 47.
|
| 293 |
|
| 294 | // (Similar to vector::reserve(n), but it takes an integer to ADD to the
|
| 295 | // capacity.)
|
| 296 |
|
| 297 | uint8_t* LengthPointer(); // start + length
|
| 298 | uint8_t* CapacityPointer(); // start + capacity
|
| 299 | void SetLengthFrom(uint8_t* length_ptr);
|
| 300 |
|
| 301 | int Length() {
|
| 302 | return len_;
|
| 303 | }
|
| 304 |
|
| 305 | // Rewind to earlier position, future writes start there
|
| 306 | void Truncate(int length);
|
| 307 |
|
| 308 | static constexpr ObjHeader obj_header() {
|
| 309 | return ObjHeader::ClassFixed(field_mask(), sizeof(BufWriter));
|
| 310 | }
|
| 311 |
|
| 312 | static constexpr unsigned field_mask() {
|
| 313 | // maskvit_v() because BufWriter has virtual methods
|
| 314 | return Writer::field_mask() | maskbit(offsetof(BufWriter, str_));
|
| 315 | }
|
| 316 |
|
| 317 | private:
|
| 318 | void WriteRaw(char* s, int n);
|
| 319 |
|
| 320 | MutableStr* str_; // getvalue() turns this directly into Str*, no copying
|
| 321 | int len_; // how many bytes have been written so far
|
| 322 | bool is_valid_ = true; // It becomes invalid after getvalue() is called
|
| 323 | };
|
| 324 |
|
| 325 | extern Writer* gStdout;
|
| 326 |
|
| 327 | inline Writer* Stdout() {
|
| 328 | if (gStdout == nullptr) {
|
| 329 | gStdout = reinterpret_cast<Writer*>(Alloc<CFile>(stdout));
|
| 330 | gHeap.RootGlobalVar(gStdout);
|
| 331 | }
|
| 332 | return gStdout;
|
| 333 | }
|
| 334 |
|
| 335 | extern Writer* gStderr;
|
| 336 |
|
| 337 | inline Writer* Stderr() {
|
| 338 | if (gStderr == nullptr) {
|
| 339 | gStderr = reinterpret_cast<Writer*>(Alloc<CFile>(stderr));
|
| 340 | gHeap.RootGlobalVar(gStderr);
|
| 341 | }
|
| 342 | return gStderr;
|
| 343 | }
|
| 344 |
|
| 345 | class UniqueObjects {
|
| 346 | // Can't be expressed in typed Python because we don't have uint64_t for
|
| 347 | // addresses
|
| 348 |
|
| 349 | public:
|
| 350 | UniqueObjects() {
|
| 351 | }
|
| 352 | void Add(void* obj) {
|
| 353 | }
|
| 354 | int Get(void* obj) {
|
| 355 | return -1;
|
| 356 | }
|
| 357 |
|
| 358 | static constexpr ObjHeader obj_header() {
|
| 359 | return ObjHeader::ClassFixed(field_mask(), sizeof(UniqueObjects));
|
| 360 | }
|
| 361 |
|
| 362 | // SPECIAL CASE? We should never have a unique reference to an object? So
|
| 363 | // don't bother tracing
|
| 364 | static constexpr uint32_t field_mask() {
|
| 365 | return kZeroMask;
|
| 366 | }
|
| 367 |
|
| 368 | private:
|
| 369 | // address -> small integer ID
|
| 370 | Dict<void*, int> addresses_;
|
| 371 | };
|
| 372 |
|
| 373 | } // namespace mylib
|
| 374 |
|
| 375 | #endif // MYCPP_GC_MYLIB_H
|