vm.c 39.6 KB
Newer Older
Damien's avatar
Damien committed
1
#include <stdio.h>
2
#include <string.h>
Damien's avatar
Damien committed
3
4
#include <assert.h>

5
#include "nlr.h"
Damien's avatar
Damien committed
6
#include "misc.h"
7
#include "mpconfig.h"
8
#include "qstr.h"
9
#include "obj.h"
10
#include "emitglue.h"
Damien's avatar
Damien committed
11
#include "runtime.h"
12
#include "bc0.h"
13
#include "bc.h"
14
#include "objgenerator.h"
Damien's avatar
Damien committed
15

16
17
#define DETECT_VM_STACK_OVERFLOW (0)

Paul Sokolovsky's avatar
Paul Sokolovsky committed
18
19
20
21
22
23
24
// Value stack grows up (this makes it incompatible with native C stack, but
// makes sure that arguments to functions are in natural order arg1..argN
// (Python semantics mandates left-to-right evaluation order, including for
// function arguments). Stack pointer is pre-incremented and points at the
// top element.
// Exception stack also grows up, top element is also pointed at.

25
// Exception stack unwind reasons (WHY_* in CPython-speak)
26
27
// TODO perhaps compress this to RETURN=0, JUMP>0, with number of unwinds
// left to do encoded in the JUMP number
28
29
typedef enum {
    UNWIND_RETURN = 1,
30
    UNWIND_JUMP,
31
32
} mp_unwind_reason_t;

33
#define DECODE_UINT do { \
34
35
36
37
    unum = 0; \
    do { \
        unum = (unum << 7) + (*ip & 0x7f); \
    } while ((*ip++ & 0x80) != 0); \
38
} while (0)
Damien's avatar
Damien committed
39
40
#define DECODE_ULABEL do { unum = (ip[0] | (ip[1] << 8)); ip += 2; } while (0)
#define DECODE_SLABEL do { unum = (ip[0] | (ip[1] << 8)) - 0x8000; ip += 2; } while (0)
41
#define DECODE_QSTR do { \
42
43
44
45
    qst = 0; \
    do { \
        qst = (qst << 7) + (*ip & 0x7f); \
    } while ((*ip++ & 0x80) != 0); \
46
47
48
49
50
51
} while (0)
#define DECODE_PTR do { \
    ip = (byte*)(((machine_uint_t)ip + sizeof(machine_uint_t) - 1) & (~(sizeof(machine_uint_t) - 1))); /* align ip */ \
    unum = *(machine_uint_t*)ip; \
    ip += sizeof(machine_uint_t); \
} while (0)
52
53
#define PUSH(val) *++sp = (val)
#define POP() (*sp--)
54
55
#define TOP() (*sp)
#define SET_TOP(val) *sp = (val)
Damien's avatar
Damien committed
56

57
#define PUSH_EXC_BLOCK() \
58
59
60
61
62
    DECODE_ULABEL; /* except labels are always forward */ \
    ++exc_sp; \
    exc_sp->opcode = op; \
    exc_sp->handler = ip + unum; \
    exc_sp->val_sp = MP_TAGPTR_MAKE(sp, currently_in_except_block); \
Damien George's avatar
Damien George committed
63
    exc_sp->prev_exc = MP_OBJ_NULL; \
64
65
    currently_in_except_block = 0; /* in a try block now */

66
67
68
69
#define POP_EXC_BLOCK() \
    currently_in_except_block = MP_TAGPTR_TAG(exc_sp->val_sp); /* restore previous state */ \
    exc_sp--; /* pop back to previous exception handler */

70
mp_vm_return_kind_t mp_execute_byte_code(const byte *code, const mp_obj_t *args, uint n_args, const mp_obj_t *args2, uint n_args2, mp_obj_t *ret) {
71
72
73
74
75
76
    const byte *ip = code;

    // get code info size, and skip line number table
    machine_uint_t code_info_size = ip[0] | (ip[1] << 8) | (ip[2] << 16) | (ip[3] << 24);
    ip += code_info_size;

77
78
79
80
    // bytecode prelude: state size and exception stack size; 16 bit uints
    machine_uint_t n_state = ip[0] | (ip[1] << 8);
    machine_uint_t n_exc_stack = ip[2] | (ip[3] << 8);
    ip += 4;
81

82
83
    // allocate state for locals and stack
    mp_obj_t temp_state[10];
84
    mp_obj_t *state = &temp_state[0];
85
#if DETECT_VM_STACK_OVERFLOW
86
87
    n_state += 1;
#endif
88
    if (n_state > 10) {
89
        state = m_new(mp_obj_t, n_state);
90
    }
91
92
    mp_obj_t *sp = &state[0] - 1;

93
    // allocate state for exceptions
94
95
    mp_exc_stack_t exc_state[4];
    mp_exc_stack_t *exc_stack = &exc_state[0];
96
    if (n_exc_stack > 4) {
97
        exc_stack = m_new(mp_exc_stack_t, n_exc_stack);
98
    }
99
    mp_exc_stack_t *exc_sp = &exc_stack[0] - 1;
100

101
    // init args
102
    for (uint i = 0; i < n_args; i++) {
103
        state[n_state - 1 - i] = args[i];
104
    }
105
106
107
    for (uint i = 0; i < n_args2; i++) {
        state[n_state - 1 - n_args - i] = args2[i];
    }
108

109
110
111
112
113
    // set rest of state to MP_OBJ_NULL
    for (uint i = 0; i < n_state - n_args - n_args2; i++) {
        state[i] = MP_OBJ_NULL;
    }

114
115
116
    // bytecode prelude: initialise closed over variables
    for (uint n_local = *ip++; n_local > 0; n_local--) {
        uint local_num = *ip++;
117
        state[n_state - 1 - local_num] = mp_obj_new_cell(state[n_state - 1 - local_num]);
Damien George's avatar
Damien George committed
118
119
120
    }

    // execute the byte code
121
    mp_vm_return_kind_t vm_return_kind = mp_execute_byte_code_2(code, &ip, &state[n_state - 1], &sp, exc_stack, &exc_sp, MP_OBJ_NULL);
122

123
#if DETECT_VM_STACK_OVERFLOW
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
    // We can't check the case when an exception is returned in state[n_state - 1]
    // and there are no arguments, because in this case our detection slot may have
    // been overwritten by the returned exception (which is allowed).
    if (!(vm_return_kind == MP_VM_RETURN_EXCEPTION && n_args == 0 && n_args2 == 0)) {
        // Just check to see that we have at least 1 null object left in the state.
        bool overflow = true;
        for (uint i = 0; i < n_state - n_args - n_args2; i++) {
            if (state[i] == MP_OBJ_NULL) {
                overflow = false;
                break;
            }
        }
        if (overflow) {
            printf("VM stack overflow state=%p n_state+1=%u\n", state, n_state);
            assert(0);
        }
140
141
142
    }
#endif

143
144
145
146
147
148
149
150
151
152
153
154
    switch (vm_return_kind) {
        case MP_VM_RETURN_NORMAL:
            *ret = *sp;
            return MP_VM_RETURN_NORMAL;
        case MP_VM_RETURN_EXCEPTION:
            *ret = state[n_state - 1];
            return MP_VM_RETURN_EXCEPTION;
        case MP_VM_RETURN_YIELD: // byte-code shouldn't yield
        default:
            assert(0);
            *ret = mp_const_none;
            return MP_VM_RETURN_NORMAL;
155
156
157
    }
}

158
159
// fastn has items in reverse order (fastn[0] is local[0], fastn[-1] is local[1], etc)
// sp points to bottom of stack which grows up
160
161
162
163
// returns:
//  MP_VM_RETURN_NORMAL, sp valid, return value in *sp
//  MP_VM_RETURN_YIELD, ip, sp valid, yielded value in *sp
//  MP_VM_RETURN_EXCEPTION, exception in fastn[0]
164
165
mp_vm_return_kind_t mp_execute_byte_code_2(const byte *code_info, const byte **ip_in_out,
                                           mp_obj_t *fastn, mp_obj_t **sp_in_out,
166
                                           mp_exc_stack_t *exc_stack, mp_exc_stack_t **exc_sp_in_out,
167
                                           volatile mp_obj_t inject_exc) {
168
#if MICROPY_USE_COMPUTED_GOTOS
AZ Huang's avatar
AZ Huang committed
169
    #include "vmentrytable.h"
170
171
172
173
    #define DISPATCH() do { \
        save_ip = ip; \
        op = *ip++; \
        goto *entry_table[op]; \
174
    } while(0)
175
176
    #define ENTRY(op) entry_##op
    #define ENTRY_DEFAULT entry_default
177
#else
178
179
180
    #define DISPATCH() break
    #define ENTRY(op) case op
    #define ENTRY_DEFAULT default
181
182
#endif

183
184
185
186
187
    // nlr_raise needs to be implemented as a goto, so that the C compiler's flow analyser
    // sees that it's possible for us to jump from the dispatch loop to the exception
    // handler.  Without this, the code may have a different stack layout in the dispatch
    // loop and the exception handler, leading to very obscure bugs.
    #define RAISE(o) do { nlr_pop(); nlr.ret_val = o; goto exception_handler; } while(0)
Damien's avatar
Damien committed
188

189
    // variables that are visible to the exception handler (declared volatile)
190
    volatile bool currently_in_except_block = MP_TAGPTR_TAG(*exc_sp_in_out); // 0 or 1, to detect nested exceptions
191
    mp_exc_stack_t *volatile exc_sp = MP_TAGPTR_PTR(*exc_sp_in_out); // stack grows up, exc_sp points to top of stack
192
193
    const byte *volatile save_ip = *ip_in_out; // this is so we can access ip in the exception handler without making ip volatile (which means the compiler can't keep it in a register in the main loop)
    mp_obj_t *volatile save_sp = *sp_in_out; // this is so we can access sp in the exception handler when needed
Damien's avatar
Damien committed
194

195
    // outer exception handling loop
Damien's avatar
Damien committed
196
    for (;;) {
197
        nlr_buf_t nlr;
198
outer_dispatch_loop:
199
        if (nlr_push(&nlr) == 0) {
200
201
202
203
204
205
206
207
            // local variables that are not visible to the exception handler
            byte op = 0;
            const byte *ip = *ip_in_out;
            mp_obj_t *sp = *sp_in_out;
            machine_uint_t unum;
            qstr qst;
            mp_obj_t obj1, obj2;

208
209
210
            // If we have exception to inject, now that we finish setting up
            // execution context, raise it. This works as if RAISE_VARARGS
            // bytecode was executed.
211
212
213
            // Injecting exc into yield from generator is a special case,
            // handled by MP_BC_YIELD_FROM itself
            if (inject_exc != MP_OBJ_NULL && *ip != MP_BC_YIELD_FROM) {
214
                obj1 = inject_exc;
215
                inject_exc = MP_OBJ_NULL;
216
217
                obj1 = mp_make_raise_obj(obj1);
                RAISE(obj1);
218
            }
219

220
221
            // loop to execute byte code
            for (;;) {
222
dispatch_loop:
223
#if MICROPY_USE_COMPUTED_GOTOS
224
225
                DISPATCH();
#else
226
                save_ip = ip;
227
                op = *ip++;
228

229
230
231
                switch (op) {
#endif
                //printf("ip=%p sp=%p op=%u\n", save_ip, sp, op);
232

233
234
235
                ENTRY(MP_BC_LOAD_CONST_FALSE):
                    PUSH(mp_const_false);
                    DISPATCH();
236

237
238
239
                ENTRY(MP_BC_LOAD_CONST_NONE):
                    PUSH(mp_const_none);
                    DISPATCH();
240

241
242
243
                ENTRY(MP_BC_LOAD_CONST_TRUE):
                    PUSH(mp_const_true);
                    DISPATCH();
244

245
246
247
                ENTRY(MP_BC_LOAD_CONST_ELLIPSIS):
                    PUSH((mp_obj_t)&mp_const_ellipsis_obj);
                    DISPATCH();
248

249
250
251
252
253
254
255
256
257
258
259
260
                ENTRY(MP_BC_LOAD_CONST_SMALL_INT): {
                    machine_int_t num = 0;
                    if ((ip[0] & 0x40) != 0) {
                        // Number is negative
                        num--;
                    }
                    do {
                        num = (num << 7) | (*ip & 0x7f);
                    } while ((*ip++ & 0x80) != 0);
                    PUSH(MP_OBJ_NEW_SMALL_INT(num));
                    DISPATCH();
                }
261

262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
                ENTRY(MP_BC_LOAD_CONST_INT):
                    DECODE_QSTR;
                    PUSH(mp_obj_new_int_from_long_str(qstr_str(qst)));
                    DISPATCH();

                ENTRY(MP_BC_LOAD_CONST_DEC):
                    DECODE_QSTR;
                    PUSH(mp_load_const_dec(qst));
                    DISPATCH();

                ENTRY(MP_BC_LOAD_CONST_ID):
                    DECODE_QSTR;
                    PUSH(mp_load_const_str(qst)); // TODO
                    DISPATCH();

                ENTRY(MP_BC_LOAD_CONST_BYTES):
                    DECODE_QSTR;
                    PUSH(mp_load_const_bytes(qst));
                    DISPATCH();

                ENTRY(MP_BC_LOAD_CONST_STRING):
                    DECODE_QSTR;
                    PUSH(mp_load_const_str(qst));
                    DISPATCH();

                ENTRY(MP_BC_LOAD_NULL):
                    PUSH(MP_OBJ_NULL);
                    DISPATCH();

                ENTRY(MP_BC_LOAD_FAST_0):
                    obj1 = fastn[0];
                    goto load_check;

                ENTRY(MP_BC_LOAD_FAST_1):
                    obj1 = fastn[-1];
                    goto load_check;

                ENTRY(MP_BC_LOAD_FAST_2):
                    obj1 = fastn[-2];
                    goto load_check;

                ENTRY(MP_BC_LOAD_FAST_N):
                    DECODE_UINT;
                    obj1 = fastn[-unum];
                    load_check:
                    if (obj1 == MP_OBJ_NULL) {
                        local_name_error:
309
310
                        obj1 = mp_obj_new_exception_msg(&mp_type_NameError, "local variable referenced before assignment");
                        RAISE(obj1);
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
                    }
                    PUSH(obj1);
                    DISPATCH();

                ENTRY(MP_BC_LOAD_DEREF):
                    DECODE_UINT;
                    obj1 = mp_obj_cell_get(fastn[-unum]);
                    goto load_check;

                ENTRY(MP_BC_LOAD_NAME):
                    DECODE_QSTR;
                    PUSH(mp_load_name(qst));
                    DISPATCH();

                ENTRY(MP_BC_LOAD_GLOBAL):
                    DECODE_QSTR;
                    PUSH(mp_load_global(qst));
                    DISPATCH();

                ENTRY(MP_BC_LOAD_ATTR):
                    DECODE_QSTR;
                    SET_TOP(mp_load_attr(TOP(), qst));
                    DISPATCH();

                ENTRY(MP_BC_LOAD_METHOD):
                    DECODE_QSTR;
                    mp_load_method(*sp, qst, sp);
                    sp += 1;
                    DISPATCH();

                ENTRY(MP_BC_LOAD_BUILD_CLASS):
                    PUSH(mp_load_build_class());
                    DISPATCH();

                ENTRY(MP_BC_STORE_FAST_0):
                    fastn[0] = POP();
                    DISPATCH();

                ENTRY(MP_BC_STORE_FAST_1):
                    fastn[-1] = POP();
                    DISPATCH();

                ENTRY(MP_BC_STORE_FAST_2):
                    fastn[-2] = POP();
                    DISPATCH();

                ENTRY(MP_BC_STORE_FAST_N):
                    DECODE_UINT;
                    fastn[-unum] = POP();
                    DISPATCH();

                ENTRY(MP_BC_STORE_DEREF):
                    DECODE_UINT;
                    mp_obj_cell_set(fastn[-unum], POP());
                    DISPATCH();

                ENTRY(MP_BC_STORE_NAME):
                    DECODE_QSTR;
                    mp_store_name(qst, POP());
                    DISPATCH();

                ENTRY(MP_BC_STORE_GLOBAL):
                    DECODE_QSTR;
                    mp_store_global(qst, POP());
                    DISPATCH();

                ENTRY(MP_BC_STORE_ATTR):
                    DECODE_QSTR;
                    mp_store_attr(sp[0], qst, sp[-1]);
                    sp -= 2;
                    DISPATCH();

                ENTRY(MP_BC_STORE_SUBSCR):
                    mp_store_subscr(sp[-1], sp[0], sp[-2]);
                    sp -= 3;
                    DISPATCH();

                ENTRY(MP_BC_DELETE_FAST):
                    DECODE_UINT;
                    if (fastn[-unum] == MP_OBJ_NULL) {
                        goto local_name_error;
                    }
                    fastn[-unum] = MP_OBJ_NULL;
                    DISPATCH();
395

396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
                ENTRY(MP_BC_DELETE_DEREF):
                    DECODE_UINT;
                    if (mp_obj_cell_get(fastn[-unum]) == MP_OBJ_NULL) {
                        goto local_name_error;
                    }
                    mp_obj_cell_set(fastn[-unum], MP_OBJ_NULL);
                    DISPATCH();

                ENTRY(MP_BC_DELETE_NAME):
                    DECODE_QSTR;
                    mp_delete_name(qst);
                    DISPATCH();

                ENTRY(MP_BC_DELETE_GLOBAL):
                    DECODE_QSTR;
                    mp_delete_global(qst);
                    DISPATCH();

                ENTRY(MP_BC_DUP_TOP):
                    obj1 = TOP();
                    PUSH(obj1);
                    DISPATCH();

                ENTRY(MP_BC_DUP_TOP_TWO):
                    sp += 2;
                    sp[0] = sp[-2];
                    sp[-1] = sp[-3];
                    DISPATCH();

                ENTRY(MP_BC_POP_TOP):
                    sp -= 1;
                    DISPATCH();

                ENTRY(MP_BC_ROT_TWO):
                    obj1 = sp[0];
                    sp[0] = sp[-1];
                    sp[-1] = obj1;
                    DISPATCH();

                ENTRY(MP_BC_ROT_THREE):
                    obj1 = sp[0];
                    sp[0] = sp[-1];
                    sp[-1] = sp[-2];
                    sp[-2] = obj1;
                    DISPATCH();

                ENTRY(MP_BC_JUMP):
                    DECODE_SLABEL;
                    ip += unum;
                    DISPATCH();

                ENTRY(MP_BC_POP_JUMP_IF_TRUE):
                    DECODE_SLABEL;
                    if (mp_obj_is_true(POP())) {
                        ip += unum;
                    }
                    DISPATCH();
Damien's avatar
Damien committed
453

454
455
456
457
458
459
                ENTRY(MP_BC_POP_JUMP_IF_FALSE):
                    DECODE_SLABEL;
                    if (!mp_obj_is_true(POP())) {
                        ip += unum;
                    }
                    DISPATCH();
Damien's avatar
Damien committed
460

461
462
463
464
465
466
467
468
                ENTRY(MP_BC_JUMP_IF_TRUE_OR_POP):
                    DECODE_SLABEL;
                    if (mp_obj_is_true(TOP())) {
                        ip += unum;
                    } else {
                        sp--;
                    }
                    DISPATCH();
469

470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
                ENTRY(MP_BC_JUMP_IF_FALSE_OR_POP):
                    DECODE_SLABEL;
                    if (mp_obj_is_true(TOP())) {
                        sp--;
                    } else {
                        ip += unum;
                    }
                    DISPATCH();

                    /* we are trying to get away without using this opcode
                ENTRY(MP_BC_SETUP_LOOP):
                    DECODE_UINT;
                    // push_block(MP_BC_SETUP_LOOP, ip + unum, sp)
                    DISPATCH();
                    */

                ENTRY(MP_BC_SETUP_WITH):
                    obj1 = TOP();
                    SET_TOP(mp_load_attr(obj1, MP_QSTR___exit__));
                    mp_load_method(obj1, MP_QSTR___enter__, sp + 1);
                    obj2 = mp_call_method_n_kw(0, 0, sp + 1);
                    PUSH_EXC_BLOCK();
                    PUSH(obj2);
                    DISPATCH();

                ENTRY(MP_BC_WITH_CLEANUP): {
                    // Arriving here, there's "exception control block" on top of stack,
                    // and __exit__ bound method underneath it. Bytecode calls __exit__,
                    // and "deletes" it off stack, shifting "exception control block"
                    // to its place.
                    static const mp_obj_t no_exc[] = {mp_const_none, mp_const_none, mp_const_none};
                    if (TOP() == mp_const_none) {
                        sp--;
503
                        obj1 = TOP();
504
505
506
507
508
509
510
511
512
513
514
                        SET_TOP(mp_const_none);
                        obj2 = mp_call_function_n_kw(obj1, 3, 0, no_exc);
                    } else if (MP_OBJ_IS_SMALL_INT(TOP())) {
                        mp_obj_t cause = POP();
                        switch (MP_OBJ_SMALL_INT_VALUE(cause)) {
                            case UNWIND_RETURN: {
                                mp_obj_t retval = POP();
                                obj2 = mp_call_function_n_kw(TOP(), 3, 0, no_exc);
                                SET_TOP(retval);
                                PUSH(cause);
                                break;
515
                            }
516
517
518
519
520
521
522
                            case UNWIND_JUMP: {
                                obj2 = mp_call_function_n_kw(sp[-2], 3, 0, no_exc);
                                // Pop __exit__ boundmethod at sp[-2]
                                sp[-2] = sp[-1];
                                sp[-1] = sp[0];
                                SET_TOP(cause);
                                break;
523
                            }
524
525
                            default:
                                assert(0);
526
                        }
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
                    } else if (mp_obj_is_exception_type(TOP())) {
                        mp_obj_t args[3] = {sp[0], sp[-1], sp[-2]};
                        obj2 = mp_call_function_n_kw(sp[-3], 3, 0, args);
                        // Pop __exit__ boundmethod at sp[-3]
                        // TODO: Once semantics is proven, optimize for case when obj2 == True
                        sp[-3] = sp[-2];
                        sp[-2] = sp[-1];
                        sp[-1] = sp[0];
                        sp--;
                        if (mp_obj_is_true(obj2)) {
                            // This is what CPython does
                            //PUSH(MP_OBJ_NEW_SMALL_INT(UNWIND_SILENCED));
                            // But what we need to do is - pop exception from value stack...
                            sp -= 3;
                            // ... pop "with" exception handler, and signal END_FINALLY
                            // to just execute finally handler normally (by pushing None
                            // on value stack)
544
                            assert(exc_sp >= exc_stack);
545
546
547
                            assert(exc_sp->opcode == MP_BC_SETUP_WITH);
                            POP_EXC_BLOCK();
                            PUSH(mp_const_none);
548
                        }
549
550
551
552
553
                    } else {
                        assert(0);
                    }
                    DISPATCH();
                }
554

555
556
557
558
559
560
561
562
                ENTRY(MP_BC_UNWIND_JUMP):
                    DECODE_SLABEL;
                    PUSH((void*)(ip + unum)); // push destination ip for jump
                    PUSH((void*)(machine_uint_t)(*ip)); // push number of exception handlers to unwind
unwind_jump:
                    unum = (machine_uint_t)POP(); // get number of exception handlers to unwind
                    while (unum > 0) {
                        unum -= 1;
563
                        assert(exc_sp >= exc_stack);
564
565
566
567
568
569
570
571
572
573
                        if (exc_sp->opcode == MP_BC_SETUP_FINALLY || exc_sp->opcode == MP_BC_SETUP_WITH) {
                            // We're going to run "finally" code as a coroutine
                            // (not calling it recursively). Set up a sentinel
                            // on a stack so it can return back to us when it is
                            // done (when END_FINALLY reached).
                            PUSH((void*)unum); // push number of exception handlers left to unwind
                            PUSH(MP_OBJ_NEW_SMALL_INT(UNWIND_JUMP)); // push sentinel
                            ip = exc_sp->handler; // get exception handler byte code address
                            exc_sp--; // pop exception handler
                            goto dispatch_loop; // run the exception handler
574
                        }
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
                        exc_sp--;
                    }
                    ip = (const byte*)POP(); // pop destination ip for jump
                    DISPATCH();

                // matched against: POP_BLOCK or POP_EXCEPT (anything else?)
                ENTRY(MP_BC_SETUP_EXCEPT):
                ENTRY(MP_BC_SETUP_FINALLY):
                    PUSH_EXC_BLOCK();
                    DISPATCH();

                ENTRY(MP_BC_END_FINALLY):
                    // not fully implemented
                    // if TOS is an exception, reraises the exception (3 values on TOS)
                    // if TOS is None, just pops it and continues
                    // if TOS is an integer, does something else
                    // else error
                    if (mp_obj_is_exception_type(TOP())) {
593
                        RAISE(sp[-1]);
594
595
                    }
                    if (TOP() == mp_const_none) {
596
                        sp--;
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
                    } else if (MP_OBJ_IS_SMALL_INT(TOP())) {
                        // We finished "finally" coroutine and now dispatch back
                        // to our caller, based on TOS value
                        mp_unwind_reason_t reason = MP_OBJ_SMALL_INT_VALUE(POP());
                        switch (reason) {
                            case UNWIND_RETURN:
                                goto unwind_return;
                            case UNWIND_JUMP:
                                goto unwind_jump;
                        }
                        assert(0);
                    } else {
                        assert(0);
                    }
                    DISPATCH();

                ENTRY(MP_BC_GET_ITER):
                    SET_TOP(mp_getiter(TOP()));
                    DISPATCH();

                ENTRY(MP_BC_FOR_ITER):
                    DECODE_ULABEL; // the jump offset if iteration finishes; for labels are always forward
619
                    save_sp = sp;
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
                    obj1 = mp_iternext_allow_raise(TOP());
                    if (obj1 == MP_OBJ_NULL) {
                        --sp; // pop the exhausted iterator
                        ip += unum; // jump to after for-block
                    } else {
                        PUSH(obj1); // push the next iteration value
                    }
                    DISPATCH();

                // matched against: SETUP_EXCEPT, SETUP_FINALLY, SETUP_WITH
                ENTRY(MP_BC_POP_BLOCK):
                    // we are exiting an exception handler, so pop the last one of the exception-stack
                    assert(exc_sp >= exc_stack);
                    POP_EXC_BLOCK();
                    DISPATCH();

                // matched against: SETUP_EXCEPT
                ENTRY(MP_BC_POP_EXCEPT):
                    // TODO need to work out how blocks work etc
                    // pops block, checks it's an exception block, and restores the stack, saving the 3 exception values to local threadstate
                    assert(exc_sp >= exc_stack);
                    assert(currently_in_except_block);
                    //sp = (mp_obj_t*)(*exc_sp--);
                    //exc_sp--; // discard ip
                    POP_EXC_BLOCK();
                    //sp -= 3; // pop 3 exception values
                    DISPATCH();

                ENTRY(MP_BC_NOT):
                    if (TOP() == mp_const_true) {
                        SET_TOP(mp_const_false);
                    } else {
                        SET_TOP(mp_const_true);
                    }
                    DISPATCH();

                ENTRY(MP_BC_UNARY_OP):
                    unum = *ip++;
                    SET_TOP(mp_unary_op(unum, TOP()));
                    DISPATCH();

                ENTRY(MP_BC_BINARY_OP):
                    unum = *ip++;
                    obj2 = POP();
                    obj1 = TOP();
                    SET_TOP(mp_binary_op(unum, obj1, obj2));
                    DISPATCH();

                ENTRY(MP_BC_BUILD_TUPLE):
                    DECODE_UINT;
                    sp -= unum - 1;
                    SET_TOP(mp_obj_new_tuple(unum, sp));
                    DISPATCH();

                ENTRY(MP_BC_BUILD_LIST):
                    DECODE_UINT;
                    sp -= unum - 1;
                    SET_TOP(mp_obj_new_list(unum, sp));
                    DISPATCH();

                ENTRY(MP_BC_LIST_APPEND):
                    DECODE_UINT;
                    // I think it's guaranteed by the compiler that sp[unum] is a list
                    mp_obj_list_append(sp[-unum], sp[0]);
                    sp--;
                    DISPATCH();

                ENTRY(MP_BC_BUILD_MAP):
                    DECODE_UINT;
                    PUSH(mp_obj_new_dict(unum));
                    DISPATCH();

                ENTRY(MP_BC_STORE_MAP):
                    sp -= 2;
                    mp_obj_dict_store(sp[0], sp[2], sp[1]);
                    DISPATCH();

                ENTRY(MP_BC_MAP_ADD):
                    DECODE_UINT;
                    // I think it's guaranteed by the compiler that sp[-unum - 1] is a map
                    mp_obj_dict_store(sp[-unum - 1], sp[0], sp[-1]);
                    sp -= 2;
                    DISPATCH();

                ENTRY(MP_BC_BUILD_SET):
                    DECODE_UINT;
                    sp -= unum - 1;
                    SET_TOP(mp_obj_new_set(unum, sp));
                    DISPATCH();

                ENTRY(MP_BC_SET_ADD):
                    DECODE_UINT;
                    // I think it's guaranteed by the compiler that sp[-unum] is a set
                    mp_obj_set_store(sp[-unum], sp[0]);
                    sp--;
                    DISPATCH();
Damien's avatar
Damien committed
716

717
#if MICROPY_ENABLE_SLICE
718
719
720
721
722
723
724
725
726
727
728
729
730
                ENTRY(MP_BC_BUILD_SLICE):
                    DECODE_UINT;
                    if (unum == 2) {
                        obj2 = POP();
                        obj1 = TOP();
                        SET_TOP(mp_obj_new_slice(obj1, obj2, NULL));
                    } else {
                        obj1 = mp_obj_new_exception_msg(&mp_type_NotImplementedError, "3-argument slice is not supported");
                        nlr_pop();
                        fastn[0] = obj1;
                        return MP_VM_RETURN_EXCEPTION;
                    }
                    DISPATCH();
731
#endif
732

733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
                ENTRY(MP_BC_UNPACK_SEQUENCE):
                    DECODE_UINT;
                    mp_unpack_sequence(sp[0], unum, sp);
                    sp += unum - 1;
                    DISPATCH();

                ENTRY(MP_BC_UNPACK_EX):
                    DECODE_UINT;
                    mp_unpack_ex(sp[0], unum, sp);
                    sp += (unum & 0xff) + ((unum >> 8) & 0xff);
                    DISPATCH();

                ENTRY(MP_BC_MAKE_FUNCTION):
                    DECODE_PTR;
                    PUSH(mp_make_function_from_raw_code((mp_raw_code_t*)unum, MP_OBJ_NULL, MP_OBJ_NULL));
                    DISPATCH();

                ENTRY(MP_BC_MAKE_FUNCTION_DEFARGS):
                    DECODE_PTR;
                    // Stack layout: def_tuple def_dict <- TOS
                    obj1 = POP();
                    SET_TOP(mp_make_function_from_raw_code((mp_raw_code_t*)unum, TOP(), obj1));
                    DISPATCH();

                ENTRY(MP_BC_MAKE_CLOSURE):
                    DECODE_PTR;
                    // Stack layout: closure_tuple <- TOS
                    SET_TOP(mp_make_closure_from_raw_code((mp_raw_code_t*)unum, TOP(), MP_OBJ_NULL, MP_OBJ_NULL));
                    DISPATCH();

                ENTRY(MP_BC_MAKE_CLOSURE_DEFARGS):
                    DECODE_PTR;
                    // Stack layout: def_tuple def_dict closure_tuple <- TOS
                    obj1 = POP();
                    obj2 = POP();
                    SET_TOP(mp_make_closure_from_raw_code((mp_raw_code_t*)unum, obj1, TOP(), obj2));
                    DISPATCH();

                ENTRY(MP_BC_CALL_FUNCTION):
                    DECODE_UINT;
                    // unum & 0xff == n_positional
                    // (unum >> 8) & 0xff == n_keyword
                    sp -= (unum & 0xff) + ((unum >> 7) & 0x1fe);
                    SET_TOP(mp_call_function_n_kw(*sp, unum & 0xff, (unum >> 8) & 0xff, sp + 1));
                    DISPATCH();

                ENTRY(MP_BC_CALL_FUNCTION_VAR_KW):
                    DECODE_UINT;
                    // unum & 0xff == n_positional
                    // (unum >> 8) & 0xff == n_keyword
                    // We have folowing stack layout here:
                    // fun arg0 arg1 ... kw0 val0 kw1 val1 ... seq dict <- TOS
                    sp -= (unum & 0xff) + ((unum >> 7) & 0x1fe) + 2;
                    SET_TOP(mp_call_method_n_kw_var(false, unum, sp));
                    DISPATCH();

                ENTRY(MP_BC_CALL_METHOD):
                    DECODE_UINT;
                    // unum & 0xff == n_positional
                    // (unum >> 8) & 0xff == n_keyword
                    sp -= (unum & 0xff) + ((unum >> 7) & 0x1fe) + 1;
                    SET_TOP(mp_call_method_n_kw(unum & 0xff, (unum >> 8) & 0xff, sp));
                    DISPATCH();

                ENTRY(MP_BC_CALL_METHOD_VAR_KW):
                    DECODE_UINT;
                    // unum & 0xff == n_positional
                    // (unum >> 8) & 0xff == n_keyword
                    // We have folowing stack layout here:
                    // fun self arg0 arg1 ... kw0 val0 kw1 val1 ... seq dict <- TOS
                    sp -= (unum & 0xff) + ((unum >> 7) & 0x1fe) + 3;
                    SET_TOP(mp_call_method_n_kw_var(true, unum, sp));
                    DISPATCH();

                ENTRY(MP_BC_RETURN_VALUE):
808
unwind_return:
809
810
811
812
813
814
815
816
817
818
819
                    while (exc_sp >= exc_stack) {
                        if (exc_sp->opcode == MP_BC_SETUP_FINALLY || exc_sp->opcode == MP_BC_SETUP_WITH) {
                            // We're going to run "finally" code as a coroutine
                            // (not calling it recursively). Set up a sentinel
                            // on a stack so it can return back to us when it is
                            // done (when END_FINALLY reached).
                            PUSH(MP_OBJ_NEW_SMALL_INT(UNWIND_RETURN));
                            ip = exc_sp->handler;
                            // We don't need to do anything with sp, finally is just
                            // syntactic sugar for sequential execution??
                            // sp =
820
                            exc_sp--;
821
                            goto dispatch_loop;
822
                        }
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
                        exc_sp--;
                    }
                    nlr_pop();
                    *sp_in_out = sp;
                    assert(exc_sp == exc_stack - 1);
                    return MP_VM_RETURN_NORMAL;

                ENTRY(MP_BC_RAISE_VARARGS):
                    unum = *ip++;
                    assert(unum <= 1);
                    if (unum == 0) {
                        // search for the inner-most previous exception, to reraise it
                        obj1 = MP_OBJ_NULL;
                        for (mp_exc_stack_t *e = exc_sp; e >= exc_stack; e--) {
                            if (e->prev_exc != MP_OBJ_NULL) {
                                obj1 = e->prev_exc;
                                break;
840
                            }
841
                        }
842
                        if (obj1 == MP_OBJ_NULL) {
843
844
                            obj1 = mp_obj_new_exception_msg(&mp_type_RuntimeError, "No active exception to reraise");
                            RAISE(obj1);
845
846
847
848
                        }
                    } else {
                        obj1 = POP();
                    }
849
850
                    obj1 = mp_make_raise_obj(obj1);
                    RAISE(obj1);
851

852
                ENTRY(MP_BC_YIELD_VALUE):
853
yield:
854
855
856
857
858
                    nlr_pop();
                    *ip_in_out = ip;
                    *sp_in_out = sp;
                    *exc_sp_in_out = MP_TAGPTR_MAKE(exc_sp, currently_in_except_block);
                    return MP_VM_RETURN_YIELD;
859

860
                ENTRY(MP_BC_YIELD_FROM): {
861
862
//#define EXC_MATCH(exc, type) MP_OBJ_IS_TYPE(exc, type)
#define EXC_MATCH(exc, type) mp_obj_exception_match(exc, type)
863
#define GENERATOR_EXIT_IF_NEEDED(t) if (t != MP_OBJ_NULL && EXC_MATCH(t, &mp_type_GeneratorExit)) { RAISE(t); }
864
865
866
867
868
869
870
871
872
873
                    mp_vm_return_kind_t ret_kind;
                    obj1 = POP();
                    mp_obj_t t_exc = MP_OBJ_NULL;
                    if (inject_exc != MP_OBJ_NULL) {
                        t_exc = inject_exc;
                        inject_exc = MP_OBJ_NULL;
                        ret_kind = mp_resume(TOP(), MP_OBJ_NULL, t_exc, &obj2);
                    } else {
                        ret_kind = mp_resume(TOP(), obj1, MP_OBJ_NULL, &obj2);
                    }
874

875
876
877
878
879
880
881
882
883
884
885
886
887
                    if (ret_kind == MP_VM_RETURN_YIELD) {
                        ip--;
                        PUSH(obj2);
                        goto yield;
                    }
                    if (ret_kind == MP_VM_RETURN_NORMAL) {
                        // Pop exhausted gen
                        sp--;
                        if (obj2 == MP_OBJ_NULL) {
                            // Optimize StopIteration
                            // TODO: get StopIteration's value
                            PUSH(mp_const_none);
                        } else {
888
889
890
                            PUSH(obj2);
                        }

891
892
893
894
895
896
897
898
899
900
                        // If we injected GeneratorExit downstream, then even
                        // if it was swallowed, we re-raise GeneratorExit
                        GENERATOR_EXIT_IF_NEEDED(t_exc);
                        DISPATCH();
                    }
                    if (ret_kind == MP_VM_RETURN_EXCEPTION) {
                        // Pop exhausted gen
                        sp--;
                        if (EXC_MATCH(obj2, &mp_type_StopIteration)) {
                            PUSH(mp_obj_exception_get_value(obj2));
901
902
903
                            // If we injected GeneratorExit downstream, then even
                            // if it was swallowed, we re-raise GeneratorExit
                            GENERATOR_EXIT_IF_NEEDED(t_exc);
904
905
                            DISPATCH();
                        } else {
906
                            RAISE(obj2);
907
908
                        }
                    }
909
                }
910

911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
                ENTRY(MP_BC_IMPORT_NAME):
                    DECODE_QSTR;
                    obj1 = POP();
                    SET_TOP(mp_import_name(qst, obj1, TOP()));
                    DISPATCH();

                ENTRY(MP_BC_IMPORT_FROM):
                    DECODE_QSTR;
                    obj1 = mp_import_from(TOP(), qst);
                    PUSH(obj1);
                    DISPATCH();

                ENTRY(MP_BC_IMPORT_STAR):
                    mp_import_all(POP());
                    DISPATCH();

                ENTRY_DEFAULT:
                    obj1 = mp_obj_new_exception_msg(&mp_type_NotImplementedError, "byte code not implemented");
                    nlr_pop();
                    fastn[0] = obj1;
                    return MP_VM_RETURN_EXCEPTION;
932

933
934
#if !MICROPY_USE_COMPUTED_GOTOS
                } // switch
935
#endif
936
            } // for loop
937
938

        } else {
939
exception_handler:
940
941
            // exception occurred

942
943
            // check if it's a StopIteration within a for block
            if (*save_ip == MP_BC_FOR_ITER && mp_obj_is_subclass_fast(mp_obj_get_type(nlr.ret_val), &mp_type_StopIteration)) {
944
945
                const byte *ip = save_ip + 1;
                machine_uint_t unum;
946
                DECODE_ULABEL; // the jump offset if iteration finishes; for labels are always forward
947
948
                *ip_in_out = ip + unum; // jump to after for-block
                *sp_in_out = save_sp - 1; // pop the exhausted iterator
949
950
951
                goto outer_dispatch_loop; // continue with dispatch loop
            }

952
            // set file and line number that the exception occurred at
953
954
            // TODO: don't set traceback for exceptions re-raised by END_FINALLY.
            // But consider how to handle nested exceptions.
955
956
            // TODO need a better way of not adding traceback to constant objects (right now, just GeneratorExit_obj and MemoryError_obj)
            if (mp_obj_is_exception_instance(nlr.ret_val) && nlr.ret_val != &mp_const_GeneratorExit_obj && nlr.ret_val != &mp_const_MemoryError_obj) {
957
                machine_uint_t code_info_size = code_info[0] | (code_info[1] << 8) | (code_info[2] << 16) | (code_info[3] << 24);
958
959
                qstr source_file = code_info[4] | (code_info[5] << 8) | (code_info[6] << 16) | (code_info[7] << 24);
                qstr block_name = code_info[8] | (code_info[9] << 8) | (code_info[10] << 16) | (code_info[11] << 24);
960
961
                machine_uint_t source_line = 1;
                machine_uint_t bc = save_ip - code_info - code_info_size;
962
                //printf("find %lu %d %d\n", bc, code_info[12], code_info[13]);
963
964
965
                for (const byte* ci = code_info + 12; *ci && bc >= ((*ci) & 31); ci++) {
                    bc -= *ci & 31;
                    source_line += *ci >> 5;
966
                }
967
                mp_obj_exception_add_traceback(nlr.ret_val, source_file, source_line, block_name);
968
969
            }

970
971
972
            while (currently_in_except_block) {
                // nested exception

973
                assert(exc_sp >= exc_stack);
974
975
976
977
978

                // TODO make a proper message for nested exception
                // at the moment we are just raising the very last exception (the one that caused the nested exception)

                // move up to previous exception handler
979
                POP_EXC_BLOCK();
980
981
            }

982
            if (exc_sp >= exc_stack) {
983
984
985
                // set flag to indicate that we are now handling an exception
                currently_in_except_block = 1;

986
                // catch exception and pass to byte code
987
988
                *ip_in_out = exc_sp->handler;
                mp_obj_t *sp = MP_TAGPTR_PTR(exc_sp->val_sp);
Damien George's avatar
Damien George committed
989
990
                // save this exception in the stack so it can be used in a reraise, if needed
                exc_sp->prev_exc = nlr.ret_val;
Damien's avatar
Damien committed
991
                // push(traceback, exc-val, exc-type)
992
                PUSH(mp_const_none);
Damien's avatar
Damien committed
993
                PUSH(nlr.ret_val);
994
                PUSH(mp_obj_get_type(nlr.ret_val));
995
                *sp_in_out = sp;
996

997
            } else {
998
999
1000
                // propagate exception to higher level
                // TODO what to do about ip and sp? they don't really make sense at this point
                fastn[0] = nlr.ret_val; // must put exception here because sp is invalid
For faster browsing, not all history is shown. View entire blame