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 <limits.h> // CHAR_BIT
|
7 |
|
8 | #include "mycpp/gc_alloc.h" // gHeap
|
9 | #include "mycpp/gc_dict.h" // for dict_erase()
|
10 | #include "mycpp/gc_tuple.h"
|
11 |
|
12 | template <class K, class V>
|
13 | class Dict;
|
14 |
|
15 | // https://stackoverflow.com/questions/3919995/determining-sprintf-buffer-size-whats-the-standard/11092994#11092994
|
16 | // Notes:
|
17 | // - Python 2.7's intobject.c has an erroneous +6
|
18 | // - This is 13, but len('-2147483648') is 11, which means we only need 12?
|
19 | // - This formula is valid for octal(), because 2^(3 bits) = 8
|
20 |
|
21 | const int kIntBufSize = CHAR_BIT * sizeof(int) / 3 + 3;
|
22 |
|
23 | namespace mylib {
|
24 |
|
25 | void InitCppOnly();
|
26 |
|
27 | // Wrappers around our C++ APIs
|
28 |
|
29 | inline void MaybeCollect() {
|
30 | gHeap.MaybeCollect();
|
31 | }
|
32 |
|
33 | void print_stderr(BigStr* s);
|
34 |
|
35 | inline int ByteAt(BigStr* s, int i) {
|
36 | DCHECK(0 <= i);
|
37 | DCHECK(i <= len(s));
|
38 |
|
39 | return static_cast<unsigned char>(s->data_[i]);
|
40 | }
|
41 |
|
42 | inline int ByteEquals(int byte, BigStr* ch) {
|
43 | DCHECK(0 <= byte);
|
44 | DCHECK(byte < 256);
|
45 |
|
46 | DCHECK(len(ch) == 1);
|
47 |
|
48 | return byte == static_cast<unsigned char>(ch->data_[0]);
|
49 | }
|
50 |
|
51 | inline int ByteInSet(int byte, BigStr* byte_set) {
|
52 | DCHECK(0 <= byte);
|
53 | DCHECK(byte < 256);
|
54 |
|
55 | int n = len(byte_set);
|
56 | for (int i = 0; i < n; ++i) {
|
57 | int b = static_cast<unsigned char>(byte_set->data_[i]);
|
58 | if (byte == b) {
|
59 | return true;
|
60 | }
|
61 | }
|
62 | return false;
|
63 | }
|
64 |
|
65 | BigStr* JoinBytes(List<int>* byte_list);
|
66 |
|
67 | // const int kStdout = 1;
|
68 | // const int kStderr = 2;
|
69 |
|
70 | // void writeln(BigStr* s, int fd = kStdout);
|
71 |
|
72 | Tuple2<BigStr*, BigStr*> split_once(BigStr* s, BigStr* delim);
|
73 |
|
74 | template <typename K, typename V>
|
75 | void dict_erase(Dict<K, V>* haystack, K needle) {
|
76 | DCHECK(haystack->obj_header().heap_tag != HeapTag::Global);
|
77 |
|
78 | int pos = haystack->hash_and_probe(needle);
|
79 | if (pos == kTooSmall) {
|
80 | return;
|
81 | }
|
82 | DCHECK(pos >= 0);
|
83 | int kv_index = haystack->index_->items_[pos];
|
84 | if (kv_index < 0) {
|
85 | return;
|
86 | }
|
87 |
|
88 | int last_kv_index = haystack->len_ - 1;
|
89 | DCHECK(kv_index <= last_kv_index);
|
90 |
|
91 | // Swap the target entry with the most recently inserted one before removing
|
92 | // it. This has two benefits.
|
93 | // (1) It keeps the entry arrays compact. All valid entries occupy a
|
94 | // contiguous region in memory.
|
95 | // (2) It prevents holes in the entry arrays. This makes iterating over
|
96 | // entries (e.g. in keys() or DictIter()) trivial and doesn't require
|
97 | // any extra validity state (like a bitset of unusable slots). This is
|
98 | // important because keys and values wont't always be pointers, so we
|
99 | // can't rely on NULL checks for validity. We also can't wrap the slab
|
100 | // entry types in some other type without modifying the garbage
|
101 | // collector to trace through unmanaged types (or paying the extra
|
102 | // allocations for the outer type).
|
103 | if (kv_index != last_kv_index) {
|
104 | K last_key = haystack->keys_->items_[last_kv_index];
|
105 | V last_val = haystack->values_->items_[last_kv_index];
|
106 | int last_pos = haystack->hash_and_probe(last_key);
|
107 | DCHECK(last_pos != kNotFound);
|
108 | haystack->keys_->items_[kv_index] = last_key;
|
109 | haystack->values_->items_[kv_index] = last_val;
|
110 | haystack->index_->items_[last_pos] = kv_index;
|
111 | }
|
112 |
|
113 | // Zero out for GC. These could be nullptr or 0
|
114 | haystack->keys_->items_[last_kv_index] = 0;
|
115 | haystack->values_->items_[last_kv_index] = 0;
|
116 | haystack->index_->items_[pos] = kDeletedEntry;
|
117 | haystack->len_--;
|
118 | DCHECK(haystack->len_ < haystack->capacity_);
|
119 | }
|
120 |
|
121 | // NOTE: Can use OverAllocatedStr for all of these, rather than copying
|
122 |
|
123 | inline BigStr* hex_lower(int i) {
|
124 | char buf[kIntBufSize];
|
125 | int len = snprintf(buf, kIntBufSize, "%x", i);
|
126 | return ::StrFromC(buf, len);
|
127 | }
|
128 |
|
129 | inline BigStr* hex_upper(int i) {
|
130 | char buf[kIntBufSize];
|
131 | int len = snprintf(buf, kIntBufSize, "%X", i);
|
132 | return ::StrFromC(buf, len);
|
133 | }
|
134 |
|
135 | inline BigStr* octal(int i) {
|
136 | char buf[kIntBufSize];
|
137 | int len = snprintf(buf, kIntBufSize, "%o", i);
|
138 | return ::StrFromC(buf, len);
|
139 | }
|
140 |
|
141 | // Abstract type: Union of LineReader and Writer
|
142 | class File {
|
143 | public:
|
144 | File() {
|
145 | }
|
146 | // Writer
|
147 | virtual void write(BigStr* s) = 0;
|
148 | virtual void flush() = 0;
|
149 |
|
150 | // Reader
|
151 | virtual BigStr* readline() = 0;
|
152 |
|
153 | // Both
|
154 | virtual bool isatty() = 0;
|
155 | virtual void close() = 0;
|
156 |
|
157 | static constexpr ObjHeader obj_header() {
|
158 | return ObjHeader::ClassFixed(field_mask(), sizeof(File));
|
159 | }
|
160 |
|
161 | static constexpr uint32_t field_mask() {
|
162 | return kZeroMask;
|
163 | }
|
164 | };
|
165 |
|
166 | // Wrap a FILE* for read and write
|
167 | class CFile : public File {
|
168 | public:
|
169 | explicit CFile(FILE* f) : File(), f_(f) {
|
170 | }
|
171 | // Writer
|
172 | void write(BigStr* s) override;
|
173 | void flush() override;
|
174 |
|
175 | // Reader
|
176 | BigStr* readline() override;
|
177 |
|
178 | // Both
|
179 | bool isatty() override;
|
180 | void close() override;
|
181 |
|
182 | static constexpr ObjHeader obj_header() {
|
183 | return ObjHeader::ClassFixed(field_mask(), sizeof(CFile));
|
184 | }
|
185 |
|
186 | static constexpr uint32_t field_mask() {
|
187 | // not mutating field_mask because FILE* isn't a GC object
|
188 | return File::field_mask();
|
189 | }
|
190 |
|
191 | private:
|
192 | FILE* f_;
|
193 |
|
194 | DISALLOW_COPY_AND_ASSIGN(CFile)
|
195 | };
|
196 |
|
197 | // Abstract File we can only read from.
|
198 | // TODO: can we get rid of DCHECK() and reinterpret_cast?
|
199 | class LineReader : public File {
|
200 | public:
|
201 | LineReader() : File() {
|
202 | }
|
203 | void write(BigStr* s) override {
|
204 | CHECK(false); // should not happen
|
205 | }
|
206 | void flush() override {
|
207 | CHECK(false); // should not happen
|
208 | }
|
209 |
|
210 | static constexpr ObjHeader obj_header() {
|
211 | return ObjHeader::ClassFixed(field_mask(), sizeof(LineReader));
|
212 | }
|
213 |
|
214 | static constexpr uint32_t field_mask() {
|
215 | return kZeroMask;
|
216 | }
|
217 | };
|
218 |
|
219 | class BufLineReader : public LineReader {
|
220 | public:
|
221 | explicit BufLineReader(BigStr* s) : LineReader(), s_(s), pos_(0) {
|
222 | }
|
223 | virtual BigStr* readline();
|
224 | virtual bool isatty() {
|
225 | return false;
|
226 | }
|
227 | virtual void close() {
|
228 | }
|
229 |
|
230 | BigStr* s_;
|
231 | int pos_;
|
232 |
|
233 | static constexpr ObjHeader obj_header() {
|
234 | return ObjHeader::ClassFixed(field_mask(), sizeof(LineReader));
|
235 | }
|
236 |
|
237 | static constexpr uint32_t field_mask() {
|
238 | return LineReader::field_mask() | maskbit(offsetof(BufLineReader, s_));
|
239 | }
|
240 |
|
241 | DISALLOW_COPY_AND_ASSIGN(BufLineReader)
|
242 | };
|
243 |
|
244 | extern LineReader* gStdin;
|
245 |
|
246 | inline LineReader* Stdin() {
|
247 | if (gStdin == nullptr) {
|
248 | gStdin = reinterpret_cast<LineReader*>(Alloc<CFile>(stdin));
|
249 | }
|
250 | return gStdin;
|
251 | }
|
252 |
|
253 | LineReader* open(BigStr* path);
|
254 |
|
255 | // Abstract File we can only write to.
|
256 | // TODO: can we get rid of DCHECK() and reinterpret_cast?
|
257 | class Writer : public File {
|
258 | public:
|
259 | Writer() : File() {
|
260 | }
|
261 | BigStr* readline() override {
|
262 | CHECK(false); // should not happen
|
263 | }
|
264 |
|
265 | static constexpr ObjHeader obj_header() {
|
266 | return ObjHeader::ClassFixed(field_mask(), sizeof(Writer));
|
267 | }
|
268 |
|
269 | static constexpr uint32_t field_mask() {
|
270 | return kZeroMask;
|
271 | }
|
272 | };
|
273 |
|
274 | class MutableStr;
|
275 |
|
276 | class BufWriter : public Writer {
|
277 | public:
|
278 | BufWriter() : Writer(), str_(nullptr), len_(0) {
|
279 | }
|
280 | void write(BigStr* s) override;
|
281 | void write_spaces(int n);
|
282 | void clear() { // Reuse this instance
|
283 | str_ = nullptr;
|
284 | len_ = 0;
|
285 | is_valid_ = true;
|
286 | }
|
287 | void close() override {
|
288 | }
|
289 | void flush() override {
|
290 | }
|
291 | bool isatty() override {
|
292 | return false;
|
293 | }
|
294 | BigStr* getvalue(); // part of cStringIO API
|
295 |
|
296 | //
|
297 | // Low Level API for C++ usage only
|
298 | //
|
299 |
|
300 | // Convenient API that avoids BigStr*
|
301 | void WriteConst(const char* c_string);
|
302 |
|
303 | // Potentially resizes the buffer.
|
304 | void EnsureMoreSpace(int n);
|
305 | // After EnsureMoreSpace(42), you can write 42 more bytes safely.
|
306 | //
|
307 | // Note that if you call EnsureMoreSpace(42), write 5 byte, and then
|
308 | // EnsureMoreSpace(42) again, the amount of additional space reserved is 47.
|
309 |
|
310 | // (Similar to vector::reserve(n), but it takes an integer to ADD to the
|
311 | // capacity.)
|
312 |
|
313 | uint8_t* LengthPointer(); // start + length
|
314 | uint8_t* CapacityPointer(); // start + capacity
|
315 | void SetLengthFrom(uint8_t* length_ptr);
|
316 |
|
317 | int Length() {
|
318 | return len_;
|
319 | }
|
320 |
|
321 | // Rewind to earlier position, future writes start there
|
322 | void Truncate(int length);
|
323 |
|
324 | static constexpr ObjHeader obj_header() {
|
325 | return ObjHeader::ClassFixed(field_mask(), sizeof(BufWriter));
|
326 | }
|
327 |
|
328 | static constexpr unsigned field_mask() {
|
329 | // maskvit_v() because BufWriter has virtual methods
|
330 | return Writer::field_mask() | maskbit(offsetof(BufWriter, str_));
|
331 | }
|
332 |
|
333 | private:
|
334 | void WriteRaw(char* s, int n);
|
335 |
|
336 | MutableStr* str_; // getvalue() turns this directly into Str*, no copying
|
337 | int len_; // how many bytes have been written so far
|
338 | bool is_valid_ = true; // It becomes invalid after getvalue() is called
|
339 | };
|
340 |
|
341 | extern Writer* gStdout;
|
342 |
|
343 | inline Writer* Stdout() {
|
344 | if (gStdout == nullptr) {
|
345 | gStdout = reinterpret_cast<Writer*>(Alloc<CFile>(stdout));
|
346 | gHeap.RootGlobalVar(gStdout);
|
347 | }
|
348 | return gStdout;
|
349 | }
|
350 |
|
351 | extern Writer* gStderr;
|
352 |
|
353 | inline Writer* Stderr() {
|
354 | if (gStderr == nullptr) {
|
355 | gStderr = reinterpret_cast<Writer*>(Alloc<CFile>(stderr));
|
356 | gHeap.RootGlobalVar(gStderr);
|
357 | }
|
358 | return gStderr;
|
359 | }
|
360 |
|
361 | class UniqueObjects {
|
362 | // Can't be expressed in typed Python because we don't have uint64_t for
|
363 | // addresses
|
364 |
|
365 | public:
|
366 | UniqueObjects() {
|
367 | }
|
368 | void Add(void* obj) {
|
369 | }
|
370 | int Get(void* obj) {
|
371 | return -1;
|
372 | }
|
373 |
|
374 | static constexpr ObjHeader obj_header() {
|
375 | return ObjHeader::ClassFixed(field_mask(), sizeof(UniqueObjects));
|
376 | }
|
377 |
|
378 | // SPECIAL CASE? We should never have a unique reference to an object? So
|
379 | // don't bother tracing
|
380 | static constexpr uint32_t field_mask() {
|
381 | return kZeroMask;
|
382 | }
|
383 |
|
384 | private:
|
385 | // address -> small integer ID
|
386 | Dict<void*, int> addresses_;
|
387 | };
|
388 |
|
389 | } // namespace mylib
|
390 |
|
391 | #endif // MYCPP_GC_MYLIB_H
|