| 1 | // Target Language Constructs
|
| 2 | //
|
| 3 | // We're generating a subset of C++.
|
| 4 | //
|
| 5 | // - Done:
|
| 6 | // - initializer lists
|
| 7 | // - exceptions
|
| 8 | // - default arguments
|
| 9 | // - namespaces
|
| 10 | //
|
| 11 | // - advanced:
|
| 12 | // - What do Python closures get translated to? Oil uses them in a few
|
| 13 | // places, e.g. for the readline callbacks.
|
| 14 | // - C++ 20 coroutines (but we're almost certainly not using this)
|
| 15 |
|
| 16 | #include <sys/mman.h> // mmap()
|
| 17 |
|
| 18 | #include <initializer_list>
|
| 19 | #include <memory> // shared_ptr
|
| 20 | #include <stdexcept>
|
| 21 | #include <unordered_map>
|
| 22 | #include <vector>
|
| 23 |
|
| 24 | #include "mycpp/runtime.h"
|
| 25 | #include "vendor/greatest.h"
|
| 26 |
|
| 27 | using std::unordered_map;
|
| 28 |
|
| 29 | class RootingScope2 {
|
| 30 | public:
|
| 31 | RootingScope2() {
|
| 32 | }
|
| 33 | RootingScope2(const char* func_name) {
|
| 34 | log(">>> %s", func_name);
|
| 35 | }
|
| 36 | ~RootingScope2() {
|
| 37 | }
|
| 38 | };
|
| 39 |
|
| 40 | #define ROOTING_REPORT 1
|
| 41 |
|
| 42 | #if ROOTING_REPORT
|
| 43 | #define FUNC_NAME_2() __PRETTY_FUNCTION__
|
| 44 | #else
|
| 45 | #define FUNC_NAME_2()
|
| 46 | #endif
|
| 47 |
|
| 48 | class MyList {
|
| 49 | public:
|
| 50 | MyList(std::initializer_list<int> init) : v_() {
|
| 51 | for (int i : init) {
|
| 52 | v_.push_back(i);
|
| 53 | }
|
| 54 | }
|
| 55 | std::vector<int> v_;
|
| 56 | };
|
| 57 |
|
| 58 | template <class T>
|
| 59 | class Array {
|
| 60 | public:
|
| 61 | Array() : v_() {
|
| 62 | }
|
| 63 |
|
| 64 | Array(std::initializer_list<T> init) : v_() {
|
| 65 | for (T i : init) {
|
| 66 | v_.push_back(i);
|
| 67 | }
|
| 68 | }
|
| 69 |
|
| 70 | void append(T item) {
|
| 71 | RootingScope2 _r(FUNC_NAME_2());
|
| 72 |
|
| 73 | v_.push_back(item);
|
| 74 | }
|
| 75 |
|
| 76 | int size() {
|
| 77 | return v_.size();
|
| 78 | }
|
| 79 |
|
| 80 | std::vector<T> v_;
|
| 81 | };
|
| 82 |
|
| 83 | class FatalError {};
|
| 84 |
|
| 85 | class ParseError : public FatalError {
|
| 86 | public:
|
| 87 | ParseError(const char* reason) : reason_(reason) {
|
| 88 | }
|
| 89 | const char* reason() const {
|
| 90 | RootingScope2 _r(FUNC_NAME_2());
|
| 91 |
|
| 92 | return reason_;
|
| 93 | }
|
| 94 |
|
| 95 | private:
|
| 96 | const char* reason_;
|
| 97 | };
|
| 98 |
|
| 99 | // https://stackoverflow.com/questions/8480640/how-to-throw-a-c-exception
|
| 100 | int compare(int a, int b) {
|
| 101 | if (a < 0 || b < 0) {
|
| 102 | throw std::invalid_argument("received negative value");
|
| 103 | }
|
| 104 | return a < b;
|
| 105 | }
|
| 106 |
|
| 107 | int parse(const char* text) {
|
| 108 | if (text[0] == 'f') {
|
| 109 | throw ParseError("started with f");
|
| 110 | }
|
| 111 | return 0;
|
| 112 | }
|
| 113 |
|
| 114 | void throw_fatal() {
|
| 115 | throw FatalError();
|
| 116 | }
|
| 117 |
|
| 118 | void except_subclass_demo() {
|
| 119 | try {
|
| 120 | throw_fatal();
|
| 121 | // parse("f");
|
| 122 | } catch (ParseError& e) {
|
| 123 | // Doesn't get caught. Does this rely on RTTI, or is it static?
|
| 124 | // I think it's static but increases the size of the exception table.
|
| 125 | log("Got ParseError: %s", e.reason());
|
| 126 | }
|
| 127 | }
|
| 128 |
|
| 129 | TEST except_demo() {
|
| 130 | int num_caught = 0;
|
| 131 |
|
| 132 | log("compare(3, 1): %d", compare(1, 3));
|
| 133 | log("compare(5, 4): %d", compare(5, 4));
|
| 134 |
|
| 135 | try {
|
| 136 | log("compare(-1, 3): %d", compare(-1, 3));
|
| 137 | } catch (const std::invalid_argument& e) {
|
| 138 | log("Got exception: %s", e.what());
|
| 139 | num_caught++;
|
| 140 | }
|
| 141 |
|
| 142 | log("");
|
| 143 |
|
| 144 | try {
|
| 145 | log("parse('foo'): %d", parse("foo"));
|
| 146 | } catch (const ParseError& e) {
|
| 147 | log("Got exception: %s", e.reason());
|
| 148 | num_caught++;
|
| 149 | }
|
| 150 |
|
| 151 | try {
|
| 152 | log("parse('bar'): %d", parse("bar"));
|
| 153 | } catch (const ParseError& e) {
|
| 154 | log("Got exception: %s", e.reason());
|
| 155 | num_caught++; // we don't get here
|
| 156 | }
|
| 157 |
|
| 158 | try {
|
| 159 | except_subclass_demo();
|
| 160 | } catch (const FatalError& e) {
|
| 161 | log("Got FatalError");
|
| 162 | num_caught++;
|
| 163 | }
|
| 164 |
|
| 165 | ASSERT_EQ_FMT(3, num_caught, "%d");
|
| 166 |
|
| 167 | PASS();
|
| 168 | }
|
| 169 |
|
| 170 | TEST template_demo() {
|
| 171 | Array<int> a;
|
| 172 | a.append(1);
|
| 173 | a.append(2);
|
| 174 | a.append(3);
|
| 175 | log("a.size() = %d", a.size());
|
| 176 |
|
| 177 | Array<MyList*> a2;
|
| 178 | a2.append(new MyList{1, 2, 3});
|
| 179 | a2.append(new MyList{4, 5, 6});
|
| 180 | log("a2.size() = %d", a2.size());
|
| 181 |
|
| 182 | PASS();
|
| 183 | }
|
| 184 |
|
| 185 | // prototype
|
| 186 | void f(int a, int b = -1, const char* s = nullptr);
|
| 187 |
|
| 188 | void f(int a, int b, const char* s) {
|
| 189 | log("");
|
| 190 | log("a = %d", a);
|
| 191 | log("b = %d", b);
|
| 192 | log("s = %p", s);
|
| 193 | }
|
| 194 |
|
| 195 | class Foo {
|
| 196 | public:
|
| 197 | // Is there any downside to these default args?
|
| 198 | // Only for virtual functions. Note that they are re-evaluated at each call
|
| 199 | // site, which is fine.
|
| 200 | //
|
| 201 | // https://google.github.io/styleguide/cppguide.html#Default_Arguments
|
| 202 | Foo(int i, bool always_strict = false);
|
| 203 |
|
| 204 | void Print() {
|
| 205 | log("i = %d", i);
|
| 206 | log("always_strict = %d", always_strict);
|
| 207 | }
|
| 208 |
|
| 209 | int i;
|
| 210 | bool always_strict;
|
| 211 | };
|
| 212 |
|
| 213 | Foo::Foo(int i, bool always_strict) : i(i), always_strict(always_strict) {
|
| 214 | }
|
| 215 |
|
| 216 | TEST default_args_demo() {
|
| 217 | f(42, 43, "foo");
|
| 218 | f(42, 43);
|
| 219 | f(42);
|
| 220 |
|
| 221 | Foo a(98);
|
| 222 | a.Print();
|
| 223 | Foo b(99, true);
|
| 224 | b.Print();
|
| 225 |
|
| 226 | PASS();
|
| 227 | }
|
| 228 |
|
| 229 | namespace core {
|
| 230 | namespace util {
|
| 231 | void p_die(const char* s) {
|
| 232 | log("p_die %s", s);
|
| 233 | }
|
| 234 | } // namespace util
|
| 235 | } // namespace core
|
| 236 |
|
| 237 | namespace tdop {
|
| 238 | using core::util::p_die;
|
| 239 |
|
| 240 | class Parser {
|
| 241 | public:
|
| 242 | Parser(int token) : token_(token) {
|
| 243 | log("Parser %d", token);
|
| 244 | p_die("Parser");
|
| 245 | }
|
| 246 | int token_;
|
| 247 | };
|
| 248 | } // namespace tdop
|
| 249 |
|
| 250 | namespace typed_arith_parse {
|
| 251 | // using namespace core; This makes EVERYTHING available.
|
| 252 |
|
| 253 | namespace util = core::util;
|
| 254 |
|
| 255 | // This lets us use "Parser""
|
| 256 | using tdop::Parser;
|
| 257 |
|
| 258 | TEST namespace_demo() {
|
| 259 | log("");
|
| 260 | log("namespace_demo()");
|
| 261 | f(42);
|
| 262 | auto unused1 = new tdop::Parser(42);
|
| 263 | auto unused2 = new Parser(43);
|
| 264 | (void)unused1;
|
| 265 | (void)unused2;
|
| 266 |
|
| 267 | util::p_die("ns");
|
| 268 |
|
| 269 | PASS();
|
| 270 | }
|
| 271 | } // namespace typed_arith_parse
|
| 272 |
|
| 273 | // Conclusion: every Python module should have is own namespace
|
| 274 | //
|
| 275 | // from core.util import log => using core::util::log
|
| 276 | // from core import util => namespace util = core::util;
|
| 277 |
|
| 278 | // test out the size of 5 uint16_t. OK it's actually padded, which is nice!
|
| 279 | // Because there is no big element.
|
| 280 | struct Extent {
|
| 281 | uint16_t s_line_id;
|
| 282 | uint16_t s_col;
|
| 283 | uint16_t e_line_id;
|
| 284 | uint16_t e_col;
|
| 285 | uint16_t src_id;
|
| 286 | };
|
| 287 |
|
| 288 | class expr__Const {
|
| 289 | public:
|
| 290 | expr__Const(int i) : i_(i) {
|
| 291 | }
|
| 292 | int i_;
|
| 293 | };
|
| 294 |
|
| 295 | namespace expr {
|
| 296 | typedef expr__Const Const;
|
| 297 | }
|
| 298 |
|
| 299 | using std::make_shared;
|
| 300 | using std::shared_ptr;
|
| 301 |
|
| 302 | shared_ptr<expr__Const> f(shared_ptr<expr__Const> arg) {
|
| 303 | log("arg.use_count() = %d", arg.use_count());
|
| 304 | return shared_ptr<expr__Const>(new expr__Const(arg->i_ + 10));
|
| 305 | }
|
| 306 |
|
| 307 | TEST shared_ptr_demo() {
|
| 308 | std::shared_ptr<expr__Const> e = make_shared<expr__Const>(5);
|
| 309 | log("e->i_ = %d", e->i_);
|
| 310 | log("e.use_count() = %d", e.use_count());
|
| 311 |
|
| 312 | // 16, not 24?
|
| 313 | // These are contiguous.
|
| 314 | log("sizeof(e) = %zu", sizeof(e));
|
| 315 | log("");
|
| 316 |
|
| 317 | std::shared_ptr<expr__Const> e2(new expr__Const(7));
|
| 318 | log("e2->i_ = %d", e2->i_);
|
| 319 | log("e2.use_count() = %d", e2.use_count());
|
| 320 | log("sizeof(e2) = %zu", sizeof(e2));
|
| 321 | log("");
|
| 322 |
|
| 323 | std::shared_ptr<expr__Const> e3 = f(e2);
|
| 324 |
|
| 325 | log("e3->i_ = %d", e3->i_);
|
| 326 | log("e3.use_count() = %d", e3.use_count());
|
| 327 | log("sizeof(e3) = %zu", sizeof(e3));
|
| 328 | log("");
|
| 329 |
|
| 330 | PASS();
|
| 331 | }
|
| 332 |
|
| 333 | TEST map_demo() {
|
| 334 | unordered_map<int, int> m;
|
| 335 | log("m.size = %d", m.size());
|
| 336 |
|
| 337 | // Hm integers have a hash function
|
| 338 | m[3] = 4;
|
| 339 | m[5] = 9;
|
| 340 | log("m.size = %d", m.size());
|
| 341 |
|
| 342 | // Hm you always get the pairs
|
| 343 | // Should this be const auto& or something?
|
| 344 | for (auto item : m) {
|
| 345 | log("iterating %d %d", item.first, item.second);
|
| 346 | }
|
| 347 |
|
| 348 | log("---");
|
| 349 |
|
| 350 | unordered_map<Extent*, int> m2;
|
| 351 | log("m2.size = %d", m2.size());
|
| 352 |
|
| 353 | // hm do I want this operator overloading?
|
| 354 | m2[nullptr] = 42;
|
| 355 | log("m2.size = %d", m2.size());
|
| 356 |
|
| 357 | log("retrieved = %d", m2[nullptr]);
|
| 358 |
|
| 359 | PASS();
|
| 360 | }
|
| 361 |
|
| 362 | TEST sizeof_demo() {
|
| 363 | log("sizeof(int): %d", sizeof(int));
|
| 364 | log("sizeof(int*): %d", sizeof(int*));
|
| 365 | log("sizeof(Extent): %d", sizeof(Extent));
|
| 366 | log("");
|
| 367 |
|
| 368 | // Good, this is 50.
|
| 369 | Extent ext_array[5];
|
| 370 | log("sizeof(ext_array): %d", sizeof(ext_array));
|
| 371 |
|
| 372 | PASS();
|
| 373 | }
|
| 374 |
|
| 375 | TEST test_misc() {
|
| 376 | MyList l{1, 2, 3};
|
| 377 | log("size: %d", l.v_.size());
|
| 378 | log("");
|
| 379 |
|
| 380 | // Dict literal syntax?
|
| 381 | // Dict d {{"key", 1}, {"val", 2}};
|
| 382 |
|
| 383 | log("");
|
| 384 | expr::Const c(42);
|
| 385 | log("expr::Const = %d", c.i_);
|
| 386 |
|
| 387 | // dumb_alloc::Summarize();
|
| 388 |
|
| 389 | PASS();
|
| 390 | }
|
| 391 |
|
| 392 | struct Point {
|
| 393 | int x;
|
| 394 | int y;
|
| 395 | };
|
| 396 |
|
| 397 | // structs don't have any constructors, so don't need any constexpr stuff
|
| 398 | constexpr Point p = {3, 4};
|
| 399 |
|
| 400 | // members must be public to allow initializer list
|
| 401 | class PointC {
|
| 402 | public:
|
| 403 | // constructor is allowed
|
| 404 | // needs to be constexpr
|
| 405 | constexpr PointC(int x, int y) : x_(x), y_(y) {
|
| 406 | }
|
| 407 | // this is allowed too
|
| 408 | int get_x() {
|
| 409 | return x_;
|
| 410 | }
|
| 411 | // this is allowed too
|
| 412 | virtual int mag() const {
|
| 413 | return x_ * x_ + y_ * y_;
|
| 414 | }
|
| 415 |
|
| 416 | int x_;
|
| 417 | int y_;
|
| 418 | };
|
| 419 |
|
| 420 | constexpr PointC pc = {5, 6};
|
| 421 |
|
| 422 | class SubPointC : public PointC {
|
| 423 | public:
|
| 424 | constexpr SubPointC(int x, int y) : PointC(x, y) {
|
| 425 | }
|
| 426 | virtual int mag() const {
|
| 427 | return 0;
|
| 428 | }
|
| 429 | };
|
| 430 |
|
| 431 | constexpr SubPointC sub = {7, 8};
|
| 432 |
|
| 433 | class Compound {
|
| 434 | public:
|
| 435 | PointC c1;
|
| 436 | PointC c2;
|
| 437 | };
|
| 438 |
|
| 439 | // This works, but what about pointers?
|
| 440 | constexpr Compound c = {{0, 1}, {8, 9}};
|
| 441 |
|
| 442 | TEST static_literals() {
|
| 443 | ASSERT_EQ(3, p.x);
|
| 444 | ASSERT_EQ(4, p.y);
|
| 445 |
|
| 446 | ASSERT_EQ(5, pc.x_);
|
| 447 | ASSERT_EQ(6, pc.y_);
|
| 448 |
|
| 449 | // I'm surprised virtual functions are allowed! We're compiling with
|
| 450 | // -std=c++11.
|
| 451 | // But this is just curiosity. We don't need this in ASDL.
|
| 452 | ASSERT_EQ_FMT(61, pc.mag(), "%d");
|
| 453 |
|
| 454 | ASSERT_EQ_FMT(0, sub.mag(), "%d");
|
| 455 |
|
| 456 | ASSERT_EQ(0, c.c1.x_);
|
| 457 | ASSERT_EQ(1, c.c1.y_);
|
| 458 | ASSERT_EQ(8, c.c2.x_);
|
| 459 | ASSERT_EQ(9, c.c2.y_);
|
| 460 |
|
| 461 | PASS();
|
| 462 | }
|
| 463 |
|
| 464 | enum class Color_e { red, blue };
|
| 465 |
|
| 466 | TEST enum_demo() {
|
| 467 | Color_e c1 = Color_e::red;
|
| 468 | Color_e c2 = Color_e::blue;
|
| 469 | int array[2] = {3, 4};
|
| 470 |
|
| 471 | // You can cast these strong enums to an integer. We don't do that in the
|
| 472 | // MyPy source, but maybe we could? It's kind of a pain though.
|
| 473 |
|
| 474 | log("c1 %d", static_cast<int>(c1));
|
| 475 | log("c2 %d", static_cast<int>(c2));
|
| 476 |
|
| 477 | log("array[c1] %d", array[static_cast<int>(c1)]);
|
| 478 |
|
| 479 | PASS();
|
| 480 | }
|
| 481 |
|
| 482 | class Node {
|
| 483 | public:
|
| 484 | int i;
|
| 485 | int j;
|
| 486 | Node* left;
|
| 487 | int k;
|
| 488 | // padding here on 64-bit, but not 32-bit
|
| 489 | Node* right;
|
| 490 | };
|
| 491 |
|
| 492 | #if 0
|
| 493 | constexpr uint16_t Node_mask() {
|
| 494 | uint16_t mask = 0;
|
| 495 |
|
| 496 | constexpr int stride = sizeof(void*);
|
| 497 |
|
| 498 | constexpr int o1 = offsetof(Node, left);
|
| 499 | static_assert(o1 % stride == 0, "oops");
|
| 500 |
|
| 501 | constexpr int o2 = offsetof(Node, right);
|
| 502 | static_assert(o2 % stride == 0, "oops");
|
| 503 |
|
| 504 | constexpr int b1 = o1 / stride;
|
| 505 | constexpr int b2 = o2 / stride;
|
| 506 |
|
| 507 | mask |= 1 << b1;
|
| 508 | mask |= 1 << b2;
|
| 509 |
|
| 510 | return mask;
|
| 511 | }
|
| 512 |
|
| 513 | #else
|
| 514 |
|
| 515 | // C++ 11 version has to be a single expression!
|
| 516 |
|
| 517 | constexpr uint16_t Node_mask() {
|
| 518 | return (1 << (offsetof(Node, left) / sizeof(void*)) |
|
| 519 | 1 << (offsetof(Node, right) / sizeof(void*)));
|
| 520 | }
|
| 521 |
|
| 522 | #endif
|
| 523 |
|
| 524 | void print_bin(int n) {
|
| 525 | for (int i = 15; i >= 0; --i) {
|
| 526 | if (n & (1 << i))
|
| 527 | putchar('1');
|
| 528 | else
|
| 529 | putchar('0');
|
| 530 | }
|
| 531 | putchar('\n');
|
| 532 | }
|
| 533 |
|
| 534 | TEST field_mask_demo() {
|
| 535 | int c1 = offsetof(Node, left);
|
| 536 | int c2 = offsetof(Node, right);
|
| 537 | log("c1 = %d, c2 = %d, sizeof(void*) = %d", c1, c2, sizeof(void*));
|
| 538 |
|
| 539 | log("Node_mask");
|
| 540 | print_bin(Node_mask());
|
| 541 |
|
| 542 | PASS();
|
| 543 | }
|
| 544 |
|
| 545 | class Base {
|
| 546 | public:
|
| 547 | Base(int i) : i(i) {
|
| 548 | }
|
| 549 | static constexpr ObjHeader obj_header() {
|
| 550 | return ObjHeader::ClassFixed(kZeroMask, sizeof(Base));
|
| 551 | }
|
| 552 | int i;
|
| 553 | Node* left;
|
| 554 | Node* right;
|
| 555 | };
|
| 556 |
|
| 557 | class Derived : public Base {
|
| 558 | public:
|
| 559 | Derived(int i, int j) : Base(i), j(j) {
|
| 560 | // annoying: should be in initializer list
|
| 561 | FIELD_MASK(*ObjHeader::FromObject(this)) |= 0x5;
|
| 562 | }
|
| 563 | int j;
|
| 564 | Node* three;
|
| 565 | };
|
| 566 |
|
| 567 | // Demonstrate problem with Local<T>
|
| 568 | #if 0
|
| 569 | TEST smartptr_inheritance_demo() {
|
| 570 | Local<Base> b = Alloc<Base>(2);
|
| 571 | Local<Derived> d = Alloc<Derived>(4, 5);
|
| 572 |
|
| 573 | ASSERT_EQ_FMT(2, b->i, "%d");
|
| 574 |
|
| 575 | ASSERT_EQ_FMT(4, d->i, "%d");
|
| 576 | ASSERT_EQ_FMT(5, d->j, "%d");
|
| 577 |
|
| 578 | ASSERT_EQ_FMT(0x9, b->field_mask_, "%d");
|
| 579 | ASSERT_EQ_FMT(0x5, d->field_mask_, "%d");
|
| 580 |
|
| 581 | PASS();
|
| 582 | }
|
| 583 | #endif
|
| 584 |
|
| 585 | char* realloc(char* buf, size_t num_bytes) {
|
| 586 | void* result = mmap(nullptr, num_bytes, PROT_READ | PROT_WRITE,
|
| 587 | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
|
| 588 | memcpy(result, buf, num_bytes);
|
| 589 |
|
| 590 | // Now make it unreadable
|
| 591 | int m = mprotect(buf, num_bytes, PROT_NONE);
|
| 592 | log("mprotect = %d", m);
|
| 593 |
|
| 594 | return static_cast<char*>(result);
|
| 595 | }
|
| 596 |
|
| 597 | TEST mmap_demo() {
|
| 598 | size_t num_bytes = 1;
|
| 599 |
|
| 600 | void* tmp = mmap(nullptr, num_bytes, PROT_READ | PROT_WRITE,
|
| 601 | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
|
| 602 | char* space = static_cast<char*>(tmp);
|
| 603 |
|
| 604 | *space = 42;
|
| 605 |
|
| 606 | log("space %p", space);
|
| 607 |
|
| 608 | log("value = %d", *space);
|
| 609 |
|
| 610 | space = realloc(space, num_bytes);
|
| 611 | log("value = %d", *space);
|
| 612 |
|
| 613 | // Can't use this anymore
|
| 614 | char* bad = static_cast<char*>(tmp);
|
| 615 | (void)bad;
|
| 616 |
|
| 617 | PASS();
|
| 618 | }
|
| 619 |
|
| 620 | TEST comma_demo() {
|
| 621 | auto i = 3;
|
| 622 | auto k = (i++, 5);
|
| 623 | log("k = %d", k);
|
| 624 |
|
| 625 | auto n = new Node();
|
| 626 | log("n = %p, n->i = %d, n->j = %d", n, n->i, n->j);
|
| 627 |
|
| 628 | // Hacky workaround ... actually this sorta works. Gah.
|
| 629 | Node* tmp;
|
| 630 | auto n2 = (tmp = new Node(), tmp->i = 42, tmp);
|
| 631 | log("n2 = %p, n2->i = %d, n2->j = %d", n2, n2->i, n2->j);
|
| 632 |
|
| 633 | PASS();
|
| 634 | }
|
| 635 |
|
| 636 | // Trick here to print types at compile time
|
| 637 | //
|
| 638 | // https://stackoverflow.com/questions/60203857/print-a-types-name-at-compile-time-without-aborting-compilation
|
| 639 |
|
| 640 | template <typename T>
|
| 641 | [[gnu::warning("your type here")]] bool print_type() {
|
| 642 | return true;
|
| 643 | }
|
| 644 |
|
| 645 | TEST signed_unsigned_demo() {
|
| 646 | char c = '\xff';
|
| 647 | log("c = %d", c);
|
| 648 | log("c = %u", c);
|
| 649 | log("c > 127 = %d", c > 127); // FALSE because it's char
|
| 650 | log("'\\xff' > 127 = %d", '\xff' > 127); // also FALSE
|
| 651 |
|
| 652 | #if 0
|
| 653 | bool b1 = print_type<decltype(c)>();
|
| 654 |
|
| 655 | // The type of literal '\xff' is 'char'
|
| 656 | bool b2 = print_type<decltype('\xff')>();
|
| 657 |
|
| 658 | log("b1 = %d", b1);
|
| 659 | log("b2 = %d", b2);
|
| 660 | #endif
|
| 661 |
|
| 662 | PASS();
|
| 663 | }
|
| 664 |
|
| 665 | class Object {
|
| 666 | public:
|
| 667 | uint32_t header;
|
| 668 | };
|
| 669 |
|
| 670 | class Writer : public Object {
|
| 671 | public:
|
| 672 | // This vtable causes the quirk!
|
| 673 | #if 1
|
| 674 | virtual int f() {
|
| 675 | return 42;
|
| 676 | }
|
| 677 | #endif
|
| 678 | };
|
| 679 |
|
| 680 | void RootGlobalVar(Object* root) {
|
| 681 | // Super weird behavior!!! The param root is 8 bytes ahead of the argument
|
| 682 | // gStdout!
|
| 683 | log("root = %p", root);
|
| 684 | }
|
| 685 |
|
| 686 | Writer* gStdout = nullptr;
|
| 687 |
|
| 688 | Writer* Stdout() {
|
| 689 | if (gStdout == nullptr) {
|
| 690 | gStdout = new Writer();
|
| 691 | log("gStdout = %p", gStdout);
|
| 692 |
|
| 693 | log("no cast");
|
| 694 | RootGlobalVar(gStdout);
|
| 695 | log("");
|
| 696 |
|
| 697 | log("reinterpret_cast");
|
| 698 | RootGlobalVar(reinterpret_cast<Object*>(gStdout));
|
| 699 | log("");
|
| 700 |
|
| 701 | log("static_cast");
|
| 702 | RootGlobalVar(static_cast<Object*>(gStdout));
|
| 703 | log("");
|
| 704 | }
|
| 705 | return gStdout;
|
| 706 | }
|
| 707 |
|
| 708 | TEST param_passing_demo() {
|
| 709 | Writer* writer = Stdout();
|
| 710 | log("writer %p", writer);
|
| 711 | log("");
|
| 712 |
|
| 713 | // Same behavior: surprising!
|
| 714 | Object* obj = writer;
|
| 715 | log("obj %p", obj);
|
| 716 | log("");
|
| 717 |
|
| 718 | PASS();
|
| 719 | }
|
| 720 |
|
| 721 | #define ENUM(name, schema)
|
| 722 |
|
| 723 | #define SUM(name, ...)
|
| 724 |
|
| 725 | #define VARIANT(...)
|
| 726 |
|
| 727 | #define USE(path)
|
| 728 |
|
| 729 | #define SUM_NS(name)
|
| 730 |
|
| 731 | #define PROD(name) struct name
|
| 732 |
|
| 733 | #define SCHEMA(name)
|
| 734 |
|
| 735 | TEST tea_macros_demo() {
|
| 736 | // The preprocessor does NOT expand this. Instead we have a separate parser
|
| 737 | // that does it. Hm not bad.
|
| 738 | //
|
| 739 | // Problem: the processor has to expand imports.
|
| 740 |
|
| 741 | USE("frontend/syntax.asdl");
|
| 742 |
|
| 743 | // Without commas
|
| 744 |
|
| 745 | ENUM(
|
| 746 | suffix_op,
|
| 747 |
|
| 748 | Nullary % Token |
|
| 749 | Unary {
|
| 750 | Token word;
|
| 751 | Word arg_word
|
| 752 | }
|
| 753 |
|
| 754 | );
|
| 755 |
|
| 756 | // More natural comma syntax. Although less consistent with C++.
|
| 757 | // TODO: See what clang-format does on these.
|
| 758 | // Oh it treats:
|
| 759 | // - % and | as binary operators
|
| 760 | // - ; breaks a line but comma doesn't , which I might not want
|
| 761 | //
|
| 762 | // OK () and , looks better, but no line breaking. Maybe there is a
|
| 763 | // clang-format option.
|
| 764 | //
|
| 765 | // Enabled WhitespaceSensitiveMacros for now.
|
| 766 |
|
| 767 | SUM(suffix_op,
|
| 768 | Nullary #Token,
|
| 769 | Unary(Token word, Word arg_word),
|
| 770 | Static(Token tok, BigStr arg)
|
| 771 | );
|
| 772 |
|
| 773 | SUM(suffix_op,
|
| 774 |
|
| 775 | Nullary #Token;
|
| 776 | Unary {
|
| 777 | Token word;
|
| 778 | Word arg_word;
|
| 779 | }
|
| 780 | Static {
|
| 781 | Token tok;
|
| 782 | BigStr arg;
|
| 783 | }
|
| 784 | );
|
| 785 |
|
| 786 | // The C++ compiler parses and validates these
|
| 787 | // Problem: recursive types and so forth. We would need forward declarations
|
| 788 | // and all that?
|
| 789 | // It's also a bit more verbose.
|
| 790 | // How to do the % reference? typedef?
|
| 791 |
|
| 792 | PROD(Token) {
|
| 793 | int id;
|
| 794 | BigStr val;
|
| 795 | };
|
| 796 | struct Word {};
|
| 797 |
|
| 798 | SUM_NS(suffix_op) {
|
| 799 | // typedef Token Nullary;
|
| 800 | struct Unary {
|
| 801 | Token op;
|
| 802 | Word arg_word;
|
| 803 | };
|
| 804 | }
|
| 805 |
|
| 806 | SCHEMA(
|
| 807 | data Token(Id id, BigStr val);
|
| 808 |
|
| 809 | enum suffix_op {
|
| 810 | Nullary %Token
|
| 811 | | Unary(Token op, Word arg_word)
|
| 812 | }
|
| 813 |
|
| 814 | // I guess we retain * for reference semantics and so forth
|
| 815 | // *out = val; can be useful
|
| 816 |
|
| 817 | data Other(Word[] words, Dict<BigStr, Word>* mydict, BigStr? option);
|
| 818 |
|
| 819 | // List<Word>* is also possible, but a bit verbose
|
| 820 | // Word words[] would be more like C++
|
| 821 | //
|
| 822 | // Probably want something more clearly different like:
|
| 823 | //
|
| 824 | // Word... words
|
| 825 | // [Word] words -- synonym for List<Word>* words
|
| 826 | // Word@ words -- not bad, for repetition
|
| 827 | //
|
| 828 | // There are also grammars with + and [] though
|
| 829 | );
|
| 830 |
|
| 831 | printf("Sum types defined");
|
| 832 |
|
| 833 | PASS();
|
| 834 | }
|
| 835 |
|
| 836 | // DEMO of putting PURE interfaces with CALLERS, like Go and TypeScript
|
| 837 | // structural types
|
| 838 |
|
| 839 | // First package
|
| 840 | class Reader1 {
|
| 841 | public:
|
| 842 | virtual int Read(int n) = 0;
|
| 843 | };
|
| 844 |
|
| 845 | // Second package
|
| 846 | class Reader2 {
|
| 847 | public:
|
| 848 | virtual int Read(int n) = 0;
|
| 849 | };
|
| 850 |
|
| 851 | class Writer2 {
|
| 852 | public:
|
| 853 | virtual int Write(int n) = 0;
|
| 854 | };
|
| 855 |
|
| 856 | // Tea could calculate implicit interfaces globally, and then emit these
|
| 857 | // explicit inheritance relationships
|
| 858 |
|
| 859 | // Multiply inheriting from abstract methods seems fine!
|
| 860 | class Concrete : public Reader1, public Reader2, public Writer2 {
|
| 861 | public:
|
| 862 | virtual int Read(int n) {
|
| 863 | printf("Concrete Read(%d)\n", n);
|
| 864 | return 0;
|
| 865 | }
|
| 866 |
|
| 867 | virtual int Write(int n) {
|
| 868 | printf("Concrete Write(%d)\n", n);
|
| 869 | return 0;
|
| 870 | }
|
| 871 | };
|
| 872 |
|
| 873 | class Concrete1 : public Reader1 {};
|
| 874 |
|
| 875 | class Concrete2 : public Reader1, public Reader2 {};
|
| 876 |
|
| 877 | /*
|
| 878 | Would be something like
|
| 879 |
|
| 880 | interface Reader1 {
|
| 881 | func Read(n Int) -> Int
|
| 882 | }
|
| 883 | interface Reader2 {
|
| 884 | func Read(n Int) -> Int
|
| 885 | }
|
| 886 | class Concrete {
|
| 887 | func Read(n Int) -> Int{
|
| 888 | log("echo")
|
| 889 | return 0
|
| 890 | }
|
| 891 | }
|
| 892 | */
|
| 893 |
|
| 894 | TEST tea_interface() {
|
| 895 | Concrete val;
|
| 896 |
|
| 897 | // 8 bytes
|
| 898 | log("sizeof(Concrete1) = %d", sizeof(Concrete1));
|
| 899 | // 16 bytes
|
| 900 | log("sizeof(Concrete2) = %d", sizeof(Concrete2));
|
| 901 |
|
| 902 | // 24 bytes because it has 3 vtables?
|
| 903 | // Hm yes, this idea doesn't scale because objects become bloated with
|
| 904 | // vtables. Though interestingly 2 different vtables can point to the same
|
| 905 | // concrete Read() method.
|
| 906 | log("sizeof(Concrete) = %d", sizeof(Concrete));
|
| 907 | log("sizeof(val) = %d", sizeof(val));
|
| 908 |
|
| 909 | Concrete* c = &val;
|
| 910 | c->Read(3);
|
| 911 |
|
| 912 | Reader1* r1 = c;
|
| 913 | r1->Read(4);
|
| 914 |
|
| 915 | Reader2* r2 = c;
|
| 916 | r2->Read(5);
|
| 917 |
|
| 918 | Writer2* w = c;
|
| 919 | w->Write(6);
|
| 920 |
|
| 921 | PASS();
|
| 922 | }
|
| 923 |
|
| 924 | namespace runtime_asdl {
|
| 925 |
|
| 926 | class lvalue_t {};
|
| 927 |
|
| 928 | class lvalue__Named : public lvalue_t {};
|
| 929 |
|
| 930 | class lvalue__Indexed : public lvalue_t {};
|
| 931 |
|
| 932 | #if 0
|
| 933 | namespace lvalue {
|
| 934 | typedef lvalue__Named Named;
|
| 935 | typedef lvalue__Indexed Indexed;
|
| 936 | }
|
| 937 | #endif
|
| 938 |
|
| 939 | // A CLASS can substitute for a namespace, but it can be "imported" with "using"
|
| 940 | struct lvalue {
|
| 941 | #if 0
|
| 942 | class Named: public lvalue_t {
|
| 943 | };
|
| 944 | class Indexed: public lvalue_t {
|
| 945 | };
|
| 946 | #endif
|
| 947 |
|
| 948 | // typedef lvalue__Named Named;
|
| 949 | // typedef lvalue__Indexed Indexed;
|
| 950 | using Named = lvalue__Named;
|
| 951 | using Indexed = lvalue__Indexed;
|
| 952 | };
|
| 953 |
|
| 954 | }; // namespace runtime_asdl
|
| 955 |
|
| 956 | namespace hnode_asdl {
|
| 957 | #if 0
|
| 958 | namespace hnode_e {
|
| 959 | const int Record = 1;
|
| 960 | const int Array = 2;
|
| 961 | const int Leaf = 3;
|
| 962 | const int External = 4;
|
| 963 | };
|
| 964 | #endif
|
| 965 |
|
| 966 | // Not enum class, a namespace
|
| 967 | struct hnode_e {
|
| 968 | #if 0
|
| 969 | static const int Record = 1;
|
| 970 | static const int Array = 2;
|
| 971 | static const int Leaf = 3;
|
| 972 | static const int External = 4;
|
| 973 | #endif
|
| 974 | enum no_name {
|
| 975 | Record = 1,
|
| 976 | Array = 2,
|
| 977 | Leaf = 3,
|
| 978 | External = 4,
|
| 979 | };
|
| 980 | };
|
| 981 |
|
| 982 | struct scope_e {
|
| 983 | enum no_name {
|
| 984 | Record = 1,
|
| 985 | Array = 2,
|
| 986 | Leaf = 3,
|
| 987 | External = 4,
|
| 988 | };
|
| 989 | };
|
| 990 |
|
| 991 | enum Other {
|
| 992 | Record = 2,
|
| 993 | };
|
| 994 |
|
| 995 | }; // namespace hnode_asdl
|
| 996 |
|
| 997 | using hnode_asdl::hnode_e;
|
| 998 | using runtime_asdl::lvalue;
|
| 999 | // namespace lvalue = runtime_asdl::lvalue;
|
| 1000 |
|
| 1001 | TEST asdl_namespace_demo() {
|
| 1002 | lvalue::Named n;
|
| 1003 | lvalue::Indexed i;
|
| 1004 |
|
| 1005 | (void)n;
|
| 1006 | (void)i;
|
| 1007 |
|
| 1008 | log("Record = %d", hnode_e::Record);
|
| 1009 | log("Array = %d", hnode_e::Array);
|
| 1010 |
|
| 1011 | // In Python, it's lvalue.Named(), not lvalue__Named
|
| 1012 | //
|
| 1013 | // Although you could change that everywhere
|
| 1014 | //
|
| 1015 | // from _devbuild.gen.runtime_asdl import lvalue
|
| 1016 | //
|
| 1017 | // can you reverse it?
|
| 1018 |
|
| 1019 | PASS();
|
| 1020 | }
|
| 1021 |
|
| 1022 | class C1 {
|
| 1023 | public:
|
| 1024 | int i_;
|
| 1025 | };
|
| 1026 |
|
| 1027 | class C2 {
|
| 1028 | public:
|
| 1029 | C2() {
|
| 1030 | }
|
| 1031 | C2(int i) : i_(i) {
|
| 1032 | }
|
| 1033 | int i_ = 42;
|
| 1034 | };
|
| 1035 |
|
| 1036 | // Demo: we can use {} initialization for all fields
|
| 1037 | //
|
| 1038 | // Later, if we turn self.i = i into a initialization list, this will be
|
| 1039 | // cheaper than memset() in theory
|
| 1040 | class C3 {
|
| 1041 | public:
|
| 1042 | C3() {
|
| 1043 | }
|
| 1044 | C3(int i) : i_(i) {
|
| 1045 | }
|
| 1046 | int i_{};
|
| 1047 | double f_{};
|
| 1048 | C2* c2_{};
|
| 1049 | C2* uninitialized;
|
| 1050 | };
|
| 1051 |
|
| 1052 | TEST member_init_demo() {
|
| 1053 | C1 c1;
|
| 1054 | // Uninitialized
|
| 1055 | log("c1.i_ = %d", c1.i_);
|
| 1056 |
|
| 1057 | C2 c2;
|
| 1058 | log("c2.i_ = %d", c2.i_); // from in-class initialization
|
| 1059 |
|
| 1060 | C2 cc2(99);
|
| 1061 | log("cc2.i_ = %d", cc2.i_); // from constructor
|
| 1062 |
|
| 1063 | C3 c3;
|
| 1064 | log("c3.i_ = %d", c3.i_); // in-class
|
| 1065 | log("c3.f_ = %f", c3.f_); // in-class
|
| 1066 | log("c3.c2_ = %p", c3.c2_); // in-class
|
| 1067 | log("c3.uninitialized = %p", c3.uninitialized); // in-class
|
| 1068 |
|
| 1069 | PASS();
|
| 1070 | }
|
| 1071 |
|
| 1072 | GREATEST_MAIN_DEFS();
|
| 1073 |
|
| 1074 | int main(int argc, char** argv) {
|
| 1075 | gHeap.Init(1 << 20);
|
| 1076 |
|
| 1077 | GREATEST_MAIN_BEGIN();
|
| 1078 |
|
| 1079 | RUN_TEST(typed_arith_parse::namespace_demo);
|
| 1080 |
|
| 1081 | RUN_TEST(test_misc);
|
| 1082 | RUN_TEST(map_demo);
|
| 1083 | RUN_TEST(shared_ptr_demo);
|
| 1084 | RUN_TEST(template_demo);
|
| 1085 | RUN_TEST(default_args_demo);
|
| 1086 | RUN_TEST(sizeof_demo);
|
| 1087 | RUN_TEST(except_demo);
|
| 1088 | RUN_TEST(static_literals);
|
| 1089 | RUN_TEST(enum_demo);
|
| 1090 | RUN_TEST(field_mask_demo);
|
| 1091 | // RUN_TEST(smartptr_inheritance_demo);
|
| 1092 |
|
| 1093 | RUN_TEST(mmap_demo);
|
| 1094 | RUN_TEST(comma_demo);
|
| 1095 | RUN_TEST(signed_unsigned_demo);
|
| 1096 | RUN_TEST(param_passing_demo);
|
| 1097 |
|
| 1098 | RUN_TEST(tea_macros_demo);
|
| 1099 | RUN_TEST(tea_interface);
|
| 1100 |
|
| 1101 | RUN_TEST(asdl_namespace_demo);
|
| 1102 |
|
| 1103 | RUN_TEST(member_init_demo);
|
| 1104 |
|
| 1105 | GREATEST_MAIN_END(); /* display results */
|
| 1106 | return 0;
|
| 1107 | }
|