1 /*
2 * This file compiles an abstract syntax tree (AST) into Python bytecode.
3 *
4 * The primary entry point is PyAST_Compile(), which returns a
5 * PyCodeObject. The compiler makes several passes to build the code
6 * object:
7 * 1. Checks for future statements. See future.c
8 * 2. Builds a symbol table. See symtable.c.
9 * 3. Generate code for basic blocks. See compiler_mod() in this file.
10 * 4. Assemble the basic blocks into final code. See assemble() in
11 * this file.
12 * 5. Optimize the byte code (peephole optimizations). See peephole.c
13 *
14 * Note that compiler_mod() suggests module, but the module ast type
15 * (mod_ty) has cases for expressions and interactive statements.
16 *
17 * CAUTION: The VISIT_* macros abort the current function when they
18 * encounter a problem. So don't invoke them when there is memory
19 * which needs to be released. Code blocks are OK, as the compiler
20 * structure takes care of releasing those. Use the arena to manage
21 * objects.
22 */
23
24 #include "Python.h"
25
26 #include "Python-ast.h"
27 #include "node.h"
28 #include "ast.h"
29 #include "code.h"
30 #include "symtable.h"
31 #include "opcode.h"
32 #include "wordcode_helpers.h"
33
34 #define DEFAULT_BLOCK_SIZE 16
35 #define DEFAULT_BLOCKS 8
36 #define DEFAULT_CODE_SIZE 128
37 #define DEFAULT_LNOTAB_SIZE 16
38
39 #define COMP_GENEXP 0
40 #define COMP_LISTCOMP 1
41 #define COMP_SETCOMP 2
42 #define COMP_DICTCOMP 3
43
44 struct instr {
45 unsigned i_jabs : 1;
46 unsigned i_jrel : 1;
47 unsigned char i_opcode;
48 int i_oparg;
49 struct basicblock_ *i_target; /* target block (if jump instruction) */
50 int i_lineno;
51 };
52
53 typedef struct basicblock_ {
54 /* Each basicblock in a compilation unit is linked via b_list in the
55 reverse order that the block are allocated. b_list points to the next
56 block, not to be confused with b_next, which is next by control flow. */
57 struct basicblock_ *b_list;
58 /* number of instructions used */
59 int b_iused;
60 /* length of instruction array (b_instr) */
61 int b_ialloc;
62 /* pointer to an array of instructions, initially NULL */
63 struct instr *b_instr;
64 /* If b_next is non-NULL, it is a pointer to the next
65 block reached by normal control flow. */
66 struct basicblock_ *b_next;
67 /* b_seen is used to perform a DFS of basicblocks. */
68 unsigned b_seen : 1;
69 /* b_return is true if a RETURN_VALUE opcode is inserted. */
70 unsigned b_return : 1;
71 /* depth of stack upon entry of block, computed by stackdepth() */
72 int b_startdepth;
73 /* instruction offset for block, computed by assemble_jump_offsets() */
74 int b_offset;
75 } basicblock;
76
77 /* fblockinfo tracks the current frame block.
78
79 A frame block is used to handle loops, try/except, and try/finally.
80 It's called a frame block to distinguish it from a basic block in the
81 compiler IR.
82 */
83
84 enum fblocktype { LOOP, EXCEPT, FINALLY_TRY, FINALLY_END };
85
86 struct fblockinfo {
87 enum fblocktype fb_type;
88 basicblock *fb_block;
89 };
90
91 enum {
92 COMPILER_SCOPE_MODULE,
93 COMPILER_SCOPE_CLASS,
94 COMPILER_SCOPE_FUNCTION,
95 COMPILER_SCOPE_ASYNC_FUNCTION,
96 COMPILER_SCOPE_LAMBDA,
97 COMPILER_SCOPE_COMPREHENSION,
98 };
99
100 /* The following items change on entry and exit of code blocks.
101 They must be saved and restored when returning to a block.
102 */
103 struct compiler_unit {
104 PySTEntryObject *u_ste;
105
106 PyObject *u_name;
107 PyObject *u_qualname; /* dot-separated qualified name (lazy) */
108 int u_scope_type;
109
110 /* The following fields are dicts that map objects to
111 the index of them in co_XXX. The index is used as
112 the argument for opcodes that refer to those collections.
113 */
114 PyObject *u_consts; /* all constants */
115 PyObject *u_names; /* all names */
116 PyObject *u_varnames; /* local variables */
117 PyObject *u_cellvars; /* cell variables */
118 PyObject *u_freevars; /* free variables */
119
120 PyObject *u_private; /* for private name mangling */
121
122 Py_ssize_t u_argcount; /* number of arguments for block */
123 Py_ssize_t u_kwonlyargcount; /* number of keyword only arguments for block */
124 /* Pointer to the most recently allocated block. By following b_list
125 members, you can reach all early allocated blocks. */
126 basicblock *u_blocks;
127 basicblock *u_curblock; /* pointer to current block */
128
129 int u_nfblocks;
130 struct fblockinfo u_fblock[CO_MAXBLOCKS];
131
132 int u_firstlineno; /* the first lineno of the block */
133 int u_lineno; /* the lineno for the current stmt */
134 int u_col_offset; /* the offset of the current stmt */
135 int u_lineno_set; /* boolean to indicate whether instr
136 has been generated with current lineno */
137 };
138
139 /* This struct captures the global state of a compilation.
140
141 The u pointer points to the current compilation unit, while units
142 for enclosing blocks are stored in c_stack. The u and c_stack are
143 managed by compiler_enter_scope() and compiler_exit_scope().
144
145 Note that we don't track recursion levels during compilation - the
146 task of detecting and rejecting excessive levels of nesting is
147 handled by the symbol analysis pass.
148
149 */
150
151 struct compiler {
152 PyObject *c_filename;
153 struct symtable *c_st;
154 PyFutureFeatures *c_future; /* pointer to module's __future__ */
155 PyCompilerFlags *c_flags;
156
157 int c_optimize; /* optimization level */
158 int c_interactive; /* true if in interactive mode */
159 int c_nestlevel;
160
161 struct compiler_unit *u; /* compiler state for current block */
162 PyObject *c_stack; /* Python list holding compiler_unit ptrs */
163 PyArena *c_arena; /* pointer to memory allocation arena */
164 };
165
166 static int compiler_enter_scope(struct compiler *, identifier, int, void *, int);
167 static void compiler_free(struct compiler *);
168 static basicblock *compiler_new_block(struct compiler *);
169 static int compiler_next_instr(struct compiler *, basicblock *);
170 static int compiler_addop(struct compiler *, int);
171 static int compiler_addop_o(struct compiler *, int, PyObject *, PyObject *);
172 static int compiler_addop_i(struct compiler *, int, Py_ssize_t);
173 static int compiler_addop_j(struct compiler *, int, basicblock *, int);
174 static int compiler_error(struct compiler *, const char *);
175 static int compiler_nameop(struct compiler *, identifier, expr_context_ty);
176
177 static PyCodeObject *compiler_mod(struct compiler *, mod_ty);
178 static int compiler_visit_stmt(struct compiler *, stmt_ty);
179 static int compiler_visit_keyword(struct compiler *, keyword_ty);
180 static int compiler_visit_expr(struct compiler *, expr_ty);
181 static int compiler_augassign(struct compiler *, stmt_ty);
182 static int compiler_annassign(struct compiler *, stmt_ty);
183 static int compiler_visit_slice(struct compiler *, slice_ty,
184 expr_context_ty);
185
186 static int compiler_push_fblock(struct compiler *, enum fblocktype,
187 basicblock *);
188 static void compiler_pop_fblock(struct compiler *, enum fblocktype,
189 basicblock *);
190 /* Returns true if there is a loop on the fblock stack. */
191 static int compiler_in_loop(struct compiler *);
192
193 static int inplace_binop(struct compiler *, operator_ty);
194 static int expr_constant(expr_ty);
195
196 static int compiler_with(struct compiler *, stmt_ty, int);
197 static int compiler_async_with(struct compiler *, stmt_ty, int);
198 static int compiler_async_for(struct compiler *, stmt_ty);
199 static int compiler_call_helper(struct compiler *c, int n,
200 asdl_seq *args,
201 asdl_seq *keywords);
202 static int compiler_try_except(struct compiler *, stmt_ty);
203 static int compiler_set_qualname(struct compiler *);
204
205 static int compiler_sync_comprehension_generator(
206 struct compiler *c,
207 asdl_seq *generators, int gen_index,
208 expr_ty elt, expr_ty val, int type);
209
210 static int compiler_async_comprehension_generator(
211 struct compiler *c,
212 asdl_seq *generators, int gen_index,
213 expr_ty elt, expr_ty val, int type);
214
215 static PyCodeObject *assemble(struct compiler *, int addNone);
216 static PyObject *__doc__, *__annotations__;
217
218 #define CAPSULE_NAME "compile.c compiler unit"
219
220 PyObject *
_Py_Mangle(PyObject * privateobj,PyObject * ident)221 _Py_Mangle(PyObject *privateobj, PyObject *ident)
222 {
223 /* Name mangling: __private becomes _classname__private.
224 This is independent from how the name is used. */
225 PyObject *result;
226 size_t nlen, plen, ipriv;
227 Py_UCS4 maxchar;
228 if (privateobj == NULL || !PyUnicode_Check(privateobj) ||
229 PyUnicode_READ_CHAR(ident, 0) != '_' ||
230 PyUnicode_READ_CHAR(ident, 1) != '_') {
231 Py_INCREF(ident);
232 return ident;
233 }
234 nlen = PyUnicode_GET_LENGTH(ident);
235 plen = PyUnicode_GET_LENGTH(privateobj);
236 /* Don't mangle __id__ or names with dots.
237
238 The only time a name with a dot can occur is when
239 we are compiling an import statement that has a
240 package name.
241
242 TODO(jhylton): Decide whether we want to support
243 mangling of the module name, e.g. __M.X.
244 */
245 if ((PyUnicode_READ_CHAR(ident, nlen-1) == '_' &&
246 PyUnicode_READ_CHAR(ident, nlen-2) == '_') ||
247 PyUnicode_FindChar(ident, '.', 0, nlen, 1) != -1) {
248 Py_INCREF(ident);
249 return ident; /* Don't mangle __whatever__ */
250 }
251 /* Strip leading underscores from class name */
252 ipriv = 0;
253 while (PyUnicode_READ_CHAR(privateobj, ipriv) == '_')
254 ipriv++;
255 if (ipriv == plen) {
256 Py_INCREF(ident);
257 return ident; /* Don't mangle if class is just underscores */
258 }
259 plen -= ipriv;
260
261 if (plen + nlen >= PY_SSIZE_T_MAX - 1) {
262 PyErr_SetString(PyExc_OverflowError,
263 "private identifier too large to be mangled");
264 return NULL;
265 }
266
267 maxchar = PyUnicode_MAX_CHAR_VALUE(ident);
268 if (PyUnicode_MAX_CHAR_VALUE(privateobj) > maxchar)
269 maxchar = PyUnicode_MAX_CHAR_VALUE(privateobj);
270
271 result = PyUnicode_New(1 + nlen + plen, maxchar);
272 if (!result)
273 return 0;
274 /* ident = "_" + priv[ipriv:] + ident # i.e. 1+plen+nlen bytes */
275 PyUnicode_WRITE(PyUnicode_KIND(result), PyUnicode_DATA(result), 0, '_');
276 if (PyUnicode_CopyCharacters(result, 1, privateobj, ipriv, plen) < 0) {
277 Py_DECREF(result);
278 return NULL;
279 }
280 if (PyUnicode_CopyCharacters(result, plen+1, ident, 0, nlen) < 0) {
281 Py_DECREF(result);
282 return NULL;
283 }
284 assert(_PyUnicode_CheckConsistency(result, 1));
285 return result;
286 }
287
288 static int
compiler_init(struct compiler * c)289 compiler_init(struct compiler *c)
290 {
291 memset(c, 0, sizeof(struct compiler));
292
293 c->c_stack = PyList_New(0);
294 if (!c->c_stack)
295 return 0;
296
297 return 1;
298 }
299
300 PyCodeObject *
PyAST_CompileObject(mod_ty mod,PyObject * filename,PyCompilerFlags * flags,int optimize,PyArena * arena)301 PyAST_CompileObject(mod_ty mod, PyObject *filename, PyCompilerFlags *flags,
302 int optimize, PyArena *arena)
303 {
304 struct compiler c;
305 PyCodeObject *co = NULL;
306 PyCompilerFlags local_flags;
307 int merged;
308
309 if (!__doc__) {
310 __doc__ = PyUnicode_InternFromString("__doc__");
311 if (!__doc__)
312 return NULL;
313 }
314 if (!__annotations__) {
315 __annotations__ = PyUnicode_InternFromString("__annotations__");
316 if (!__annotations__)
317 return NULL;
318 }
319 if (!compiler_init(&c))
320 return NULL;
321 Py_INCREF(filename);
322 c.c_filename = filename;
323 c.c_arena = arena;
324 c.c_future = PyFuture_FromASTObject(mod, filename);
325 if (c.c_future == NULL)
326 goto finally;
327 if (!flags) {
328 local_flags.cf_flags = 0;
329 flags = &local_flags;
330 }
331 merged = c.c_future->ff_features | flags->cf_flags;
332 c.c_future->ff_features = merged;
333 flags->cf_flags = merged;
334 c.c_flags = flags;
335 c.c_optimize = (optimize == -1) ? Py_OptimizeFlag : optimize;
336 c.c_nestlevel = 0;
337
338 if (!_PyAST_Optimize(mod, arena, c.c_optimize)) {
339 goto finally;
340 }
341
342 c.c_st = PySymtable_BuildObject(mod, filename, c.c_future);
343 if (c.c_st == NULL) {
344 if (!PyErr_Occurred())
345 PyErr_SetString(PyExc_SystemError, "no symtable");
346 goto finally;
347 }
348
349 co = compiler_mod(&c, mod);
350
351 finally:
352 compiler_free(&c);
353 assert(co || PyErr_Occurred());
354 return co;
355 }
356
357 PyCodeObject *
PyAST_CompileEx(mod_ty mod,const char * filename_str,PyCompilerFlags * flags,int optimize,PyArena * arena)358 PyAST_CompileEx(mod_ty mod, const char *filename_str, PyCompilerFlags *flags,
359 int optimize, PyArena *arena)
360 {
361 PyObject *filename;
362 PyCodeObject *co;
363 filename = PyUnicode_DecodeFSDefault(filename_str);
364 if (filename == NULL)
365 return NULL;
366 co = PyAST_CompileObject(mod, filename, flags, optimize, arena);
367 Py_DECREF(filename);
368 return co;
369
370 }
371
372 PyCodeObject *
PyNode_Compile(struct _node * n,const char * filename)373 PyNode_Compile(struct _node *n, const char *filename)
374 {
375 PyCodeObject *co = NULL;
376 mod_ty mod;
377 PyArena *arena = PyArena_New();
378 if (!arena)
379 return NULL;
380 mod = PyAST_FromNode(n, NULL, filename, arena);
381 if (mod)
382 co = PyAST_Compile(mod, filename, NULL, arena);
383 PyArena_Free(arena);
384 return co;
385 }
386
387 static void
compiler_free(struct compiler * c)388 compiler_free(struct compiler *c)
389 {
390 if (c->c_st)
391 PySymtable_Free(c->c_st);
392 if (c->c_future)
393 PyObject_Free(c->c_future);
394 Py_XDECREF(c->c_filename);
395 Py_DECREF(c->c_stack);
396 }
397
398 static PyObject *
list2dict(PyObject * list)399 list2dict(PyObject *list)
400 {
401 Py_ssize_t i, n;
402 PyObject *v, *k;
403 PyObject *dict = PyDict_New();
404 if (!dict) return NULL;
405
406 n = PyList_Size(list);
407 for (i = 0; i < n; i++) {
408 v = PyLong_FromSsize_t(i);
409 if (!v) {
410 Py_DECREF(dict);
411 return NULL;
412 }
413 k = PyList_GET_ITEM(list, i);
414 k = _PyCode_ConstantKey(k);
415 if (k == NULL || PyDict_SetItem(dict, k, v) < 0) {
416 Py_XDECREF(k);
417 Py_DECREF(v);
418 Py_DECREF(dict);
419 return NULL;
420 }
421 Py_DECREF(k);
422 Py_DECREF(v);
423 }
424 return dict;
425 }
426
427 /* Return new dict containing names from src that match scope(s).
428
429 src is a symbol table dictionary. If the scope of a name matches
430 either scope_type or flag is set, insert it into the new dict. The
431 values are integers, starting at offset and increasing by one for
432 each key.
433 */
434
435 static PyObject *
dictbytype(PyObject * src,int scope_type,int flag,Py_ssize_t offset)436 dictbytype(PyObject *src, int scope_type, int flag, Py_ssize_t offset)
437 {
438 Py_ssize_t i = offset, scope, num_keys, key_i;
439 PyObject *k, *v, *dest = PyDict_New();
440 PyObject *sorted_keys;
441
442 assert(offset >= 0);
443 if (dest == NULL)
444 return NULL;
445
446 /* Sort the keys so that we have a deterministic order on the indexes
447 saved in the returned dictionary. These indexes are used as indexes
448 into the free and cell var storage. Therefore if they aren't
449 deterministic, then the generated bytecode is not deterministic.
450 */
451 sorted_keys = PyDict_Keys(src);
452 if (sorted_keys == NULL)
453 return NULL;
454 if (PyList_Sort(sorted_keys) != 0) {
455 Py_DECREF(sorted_keys);
456 return NULL;
457 }
458 num_keys = PyList_GET_SIZE(sorted_keys);
459
460 for (key_i = 0; key_i < num_keys; key_i++) {
461 /* XXX this should probably be a macro in symtable.h */
462 long vi;
463 k = PyList_GET_ITEM(sorted_keys, key_i);
464 v = PyDict_GetItem(src, k);
465 assert(PyLong_Check(v));
466 vi = PyLong_AS_LONG(v);
467 scope = (vi >> SCOPE_OFFSET) & SCOPE_MASK;
468
469 if (scope == scope_type || vi & flag) {
470 PyObject *tuple, *item = PyLong_FromSsize_t(i);
471 if (item == NULL) {
472 Py_DECREF(sorted_keys);
473 Py_DECREF(dest);
474 return NULL;
475 }
476 i++;
477 tuple = _PyCode_ConstantKey(k);
478 if (!tuple || PyDict_SetItem(dest, tuple, item) < 0) {
479 Py_DECREF(sorted_keys);
480 Py_DECREF(item);
481 Py_DECREF(dest);
482 Py_XDECREF(tuple);
483 return NULL;
484 }
485 Py_DECREF(item);
486 Py_DECREF(tuple);
487 }
488 }
489 Py_DECREF(sorted_keys);
490 return dest;
491 }
492
493 static void
compiler_unit_check(struct compiler_unit * u)494 compiler_unit_check(struct compiler_unit *u)
495 {
496 basicblock *block;
497 for (block = u->u_blocks; block != NULL; block = block->b_list) {
498 assert((uintptr_t)block != 0xcbcbcbcbU);
499 assert((uintptr_t)block != 0xfbfbfbfbU);
500 assert((uintptr_t)block != 0xdbdbdbdbU);
501 if (block->b_instr != NULL) {
502 assert(block->b_ialloc > 0);
503 assert(block->b_iused > 0);
504 assert(block->b_ialloc >= block->b_iused);
505 }
506 else {
507 assert (block->b_iused == 0);
508 assert (block->b_ialloc == 0);
509 }
510 }
511 }
512
513 static void
compiler_unit_free(struct compiler_unit * u)514 compiler_unit_free(struct compiler_unit *u)
515 {
516 basicblock *b, *next;
517
518 compiler_unit_check(u);
519 b = u->u_blocks;
520 while (b != NULL) {
521 if (b->b_instr)
522 PyObject_Free((void *)b->b_instr);
523 next = b->b_list;
524 PyObject_Free((void *)b);
525 b = next;
526 }
527 Py_CLEAR(u->u_ste);
528 Py_CLEAR(u->u_name);
529 Py_CLEAR(u->u_qualname);
530 Py_CLEAR(u->u_consts);
531 Py_CLEAR(u->u_names);
532 Py_CLEAR(u->u_varnames);
533 Py_CLEAR(u->u_freevars);
534 Py_CLEAR(u->u_cellvars);
535 Py_CLEAR(u->u_private);
536 PyObject_Free(u);
537 }
538
539 static int
compiler_enter_scope(struct compiler * c,identifier name,int scope_type,void * key,int lineno)540 compiler_enter_scope(struct compiler *c, identifier name,
541 int scope_type, void *key, int lineno)
542 {
543 struct compiler_unit *u;
544 basicblock *block;
545
546 u = (struct compiler_unit *)PyObject_Malloc(sizeof(
547 struct compiler_unit));
548 if (!u) {
549 PyErr_NoMemory();
550 return 0;
551 }
552 memset(u, 0, sizeof(struct compiler_unit));
553 u->u_scope_type = scope_type;
554 u->u_argcount = 0;
555 u->u_kwonlyargcount = 0;
556 u->u_ste = PySymtable_Lookup(c->c_st, key);
557 if (!u->u_ste) {
558 compiler_unit_free(u);
559 return 0;
560 }
561 Py_INCREF(name);
562 u->u_name = name;
563 u->u_varnames = list2dict(u->u_ste->ste_varnames);
564 u->u_cellvars = dictbytype(u->u_ste->ste_symbols, CELL, 0, 0);
565 if (!u->u_varnames || !u->u_cellvars) {
566 compiler_unit_free(u);
567 return 0;
568 }
569 if (u->u_ste->ste_needs_class_closure) {
570 /* Cook up an implicit __class__ cell. */
571 _Py_IDENTIFIER(__class__);
572 PyObject *tuple, *name;
573 int res;
574 assert(u->u_scope_type == COMPILER_SCOPE_CLASS);
575 assert(PyDict_GET_SIZE(u->u_cellvars) == 0);
576 name = _PyUnicode_FromId(&PyId___class__);
577 if (!name) {
578 compiler_unit_free(u);
579 return 0;
580 }
581 tuple = _PyCode_ConstantKey(name);
582 if (!tuple) {
583 compiler_unit_free(u);
584 return 0;
585 }
586 res = PyDict_SetItem(u->u_cellvars, tuple, _PyLong_Zero);
587 Py_DECREF(tuple);
588 if (res < 0) {
589 compiler_unit_free(u);
590 return 0;
591 }
592 }
593
594 u->u_freevars = dictbytype(u->u_ste->ste_symbols, FREE, DEF_FREE_CLASS,
595 PyDict_GET_SIZE(u->u_cellvars));
596 if (!u->u_freevars) {
597 compiler_unit_free(u);
598 return 0;
599 }
600
601 u->u_blocks = NULL;
602 u->u_nfblocks = 0;
603 u->u_firstlineno = lineno;
604 u->u_lineno = 0;
605 u->u_col_offset = 0;
606 u->u_lineno_set = 0;
607 u->u_consts = PyDict_New();
608 if (!u->u_consts) {
609 compiler_unit_free(u);
610 return 0;
611 }
612 u->u_names = PyDict_New();
613 if (!u->u_names) {
614 compiler_unit_free(u);
615 return 0;
616 }
617
618 u->u_private = NULL;
619
620 /* Push the old compiler_unit on the stack. */
621 if (c->u) {
622 PyObject *capsule = PyCapsule_New(c->u, CAPSULE_NAME, NULL);
623 if (!capsule || PyList_Append(c->c_stack, capsule) < 0) {
624 Py_XDECREF(capsule);
625 compiler_unit_free(u);
626 return 0;
627 }
628 Py_DECREF(capsule);
629 u->u_private = c->u->u_private;
630 Py_XINCREF(u->u_private);
631 }
632 c->u = u;
633
634 c->c_nestlevel++;
635
636 block = compiler_new_block(c);
637 if (block == NULL)
638 return 0;
639 c->u->u_curblock = block;
640
641 if (u->u_scope_type != COMPILER_SCOPE_MODULE) {
642 if (!compiler_set_qualname(c))
643 return 0;
644 }
645
646 return 1;
647 }
648
649 static void
compiler_exit_scope(struct compiler * c)650 compiler_exit_scope(struct compiler *c)
651 {
652 Py_ssize_t n;
653 PyObject *capsule;
654
655 c->c_nestlevel--;
656 compiler_unit_free(c->u);
657 /* Restore c->u to the parent unit. */
658 n = PyList_GET_SIZE(c->c_stack) - 1;
659 if (n >= 0) {
660 capsule = PyList_GET_ITEM(c->c_stack, n);
661 c->u = (struct compiler_unit *)PyCapsule_GetPointer(capsule, CAPSULE_NAME);
662 assert(c->u);
663 /* we are deleting from a list so this really shouldn't fail */
664 if (PySequence_DelItem(c->c_stack, n) < 0)
665 Py_FatalError("compiler_exit_scope()");
666 compiler_unit_check(c->u);
667 }
668 else
669 c->u = NULL;
670
671 }
672
673 static int
compiler_set_qualname(struct compiler * c)674 compiler_set_qualname(struct compiler *c)
675 {
676 _Py_static_string(dot, ".");
677 _Py_static_string(dot_locals, ".<locals>");
678 Py_ssize_t stack_size;
679 struct compiler_unit *u = c->u;
680 PyObject *name, *base, *dot_str, *dot_locals_str;
681
682 base = NULL;
683 stack_size = PyList_GET_SIZE(c->c_stack);
684 assert(stack_size >= 1);
685 if (stack_size > 1) {
686 int scope, force_global = 0;
687 struct compiler_unit *parent;
688 PyObject *mangled, *capsule;
689
690 capsule = PyList_GET_ITEM(c->c_stack, stack_size - 1);
691 parent = (struct compiler_unit *)PyCapsule_GetPointer(capsule, CAPSULE_NAME);
692 assert(parent);
693
694 if (u->u_scope_type == COMPILER_SCOPE_FUNCTION
695 || u->u_scope_type == COMPILER_SCOPE_ASYNC_FUNCTION
696 || u->u_scope_type == COMPILER_SCOPE_CLASS) {
697 assert(u->u_name);
698 mangled = _Py_Mangle(parent->u_private, u->u_name);
699 if (!mangled)
700 return 0;
701 scope = PyST_GetScope(parent->u_ste, mangled);
702 Py_DECREF(mangled);
703 assert(scope != GLOBAL_IMPLICIT);
704 if (scope == GLOBAL_EXPLICIT)
705 force_global = 1;
706 }
707
708 if (!force_global) {
709 if (parent->u_scope_type == COMPILER_SCOPE_FUNCTION
710 || parent->u_scope_type == COMPILER_SCOPE_ASYNC_FUNCTION
711 || parent->u_scope_type == COMPILER_SCOPE_LAMBDA) {
712 dot_locals_str = _PyUnicode_FromId(&dot_locals);
713 if (dot_locals_str == NULL)
714 return 0;
715 base = PyUnicode_Concat(parent->u_qualname, dot_locals_str);
716 if (base == NULL)
717 return 0;
718 }
719 else {
720 Py_INCREF(parent->u_qualname);
721 base = parent->u_qualname;
722 }
723 }
724 }
725
726 if (base != NULL) {
727 dot_str = _PyUnicode_FromId(&dot);
728 if (dot_str == NULL) {
729 Py_DECREF(base);
730 return 0;
731 }
732 name = PyUnicode_Concat(base, dot_str);
733 Py_DECREF(base);
734 if (name == NULL)
735 return 0;
736 PyUnicode_Append(&name, u->u_name);
737 if (name == NULL)
738 return 0;
739 }
740 else {
741 Py_INCREF(u->u_name);
742 name = u->u_name;
743 }
744 u->u_qualname = name;
745
746 return 1;
747 }
748
749
750 /* Allocate a new block and return a pointer to it.
751 Returns NULL on error.
752 */
753
754 static basicblock *
compiler_new_block(struct compiler * c)755 compiler_new_block(struct compiler *c)
756 {
757 basicblock *b;
758 struct compiler_unit *u;
759
760 u = c->u;
761 b = (basicblock *)PyObject_Malloc(sizeof(basicblock));
762 if (b == NULL) {
763 PyErr_NoMemory();
764 return NULL;
765 }
766 memset((void *)b, 0, sizeof(basicblock));
767 /* Extend the singly linked list of blocks with new block. */
768 b->b_list = u->u_blocks;
769 u->u_blocks = b;
770 return b;
771 }
772
773 static basicblock *
compiler_next_block(struct compiler * c)774 compiler_next_block(struct compiler *c)
775 {
776 basicblock *block = compiler_new_block(c);
777 if (block == NULL)
778 return NULL;
779 c->u->u_curblock->b_next = block;
780 c->u->u_curblock = block;
781 return block;
782 }
783
784 static basicblock *
compiler_use_next_block(struct compiler * c,basicblock * block)785 compiler_use_next_block(struct compiler *c, basicblock *block)
786 {
787 assert(block != NULL);
788 c->u->u_curblock->b_next = block;
789 c->u->u_curblock = block;
790 return block;
791 }
792
793 /* Returns the offset of the next instruction in the current block's
794 b_instr array. Resizes the b_instr as necessary.
795 Returns -1 on failure.
796 */
797
798 static int
compiler_next_instr(struct compiler * c,basicblock * b)799 compiler_next_instr(struct compiler *c, basicblock *b)
800 {
801 assert(b != NULL);
802 if (b->b_instr == NULL) {
803 b->b_instr = (struct instr *)PyObject_Malloc(
804 sizeof(struct instr) * DEFAULT_BLOCK_SIZE);
805 if (b->b_instr == NULL) {
806 PyErr_NoMemory();
807 return -1;
808 }
809 b->b_ialloc = DEFAULT_BLOCK_SIZE;
810 memset((char *)b->b_instr, 0,
811 sizeof(struct instr) * DEFAULT_BLOCK_SIZE);
812 }
813 else if (b->b_iused == b->b_ialloc) {
814 struct instr *tmp;
815 size_t oldsize, newsize;
816 oldsize = b->b_ialloc * sizeof(struct instr);
817 newsize = oldsize << 1;
818
819 if (oldsize > (SIZE_MAX >> 1)) {
820 PyErr_NoMemory();
821 return -1;
822 }
823
824 if (newsize == 0) {
825 PyErr_NoMemory();
826 return -1;
827 }
828 b->b_ialloc <<= 1;
829 tmp = (struct instr *)PyObject_Realloc(
830 (void *)b->b_instr, newsize);
831 if (tmp == NULL) {
832 PyErr_NoMemory();
833 return -1;
834 }
835 b->b_instr = tmp;
836 memset((char *)b->b_instr + oldsize, 0, newsize - oldsize);
837 }
838 return b->b_iused++;
839 }
840
841 /* Set the i_lineno member of the instruction at offset off if the
842 line number for the current expression/statement has not
843 already been set. If it has been set, the call has no effect.
844
845 The line number is reset in the following cases:
846 - when entering a new scope
847 - on each statement
848 - on each expression that start a new line
849 - before the "except" clause
850 - before the "for" and "while" expressions
851 */
852
853 static void
compiler_set_lineno(struct compiler * c,int off)854 compiler_set_lineno(struct compiler *c, int off)
855 {
856 basicblock *b;
857 if (c->u->u_lineno_set)
858 return;
859 c->u->u_lineno_set = 1;
860 b = c->u->u_curblock;
861 b->b_instr[off].i_lineno = c->u->u_lineno;
862 }
863
864 /* Return the stack effect of opcode with argument oparg.
865
866 Some opcodes have different stack effect when jump to the target and
867 when not jump. The 'jump' parameter specifies the case:
868
869 * 0 -- when not jump
870 * 1 -- when jump
871 * -1 -- maximal
872 */
873 /* XXX Make the stack effect of WITH_CLEANUP_START and
874 WITH_CLEANUP_FINISH deterministic. */
875 static int
stack_effect(int opcode,int oparg,int jump)876 stack_effect(int opcode, int oparg, int jump)
877 {
878 switch (opcode) {
879 /* Stack manipulation */
880 case POP_TOP:
881 return -1;
882 case ROT_TWO:
883 case ROT_THREE:
884 return 0;
885 case DUP_TOP:
886 return 1;
887 case DUP_TOP_TWO:
888 return 2;
889
890 /* Unary operators */
891 case UNARY_POSITIVE:
892 case UNARY_NEGATIVE:
893 case UNARY_NOT:
894 case UNARY_INVERT:
895 return 0;
896
897 case SET_ADD:
898 case LIST_APPEND:
899 return -1;
900 case MAP_ADD:
901 return -2;
902
903 /* Binary operators */
904 case BINARY_POWER:
905 case BINARY_MULTIPLY:
906 case BINARY_MATRIX_MULTIPLY:
907 case BINARY_MODULO:
908 case BINARY_ADD:
909 case BINARY_SUBTRACT:
910 case BINARY_SUBSCR:
911 case BINARY_FLOOR_DIVIDE:
912 case BINARY_TRUE_DIVIDE:
913 return -1;
914 case INPLACE_FLOOR_DIVIDE:
915 case INPLACE_TRUE_DIVIDE:
916 return -1;
917
918 case INPLACE_ADD:
919 case INPLACE_SUBTRACT:
920 case INPLACE_MULTIPLY:
921 case INPLACE_MATRIX_MULTIPLY:
922 case INPLACE_MODULO:
923 return -1;
924 case STORE_SUBSCR:
925 return -3;
926 case DELETE_SUBSCR:
927 return -2;
928
929 case BINARY_LSHIFT:
930 case BINARY_RSHIFT:
931 case BINARY_AND:
932 case BINARY_XOR:
933 case BINARY_OR:
934 return -1;
935 case INPLACE_POWER:
936 return -1;
937 case GET_ITER:
938 return 0;
939
940 case PRINT_EXPR:
941 return -1;
942 case LOAD_BUILD_CLASS:
943 return 1;
944 case INPLACE_LSHIFT:
945 case INPLACE_RSHIFT:
946 case INPLACE_AND:
947 case INPLACE_XOR:
948 case INPLACE_OR:
949 return -1;
950 case BREAK_LOOP:
951 return 0;
952 case SETUP_WITH:
953 /* 1 in the normal flow.
954 * Restore the stack position and push 6 values before jumping to
955 * the handler if an exception be raised. */
956 return jump ? 6 : 1;
957 case WITH_CLEANUP_START:
958 return 2; /* or 1, depending on TOS */
959 case WITH_CLEANUP_FINISH:
960 /* Pop a variable number of values pushed by WITH_CLEANUP_START
961 * + __exit__ or __aexit__. */
962 return -3;
963 case RETURN_VALUE:
964 return -1;
965 case IMPORT_STAR:
966 return -1;
967 case SETUP_ANNOTATIONS:
968 return 0;
969 case YIELD_VALUE:
970 return 0;
971 case YIELD_FROM:
972 return -1;
973 case POP_BLOCK:
974 return 0;
975 case POP_EXCEPT:
976 return -3;
977 case END_FINALLY:
978 /* Pop 6 values when an exception was raised. */
979 return -6;
980
981 case STORE_NAME:
982 return -1;
983 case DELETE_NAME:
984 return 0;
985 case UNPACK_SEQUENCE:
986 return oparg-1;
987 case UNPACK_EX:
988 return (oparg&0xFF) + (oparg>>8);
989 case FOR_ITER:
990 /* -1 at end of iterator, 1 if continue iterating. */
991 return jump > 0 ? -1 : 1;
992
993 case STORE_ATTR:
994 return -2;
995 case DELETE_ATTR:
996 return -1;
997 case STORE_GLOBAL:
998 return -1;
999 case DELETE_GLOBAL:
1000 return 0;
1001 case LOAD_CONST:
1002 return 1;
1003 case LOAD_NAME:
1004 return 1;
1005 case BUILD_TUPLE:
1006 case BUILD_LIST:
1007 case BUILD_SET:
1008 case BUILD_STRING:
1009 return 1-oparg;
1010 case BUILD_LIST_UNPACK:
1011 case BUILD_TUPLE_UNPACK:
1012 case BUILD_TUPLE_UNPACK_WITH_CALL:
1013 case BUILD_SET_UNPACK:
1014 case BUILD_MAP_UNPACK:
1015 case BUILD_MAP_UNPACK_WITH_CALL:
1016 return 1 - oparg;
1017 case BUILD_MAP:
1018 return 1 - 2*oparg;
1019 case BUILD_CONST_KEY_MAP:
1020 return -oparg;
1021 case LOAD_ATTR:
1022 return 0;
1023 case COMPARE_OP:
1024 return -1;
1025 case IMPORT_NAME:
1026 return -1;
1027 case IMPORT_FROM:
1028 return 1;
1029
1030 /* Jumps */
1031 case JUMP_FORWARD:
1032 case JUMP_ABSOLUTE:
1033 return 0;
1034
1035 case JUMP_IF_TRUE_OR_POP:
1036 case JUMP_IF_FALSE_OR_POP:
1037 return jump ? 0 : -1;
1038
1039 case POP_JUMP_IF_FALSE:
1040 case POP_JUMP_IF_TRUE:
1041 return -1;
1042
1043 case LOAD_GLOBAL:
1044 return 1;
1045
1046 case CONTINUE_LOOP:
1047 return 0;
1048 case SETUP_LOOP:
1049 return 0;
1050 case SETUP_EXCEPT:
1051 case SETUP_FINALLY:
1052 /* 0 in the normal flow.
1053 * Restore the stack position and push 6 values before jumping to
1054 * the handler if an exception be raised. */
1055 return jump ? 6 : 0;
1056
1057 case LOAD_FAST:
1058 return 1;
1059 case STORE_FAST:
1060 return -1;
1061 case DELETE_FAST:
1062 return 0;
1063
1064 case RAISE_VARARGS:
1065 return -oparg;
1066
1067 /* Functions and calls */
1068 case CALL_FUNCTION:
1069 return -oparg;
1070 case CALL_METHOD:
1071 return -oparg-1;
1072 case CALL_FUNCTION_KW:
1073 return -oparg-1;
1074 case CALL_FUNCTION_EX:
1075 return -1 - ((oparg & 0x01) != 0);
1076 case MAKE_FUNCTION:
1077 return -1 - ((oparg & 0x01) != 0) - ((oparg & 0x02) != 0) -
1078 ((oparg & 0x04) != 0) - ((oparg & 0x08) != 0);
1079 case BUILD_SLICE:
1080 if (oparg == 3)
1081 return -2;
1082 else
1083 return -1;
1084
1085 /* Closures */
1086 case LOAD_CLOSURE:
1087 return 1;
1088 case LOAD_DEREF:
1089 case LOAD_CLASSDEREF:
1090 return 1;
1091 case STORE_DEREF:
1092 return -1;
1093 case DELETE_DEREF:
1094 return 0;
1095
1096 /* Iterators and generators */
1097 case GET_AWAITABLE:
1098 return 0;
1099 case SETUP_ASYNC_WITH:
1100 /* 0 in the normal flow.
1101 * Restore the stack position to the position before the result
1102 * of __aenter__ and push 6 values before jumping to the handler
1103 * if an exception be raised. */
1104 return jump ? -1 + 6 : 0;
1105 case BEFORE_ASYNC_WITH:
1106 return 1;
1107 case GET_AITER:
1108 return 0;
1109 case GET_ANEXT:
1110 return 1;
1111 case GET_YIELD_FROM_ITER:
1112 return 0;
1113 case FORMAT_VALUE:
1114 /* If there's a fmt_spec on the stack, we go from 2->1,
1115 else 1->1. */
1116 return (oparg & FVS_MASK) == FVS_HAVE_SPEC ? -1 : 0;
1117 case LOAD_METHOD:
1118 return 1;
1119 default:
1120 return PY_INVALID_STACK_EFFECT;
1121 }
1122 return PY_INVALID_STACK_EFFECT; /* not reachable */
1123 }
1124
1125 int
PyCompile_OpcodeStackEffect(int opcode,int oparg)1126 PyCompile_OpcodeStackEffect(int opcode, int oparg)
1127 {
1128 return stack_effect(opcode, oparg, -1);
1129 }
1130
1131 /* Add an opcode with no argument.
1132 Returns 0 on failure, 1 on success.
1133 */
1134
1135 static int
compiler_addop(struct compiler * c,int opcode)1136 compiler_addop(struct compiler *c, int opcode)
1137 {
1138 basicblock *b;
1139 struct instr *i;
1140 int off;
1141 assert(!HAS_ARG(opcode));
1142 off = compiler_next_instr(c, c->u->u_curblock);
1143 if (off < 0)
1144 return 0;
1145 b = c->u->u_curblock;
1146 i = &b->b_instr[off];
1147 i->i_opcode = opcode;
1148 i->i_oparg = 0;
1149 if (opcode == RETURN_VALUE)
1150 b->b_return = 1;
1151 compiler_set_lineno(c, off);
1152 return 1;
1153 }
1154
1155 static Py_ssize_t
compiler_add_o(struct compiler * c,PyObject * dict,PyObject * o)1156 compiler_add_o(struct compiler *c, PyObject *dict, PyObject *o)
1157 {
1158 PyObject *t, *v;
1159 Py_ssize_t arg;
1160
1161 t = _PyCode_ConstantKey(o);
1162 if (t == NULL)
1163 return -1;
1164
1165 v = PyDict_GetItem(dict, t);
1166 if (!v) {
1167 if (PyErr_Occurred()) {
1168 Py_DECREF(t);
1169 return -1;
1170 }
1171 arg = PyDict_GET_SIZE(dict);
1172 v = PyLong_FromSsize_t(arg);
1173 if (!v) {
1174 Py_DECREF(t);
1175 return -1;
1176 }
1177 if (PyDict_SetItem(dict, t, v) < 0) {
1178 Py_DECREF(t);
1179 Py_DECREF(v);
1180 return -1;
1181 }
1182 Py_DECREF(v);
1183 }
1184 else
1185 arg = PyLong_AsLong(v);
1186 Py_DECREF(t);
1187 return arg;
1188 }
1189
1190 static int
compiler_addop_o(struct compiler * c,int opcode,PyObject * dict,PyObject * o)1191 compiler_addop_o(struct compiler *c, int opcode, PyObject *dict,
1192 PyObject *o)
1193 {
1194 Py_ssize_t arg = compiler_add_o(c, dict, o);
1195 if (arg < 0)
1196 return 0;
1197 return compiler_addop_i(c, opcode, arg);
1198 }
1199
1200 static int
compiler_addop_name(struct compiler * c,int opcode,PyObject * dict,PyObject * o)1201 compiler_addop_name(struct compiler *c, int opcode, PyObject *dict,
1202 PyObject *o)
1203 {
1204 Py_ssize_t arg;
1205 PyObject *mangled = _Py_Mangle(c->u->u_private, o);
1206 if (!mangled)
1207 return 0;
1208 arg = compiler_add_o(c, dict, mangled);
1209 Py_DECREF(mangled);
1210 if (arg < 0)
1211 return 0;
1212 return compiler_addop_i(c, opcode, arg);
1213 }
1214
1215 /* Add an opcode with an integer argument.
1216 Returns 0 on failure, 1 on success.
1217 */
1218
1219 static int
compiler_addop_i(struct compiler * c,int opcode,Py_ssize_t oparg)1220 compiler_addop_i(struct compiler *c, int opcode, Py_ssize_t oparg)
1221 {
1222 struct instr *i;
1223 int off;
1224
1225 /* oparg value is unsigned, but a signed C int is usually used to store
1226 it in the C code (like Python/ceval.c).
1227
1228 Limit to 32-bit signed C int (rather than INT_MAX) for portability.
1229
1230 The argument of a concrete bytecode instruction is limited to 8-bit.
1231 EXTENDED_ARG is used for 16, 24, and 32-bit arguments. */
1232 assert(HAS_ARG(opcode));
1233 assert(0 <= oparg && oparg <= 2147483647);
1234
1235 off = compiler_next_instr(c, c->u->u_curblock);
1236 if (off < 0)
1237 return 0;
1238 i = &c->u->u_curblock->b_instr[off];
1239 i->i_opcode = opcode;
1240 i->i_oparg = Py_SAFE_DOWNCAST(oparg, Py_ssize_t, int);
1241 compiler_set_lineno(c, off);
1242 return 1;
1243 }
1244
1245 static int
compiler_addop_j(struct compiler * c,int opcode,basicblock * b,int absolute)1246 compiler_addop_j(struct compiler *c, int opcode, basicblock *b, int absolute)
1247 {
1248 struct instr *i;
1249 int off;
1250
1251 assert(HAS_ARG(opcode));
1252 assert(b != NULL);
1253 off = compiler_next_instr(c, c->u->u_curblock);
1254 if (off < 0)
1255 return 0;
1256 i = &c->u->u_curblock->b_instr[off];
1257 i->i_opcode = opcode;
1258 i->i_target = b;
1259 if (absolute)
1260 i->i_jabs = 1;
1261 else
1262 i->i_jrel = 1;
1263 compiler_set_lineno(c, off);
1264 return 1;
1265 }
1266
1267 /* NEXT_BLOCK() creates an implicit jump from the current block
1268 to the new block.
1269
1270 The returns inside this macro make it impossible to decref objects
1271 created in the local function. Local objects should use the arena.
1272 */
1273 #define NEXT_BLOCK(C) { \
1274 if (compiler_next_block((C)) == NULL) \
1275 return 0; \
1276 }
1277
1278 #define ADDOP(C, OP) { \
1279 if (!compiler_addop((C), (OP))) \
1280 return 0; \
1281 }
1282
1283 #define ADDOP_IN_SCOPE(C, OP) { \
1284 if (!compiler_addop((C), (OP))) { \
1285 compiler_exit_scope(c); \
1286 return 0; \
1287 } \
1288 }
1289
1290 #define ADDOP_O(C, OP, O, TYPE) { \
1291 if (!compiler_addop_o((C), (OP), (C)->u->u_ ## TYPE, (O))) \
1292 return 0; \
1293 }
1294
1295 /* Same as ADDOP_O, but steals a reference. */
1296 #define ADDOP_N(C, OP, O, TYPE) { \
1297 if (!compiler_addop_o((C), (OP), (C)->u->u_ ## TYPE, (O))) { \
1298 Py_DECREF((O)); \
1299 return 0; \
1300 } \
1301 Py_DECREF((O)); \
1302 }
1303
1304 #define ADDOP_NAME(C, OP, O, TYPE) { \
1305 if (!compiler_addop_name((C), (OP), (C)->u->u_ ## TYPE, (O))) \
1306 return 0; \
1307 }
1308
1309 #define ADDOP_I(C, OP, O) { \
1310 if (!compiler_addop_i((C), (OP), (O))) \
1311 return 0; \
1312 }
1313
1314 #define ADDOP_JABS(C, OP, O) { \
1315 if (!compiler_addop_j((C), (OP), (O), 1)) \
1316 return 0; \
1317 }
1318
1319 #define ADDOP_JREL(C, OP, O) { \
1320 if (!compiler_addop_j((C), (OP), (O), 0)) \
1321 return 0; \
1322 }
1323
1324 /* VISIT and VISIT_SEQ takes an ASDL type as their second argument. They use
1325 the ASDL name to synthesize the name of the C type and the visit function.
1326 */
1327
1328 #define VISIT(C, TYPE, V) {\
1329 if (!compiler_visit_ ## TYPE((C), (V))) \
1330 return 0; \
1331 }
1332
1333 #define VISIT_IN_SCOPE(C, TYPE, V) {\
1334 if (!compiler_visit_ ## TYPE((C), (V))) { \
1335 compiler_exit_scope(c); \
1336 return 0; \
1337 } \
1338 }
1339
1340 #define VISIT_SLICE(C, V, CTX) {\
1341 if (!compiler_visit_slice((C), (V), (CTX))) \
1342 return 0; \
1343 }
1344
1345 #define VISIT_SEQ(C, TYPE, SEQ) { \
1346 int _i; \
1347 asdl_seq *seq = (SEQ); /* avoid variable capture */ \
1348 for (_i = 0; _i < asdl_seq_LEN(seq); _i++) { \
1349 TYPE ## _ty elt = (TYPE ## _ty)asdl_seq_GET(seq, _i); \
1350 if (!compiler_visit_ ## TYPE((C), elt)) \
1351 return 0; \
1352 } \
1353 }
1354
1355 #define VISIT_SEQ_IN_SCOPE(C, TYPE, SEQ) { \
1356 int _i; \
1357 asdl_seq *seq = (SEQ); /* avoid variable capture */ \
1358 for (_i = 0; _i < asdl_seq_LEN(seq); _i++) { \
1359 TYPE ## _ty elt = (TYPE ## _ty)asdl_seq_GET(seq, _i); \
1360 if (!compiler_visit_ ## TYPE((C), elt)) { \
1361 compiler_exit_scope(c); \
1362 return 0; \
1363 } \
1364 } \
1365 }
1366
1367 static int
compiler_isdocstring(stmt_ty s)1368 compiler_isdocstring(stmt_ty s)
1369 {
1370 if (s->kind != Expr_kind)
1371 return 0;
1372 if (s->v.Expr.value->kind == Str_kind)
1373 return 1;
1374 if (s->v.Expr.value->kind == Constant_kind)
1375 return PyUnicode_CheckExact(s->v.Expr.value->v.Constant.value);
1376 return 0;
1377 }
1378
1379 static int
is_const(expr_ty e)1380 is_const(expr_ty e)
1381 {
1382 switch (e->kind) {
1383 case Constant_kind:
1384 case Num_kind:
1385 case Str_kind:
1386 case Bytes_kind:
1387 case Ellipsis_kind:
1388 case NameConstant_kind:
1389 return 1;
1390 default:
1391 return 0;
1392 }
1393 }
1394
1395 static PyObject *
get_const_value(expr_ty e)1396 get_const_value(expr_ty e)
1397 {
1398 switch (e->kind) {
1399 case Constant_kind:
1400 return e->v.Constant.value;
1401 case Num_kind:
1402 return e->v.Num.n;
1403 case Str_kind:
1404 return e->v.Str.s;
1405 case Bytes_kind:
1406 return e->v.Bytes.s;
1407 case Ellipsis_kind:
1408 return Py_Ellipsis;
1409 case NameConstant_kind:
1410 return e->v.NameConstant.value;
1411 default:
1412 Py_UNREACHABLE();
1413 }
1414 }
1415
1416 /* Search if variable annotations are present statically in a block. */
1417
1418 static int
find_ann(asdl_seq * stmts)1419 find_ann(asdl_seq *stmts)
1420 {
1421 int i, j, res = 0;
1422 stmt_ty st;
1423
1424 for (i = 0; i < asdl_seq_LEN(stmts); i++) {
1425 st = (stmt_ty)asdl_seq_GET(stmts, i);
1426 switch (st->kind) {
1427 case AnnAssign_kind:
1428 return 1;
1429 case For_kind:
1430 res = find_ann(st->v.For.body) ||
1431 find_ann(st->v.For.orelse);
1432 break;
1433 case AsyncFor_kind:
1434 res = find_ann(st->v.AsyncFor.body) ||
1435 find_ann(st->v.AsyncFor.orelse);
1436 break;
1437 case While_kind:
1438 res = find_ann(st->v.While.body) ||
1439 find_ann(st->v.While.orelse);
1440 break;
1441 case If_kind:
1442 res = find_ann(st->v.If.body) ||
1443 find_ann(st->v.If.orelse);
1444 break;
1445 case With_kind:
1446 res = find_ann(st->v.With.body);
1447 break;
1448 case AsyncWith_kind:
1449 res = find_ann(st->v.AsyncWith.body);
1450 break;
1451 case Try_kind:
1452 for (j = 0; j < asdl_seq_LEN(st->v.Try.handlers); j++) {
1453 excepthandler_ty handler = (excepthandler_ty)asdl_seq_GET(
1454 st->v.Try.handlers, j);
1455 if (find_ann(handler->v.ExceptHandler.body)) {
1456 return 1;
1457 }
1458 }
1459 res = find_ann(st->v.Try.body) ||
1460 find_ann(st->v.Try.finalbody) ||
1461 find_ann(st->v.Try.orelse);
1462 break;
1463 default:
1464 res = 0;
1465 }
1466 if (res) {
1467 break;
1468 }
1469 }
1470 return res;
1471 }
1472
1473 /* Compile a sequence of statements, checking for a docstring
1474 and for annotations. */
1475
1476 static int
compiler_body(struct compiler * c,asdl_seq * stmts)1477 compiler_body(struct compiler *c, asdl_seq *stmts)
1478 {
1479 int i = 0;
1480 stmt_ty st;
1481
1482 /* Set current line number to the line number of first statement.
1483 This way line number for SETUP_ANNOTATIONS will always
1484 coincide with the line number of first "real" statement in module.
1485 If body is empy, then lineno will be set later in assemble. */
1486 if (c->u->u_scope_type == COMPILER_SCOPE_MODULE &&
1487 !c->u->u_lineno && asdl_seq_LEN(stmts)) {
1488 st = (stmt_ty)asdl_seq_GET(stmts, 0);
1489 c->u->u_lineno = st->lineno;
1490 }
1491 /* Every annotated class and module should have __annotations__. */
1492 if (find_ann(stmts)) {
1493 ADDOP(c, SETUP_ANNOTATIONS);
1494 }
1495 if (!asdl_seq_LEN(stmts))
1496 return 1;
1497 st = (stmt_ty)asdl_seq_GET(stmts, 0);
1498 /* if not -OO mode, set docstring */
1499 if (compiler_isdocstring(st) && c->c_optimize < 2) {
1500 /* don't generate docstrings if -OO */
1501 i = 1;
1502 VISIT(c, expr, st->v.Expr.value);
1503 if (!compiler_nameop(c, __doc__, Store))
1504 return 0;
1505 }
1506 for (; i < asdl_seq_LEN(stmts); i++)
1507 VISIT(c, stmt, (stmt_ty)asdl_seq_GET(stmts, i));
1508 return 1;
1509 }
1510
1511 static PyCodeObject *
compiler_mod(struct compiler * c,mod_ty mod)1512 compiler_mod(struct compiler *c, mod_ty mod)
1513 {
1514 PyCodeObject *co;
1515 int addNone = 1;
1516 static PyObject *module;
1517 if (!module) {
1518 module = PyUnicode_InternFromString("<module>");
1519 if (!module)
1520 return NULL;
1521 }
1522 /* Use 0 for firstlineno initially, will fixup in assemble(). */
1523 if (!compiler_enter_scope(c, module, COMPILER_SCOPE_MODULE, mod, 0))
1524 return NULL;
1525 switch (mod->kind) {
1526 case Module_kind:
1527 if (!compiler_body(c, mod->v.Module.body)) {
1528 compiler_exit_scope(c);
1529 return 0;
1530 }
1531 break;
1532 case Interactive_kind:
1533 if (find_ann(mod->v.Interactive.body)) {
1534 ADDOP(c, SETUP_ANNOTATIONS);
1535 }
1536 c->c_interactive = 1;
1537 VISIT_SEQ_IN_SCOPE(c, stmt,
1538 mod->v.Interactive.body);
1539 break;
1540 case Expression_kind:
1541 VISIT_IN_SCOPE(c, expr, mod->v.Expression.body);
1542 addNone = 0;
1543 break;
1544 case Suite_kind:
1545 PyErr_SetString(PyExc_SystemError,
1546 "suite should not be possible");
1547 return 0;
1548 default:
1549 PyErr_Format(PyExc_SystemError,
1550 "module kind %d should not be possible",
1551 mod->kind);
1552 return 0;
1553 }
1554 co = assemble(c, addNone);
1555 compiler_exit_scope(c);
1556 return co;
1557 }
1558
1559 /* The test for LOCAL must come before the test for FREE in order to
1560 handle classes where name is both local and free. The local var is
1561 a method and the free var is a free var referenced within a method.
1562 */
1563
1564 static int
get_ref_type(struct compiler * c,PyObject * name)1565 get_ref_type(struct compiler *c, PyObject *name)
1566 {
1567 int scope;
1568 if (c->u->u_scope_type == COMPILER_SCOPE_CLASS &&
1569 _PyUnicode_EqualToASCIIString(name, "__class__"))
1570 return CELL;
1571 scope = PyST_GetScope(c->u->u_ste, name);
1572 if (scope == 0) {
1573 char buf[350];
1574 PyOS_snprintf(buf, sizeof(buf),
1575 "unknown scope for %.100s in %.100s(%s)\n"
1576 "symbols: %s\nlocals: %s\nglobals: %s",
1577 PyUnicode_AsUTF8(name),
1578 PyUnicode_AsUTF8(c->u->u_name),
1579 PyUnicode_AsUTF8(PyObject_Repr(c->u->u_ste->ste_id)),
1580 PyUnicode_AsUTF8(PyObject_Repr(c->u->u_ste->ste_symbols)),
1581 PyUnicode_AsUTF8(PyObject_Repr(c->u->u_varnames)),
1582 PyUnicode_AsUTF8(PyObject_Repr(c->u->u_names))
1583 );
1584 Py_FatalError(buf);
1585 }
1586
1587 return scope;
1588 }
1589
1590 static int
compiler_lookup_arg(PyObject * dict,PyObject * name)1591 compiler_lookup_arg(PyObject *dict, PyObject *name)
1592 {
1593 PyObject *k, *v;
1594 k = _PyCode_ConstantKey(name);
1595 if (k == NULL)
1596 return -1;
1597 v = PyDict_GetItem(dict, k);
1598 Py_DECREF(k);
1599 if (v == NULL)
1600 return -1;
1601 return PyLong_AS_LONG(v);
1602 }
1603
1604 static int
compiler_make_closure(struct compiler * c,PyCodeObject * co,Py_ssize_t flags,PyObject * qualname)1605 compiler_make_closure(struct compiler *c, PyCodeObject *co, Py_ssize_t flags, PyObject *qualname)
1606 {
1607 Py_ssize_t i, free = PyCode_GetNumFree(co);
1608 if (qualname == NULL)
1609 qualname = co->co_name;
1610
1611 if (free) {
1612 for (i = 0; i < free; ++i) {
1613 /* Bypass com_addop_varname because it will generate
1614 LOAD_DEREF but LOAD_CLOSURE is needed.
1615 */
1616 PyObject *name = PyTuple_GET_ITEM(co->co_freevars, i);
1617 int arg, reftype;
1618
1619 /* Special case: If a class contains a method with a
1620 free variable that has the same name as a method,
1621 the name will be considered free *and* local in the
1622 class. It should be handled by the closure, as
1623 well as by the normal name loookup logic.
1624 */
1625 reftype = get_ref_type(c, name);
1626 if (reftype == CELL)
1627 arg = compiler_lookup_arg(c->u->u_cellvars, name);
1628 else /* (reftype == FREE) */
1629 arg = compiler_lookup_arg(c->u->u_freevars, name);
1630 if (arg == -1) {
1631 fprintf(stderr,
1632 "lookup %s in %s %d %d\n"
1633 "freevars of %s: %s\n",
1634 PyUnicode_AsUTF8(PyObject_Repr(name)),
1635 PyUnicode_AsUTF8(c->u->u_name),
1636 reftype, arg,
1637 PyUnicode_AsUTF8(co->co_name),
1638 PyUnicode_AsUTF8(PyObject_Repr(co->co_freevars)));
1639 Py_FatalError("compiler_make_closure()");
1640 }
1641 ADDOP_I(c, LOAD_CLOSURE, arg);
1642 }
1643 flags |= 0x08;
1644 ADDOP_I(c, BUILD_TUPLE, free);
1645 }
1646 ADDOP_O(c, LOAD_CONST, (PyObject*)co, consts);
1647 ADDOP_O(c, LOAD_CONST, qualname, consts);
1648 ADDOP_I(c, MAKE_FUNCTION, flags);
1649 return 1;
1650 }
1651
1652 static int
compiler_decorators(struct compiler * c,asdl_seq * decos)1653 compiler_decorators(struct compiler *c, asdl_seq* decos)
1654 {
1655 int i;
1656
1657 if (!decos)
1658 return 1;
1659
1660 for (i = 0; i < asdl_seq_LEN(decos); i++) {
1661 VISIT(c, expr, (expr_ty)asdl_seq_GET(decos, i));
1662 }
1663 return 1;
1664 }
1665
1666 static int
compiler_visit_kwonlydefaults(struct compiler * c,asdl_seq * kwonlyargs,asdl_seq * kw_defaults)1667 compiler_visit_kwonlydefaults(struct compiler *c, asdl_seq *kwonlyargs,
1668 asdl_seq *kw_defaults)
1669 {
1670 /* Push a dict of keyword-only default values.
1671
1672 Return 0 on error, -1 if no dict pushed, 1 if a dict is pushed.
1673 */
1674 int i;
1675 PyObject *keys = NULL;
1676
1677 for (i = 0; i < asdl_seq_LEN(kwonlyargs); i++) {
1678 arg_ty arg = asdl_seq_GET(kwonlyargs, i);
1679 expr_ty default_ = asdl_seq_GET(kw_defaults, i);
1680 if (default_) {
1681 PyObject *mangled = _Py_Mangle(c->u->u_private, arg->arg);
1682 if (!mangled) {
1683 goto error;
1684 }
1685 if (keys == NULL) {
1686 keys = PyList_New(1);
1687 if (keys == NULL) {
1688 Py_DECREF(mangled);
1689 return 0;
1690 }
1691 PyList_SET_ITEM(keys, 0, mangled);
1692 }
1693 else {
1694 int res = PyList_Append(keys, mangled);
1695 Py_DECREF(mangled);
1696 if (res == -1) {
1697 goto error;
1698 }
1699 }
1700 if (!compiler_visit_expr(c, default_)) {
1701 goto error;
1702 }
1703 }
1704 }
1705 if (keys != NULL) {
1706 Py_ssize_t default_count = PyList_GET_SIZE(keys);
1707 PyObject *keys_tuple = PyList_AsTuple(keys);
1708 Py_DECREF(keys);
1709 if (keys_tuple == NULL) {
1710 return 0;
1711 }
1712 ADDOP_N(c, LOAD_CONST, keys_tuple, consts);
1713 ADDOP_I(c, BUILD_CONST_KEY_MAP, default_count);
1714 assert(default_count > 0);
1715 return 1;
1716 }
1717 else {
1718 return -1;
1719 }
1720
1721 error:
1722 Py_XDECREF(keys);
1723 return 0;
1724 }
1725
1726 static int
compiler_visit_annexpr(struct compiler * c,expr_ty annotation)1727 compiler_visit_annexpr(struct compiler *c, expr_ty annotation)
1728 {
1729 PyObject *ann_as_str;
1730 ann_as_str = _PyAST_ExprAsUnicode(annotation);
1731 if (!ann_as_str) {
1732 return 0;
1733 }
1734 ADDOP_N(c, LOAD_CONST, ann_as_str, consts);
1735 return 1;
1736 }
1737
1738 static int
compiler_visit_argannotation(struct compiler * c,identifier id,expr_ty annotation,PyObject * names)1739 compiler_visit_argannotation(struct compiler *c, identifier id,
1740 expr_ty annotation, PyObject *names)
1741 {
1742 if (annotation) {
1743 PyObject *mangled;
1744 if (c->c_future->ff_features & CO_FUTURE_ANNOTATIONS) {
1745 VISIT(c, annexpr, annotation)
1746 }
1747 else {
1748 VISIT(c, expr, annotation);
1749 }
1750 mangled = _Py_Mangle(c->u->u_private, id);
1751 if (!mangled)
1752 return 0;
1753 if (PyList_Append(names, mangled) < 0) {
1754 Py_DECREF(mangled);
1755 return 0;
1756 }
1757 Py_DECREF(mangled);
1758 }
1759 return 1;
1760 }
1761
1762 static int
compiler_visit_argannotations(struct compiler * c,asdl_seq * args,PyObject * names)1763 compiler_visit_argannotations(struct compiler *c, asdl_seq* args,
1764 PyObject *names)
1765 {
1766 int i;
1767 for (i = 0; i < asdl_seq_LEN(args); i++) {
1768 arg_ty arg = (arg_ty)asdl_seq_GET(args, i);
1769 if (!compiler_visit_argannotation(
1770 c,
1771 arg->arg,
1772 arg->annotation,
1773 names))
1774 return 0;
1775 }
1776 return 1;
1777 }
1778
1779 static int
compiler_visit_annotations(struct compiler * c,arguments_ty args,expr_ty returns)1780 compiler_visit_annotations(struct compiler *c, arguments_ty args,
1781 expr_ty returns)
1782 {
1783 /* Push arg annotation dict.
1784 The expressions are evaluated out-of-order wrt the source code.
1785
1786 Return 0 on error, -1 if no dict pushed, 1 if a dict is pushed.
1787 */
1788 static identifier return_str;
1789 PyObject *names;
1790 Py_ssize_t len;
1791 names = PyList_New(0);
1792 if (!names)
1793 return 0;
1794
1795 if (!compiler_visit_argannotations(c, args->args, names))
1796 goto error;
1797 if (args->vararg && args->vararg->annotation &&
1798 !compiler_visit_argannotation(c, args->vararg->arg,
1799 args->vararg->annotation, names))
1800 goto error;
1801 if (!compiler_visit_argannotations(c, args->kwonlyargs, names))
1802 goto error;
1803 if (args->kwarg && args->kwarg->annotation &&
1804 !compiler_visit_argannotation(c, args->kwarg->arg,
1805 args->kwarg->annotation, names))
1806 goto error;
1807
1808 if (!return_str) {
1809 return_str = PyUnicode_InternFromString("return");
1810 if (!return_str)
1811 goto error;
1812 }
1813 if (!compiler_visit_argannotation(c, return_str, returns, names)) {
1814 goto error;
1815 }
1816
1817 len = PyList_GET_SIZE(names);
1818 if (len) {
1819 PyObject *keytuple = PyList_AsTuple(names);
1820 Py_DECREF(names);
1821 if (keytuple == NULL) {
1822 return 0;
1823 }
1824 ADDOP_N(c, LOAD_CONST, keytuple, consts);
1825 ADDOP_I(c, BUILD_CONST_KEY_MAP, len);
1826 return 1;
1827 }
1828 else {
1829 Py_DECREF(names);
1830 return -1;
1831 }
1832
1833 error:
1834 Py_DECREF(names);
1835 return 0;
1836 }
1837
1838 static int
compiler_visit_defaults(struct compiler * c,arguments_ty args)1839 compiler_visit_defaults(struct compiler *c, arguments_ty args)
1840 {
1841 VISIT_SEQ(c, expr, args->defaults);
1842 ADDOP_I(c, BUILD_TUPLE, asdl_seq_LEN(args->defaults));
1843 return 1;
1844 }
1845
1846 static Py_ssize_t
compiler_default_arguments(struct compiler * c,arguments_ty args)1847 compiler_default_arguments(struct compiler *c, arguments_ty args)
1848 {
1849 Py_ssize_t funcflags = 0;
1850 if (args->defaults && asdl_seq_LEN(args->defaults) > 0) {
1851 if (!compiler_visit_defaults(c, args))
1852 return -1;
1853 funcflags |= 0x01;
1854 }
1855 if (args->kwonlyargs) {
1856 int res = compiler_visit_kwonlydefaults(c, args->kwonlyargs,
1857 args->kw_defaults);
1858 if (res == 0) {
1859 return -1;
1860 }
1861 else if (res > 0) {
1862 funcflags |= 0x02;
1863 }
1864 }
1865 return funcflags;
1866 }
1867
1868 static int
compiler_function(struct compiler * c,stmt_ty s,int is_async)1869 compiler_function(struct compiler *c, stmt_ty s, int is_async)
1870 {
1871 PyCodeObject *co;
1872 PyObject *qualname, *first_const = Py_None;
1873 arguments_ty args;
1874 expr_ty returns;
1875 identifier name;
1876 asdl_seq* decos;
1877 asdl_seq *body;
1878 stmt_ty st;
1879 Py_ssize_t i, funcflags;
1880 int docstring;
1881 int annotations;
1882 int scope_type;
1883
1884 if (is_async) {
1885 assert(s->kind == AsyncFunctionDef_kind);
1886
1887 args = s->v.AsyncFunctionDef.args;
1888 returns = s->v.AsyncFunctionDef.returns;
1889 decos = s->v.AsyncFunctionDef.decorator_list;
1890 name = s->v.AsyncFunctionDef.name;
1891 body = s->v.AsyncFunctionDef.body;
1892
1893 scope_type = COMPILER_SCOPE_ASYNC_FUNCTION;
1894 } else {
1895 assert(s->kind == FunctionDef_kind);
1896
1897 args = s->v.FunctionDef.args;
1898 returns = s->v.FunctionDef.returns;
1899 decos = s->v.FunctionDef.decorator_list;
1900 name = s->v.FunctionDef.name;
1901 body = s->v.FunctionDef.body;
1902
1903 scope_type = COMPILER_SCOPE_FUNCTION;
1904 }
1905
1906 if (!compiler_decorators(c, decos))
1907 return 0;
1908
1909 funcflags = compiler_default_arguments(c, args);
1910 if (funcflags == -1) {
1911 return 0;
1912 }
1913
1914 annotations = compiler_visit_annotations(c, args, returns);
1915 if (annotations == 0) {
1916 return 0;
1917 }
1918 else if (annotations > 0) {
1919 funcflags |= 0x04;
1920 }
1921
1922 if (!compiler_enter_scope(c, name, scope_type, (void *)s, s->lineno)) {
1923 return 0;
1924 }
1925
1926 st = (stmt_ty)asdl_seq_GET(body, 0);
1927 docstring = compiler_isdocstring(st);
1928 if (docstring && c->c_optimize < 2) {
1929 if (st->v.Expr.value->kind == Constant_kind)
1930 first_const = st->v.Expr.value->v.Constant.value;
1931 else
1932 first_const = st->v.Expr.value->v.Str.s;
1933 }
1934 if (compiler_add_o(c, c->u->u_consts, first_const) < 0) {
1935 compiler_exit_scope(c);
1936 return 0;
1937 }
1938
1939 c->u->u_argcount = asdl_seq_LEN(args->args);
1940 c->u->u_kwonlyargcount = asdl_seq_LEN(args->kwonlyargs);
1941 VISIT_SEQ_IN_SCOPE(c, stmt, body);
1942 co = assemble(c, 1);
1943 qualname = c->u->u_qualname;
1944 Py_INCREF(qualname);
1945 compiler_exit_scope(c);
1946 if (co == NULL) {
1947 Py_XDECREF(qualname);
1948 Py_XDECREF(co);
1949 return 0;
1950 }
1951
1952 compiler_make_closure(c, co, funcflags, qualname);
1953 Py_DECREF(qualname);
1954 Py_DECREF(co);
1955
1956 /* decorators */
1957 for (i = 0; i < asdl_seq_LEN(decos); i++) {
1958 ADDOP_I(c, CALL_FUNCTION, 1);
1959 }
1960
1961 return compiler_nameop(c, name, Store);
1962 }
1963
1964 static int
compiler_class(struct compiler * c,stmt_ty s)1965 compiler_class(struct compiler *c, stmt_ty s)
1966 {
1967 PyCodeObject *co;
1968 PyObject *str;
1969 int i;
1970 asdl_seq* decos = s->v.ClassDef.decorator_list;
1971
1972 if (!compiler_decorators(c, decos))
1973 return 0;
1974
1975 /* ultimately generate code for:
1976 <name> = __build_class__(<func>, <name>, *<bases>, **<keywords>)
1977 where:
1978 <func> is a function/closure created from the class body;
1979 it has a single argument (__locals__) where the dict
1980 (or MutableSequence) representing the locals is passed
1981 <name> is the class name
1982 <bases> is the positional arguments and *varargs argument
1983 <keywords> is the keyword arguments and **kwds argument
1984 This borrows from compiler_call.
1985 */
1986
1987 /* 1. compile the class body into a code object */
1988 if (!compiler_enter_scope(c, s->v.ClassDef.name,
1989 COMPILER_SCOPE_CLASS, (void *)s, s->lineno))
1990 return 0;
1991 /* this block represents what we do in the new scope */
1992 {
1993 /* use the class name for name mangling */
1994 Py_INCREF(s->v.ClassDef.name);
1995 Py_XSETREF(c->u->u_private, s->v.ClassDef.name);
1996 /* load (global) __name__ ... */
1997 str = PyUnicode_InternFromString("__name__");
1998 if (!str || !compiler_nameop(c, str, Load)) {
1999 Py_XDECREF(str);
2000 compiler_exit_scope(c);
2001 return 0;
2002 }
2003 Py_DECREF(str);
2004 /* ... and store it as __module__ */
2005 str = PyUnicode_InternFromString("__module__");
2006 if (!str || !compiler_nameop(c, str, Store)) {
2007 Py_XDECREF(str);
2008 compiler_exit_scope(c);
2009 return 0;
2010 }
2011 Py_DECREF(str);
2012 assert(c->u->u_qualname);
2013 ADDOP_O(c, LOAD_CONST, c->u->u_qualname, consts);
2014 str = PyUnicode_InternFromString("__qualname__");
2015 if (!str || !compiler_nameop(c, str, Store)) {
2016 Py_XDECREF(str);
2017 compiler_exit_scope(c);
2018 return 0;
2019 }
2020 Py_DECREF(str);
2021 /* compile the body proper */
2022 if (!compiler_body(c, s->v.ClassDef.body)) {
2023 compiler_exit_scope(c);
2024 return 0;
2025 }
2026 /* Return __classcell__ if it is referenced, otherwise return None */
2027 if (c->u->u_ste->ste_needs_class_closure) {
2028 /* Store __classcell__ into class namespace & return it */
2029 str = PyUnicode_InternFromString("__class__");
2030 if (str == NULL) {
2031 compiler_exit_scope(c);
2032 return 0;
2033 }
2034 i = compiler_lookup_arg(c->u->u_cellvars, str);
2035 Py_DECREF(str);
2036 if (i < 0) {
2037 compiler_exit_scope(c);
2038 return 0;
2039 }
2040 assert(i == 0);
2041
2042 ADDOP_I(c, LOAD_CLOSURE, i);
2043 ADDOP(c, DUP_TOP);
2044 str = PyUnicode_InternFromString("__classcell__");
2045 if (!str || !compiler_nameop(c, str, Store)) {
2046 Py_XDECREF(str);
2047 compiler_exit_scope(c);
2048 return 0;
2049 }
2050 Py_DECREF(str);
2051 }
2052 else {
2053 /* No methods referenced __class__, so just return None */
2054 assert(PyDict_GET_SIZE(c->u->u_cellvars) == 0);
2055 ADDOP_O(c, LOAD_CONST, Py_None, consts);
2056 }
2057 ADDOP_IN_SCOPE(c, RETURN_VALUE);
2058 /* create the code object */
2059 co = assemble(c, 1);
2060 }
2061 /* leave the new scope */
2062 compiler_exit_scope(c);
2063 if (co == NULL)
2064 return 0;
2065
2066 /* 2. load the 'build_class' function */
2067 ADDOP(c, LOAD_BUILD_CLASS);
2068
2069 /* 3. load a function (or closure) made from the code object */
2070 compiler_make_closure(c, co, 0, NULL);
2071 Py_DECREF(co);
2072
2073 /* 4. load class name */
2074 ADDOP_O(c, LOAD_CONST, s->v.ClassDef.name, consts);
2075
2076 /* 5. generate the rest of the code for the call */
2077 if (!compiler_call_helper(c, 2,
2078 s->v.ClassDef.bases,
2079 s->v.ClassDef.keywords))
2080 return 0;
2081
2082 /* 6. apply decorators */
2083 for (i = 0; i < asdl_seq_LEN(decos); i++) {
2084 ADDOP_I(c, CALL_FUNCTION, 1);
2085 }
2086
2087 /* 7. store into <name> */
2088 if (!compiler_nameop(c, s->v.ClassDef.name, Store))
2089 return 0;
2090 return 1;
2091 }
2092
2093 static int
cmpop(cmpop_ty op)2094 cmpop(cmpop_ty op)
2095 {
2096 switch (op) {
2097 case Eq:
2098 return PyCmp_EQ;
2099 case NotEq:
2100 return PyCmp_NE;
2101 case Lt:
2102 return PyCmp_LT;
2103 case LtE:
2104 return PyCmp_LE;
2105 case Gt:
2106 return PyCmp_GT;
2107 case GtE:
2108 return PyCmp_GE;
2109 case Is:
2110 return PyCmp_IS;
2111 case IsNot:
2112 return PyCmp_IS_NOT;
2113 case In:
2114 return PyCmp_IN;
2115 case NotIn:
2116 return PyCmp_NOT_IN;
2117 default:
2118 return PyCmp_BAD;
2119 }
2120 }
2121
2122 static int
compiler_jump_if(struct compiler * c,expr_ty e,basicblock * next,int cond)2123 compiler_jump_if(struct compiler *c, expr_ty e, basicblock *next, int cond)
2124 {
2125 switch (e->kind) {
2126 case UnaryOp_kind:
2127 if (e->v.UnaryOp.op == Not)
2128 return compiler_jump_if(c, e->v.UnaryOp.operand, next, !cond);
2129 /* fallback to general implementation */
2130 break;
2131 case BoolOp_kind: {
2132 asdl_seq *s = e->v.BoolOp.values;
2133 Py_ssize_t i, n = asdl_seq_LEN(s) - 1;
2134 assert(n >= 0);
2135 int cond2 = e->v.BoolOp.op == Or;
2136 basicblock *next2 = next;
2137 if (!cond2 != !cond) {
2138 next2 = compiler_new_block(c);
2139 if (next2 == NULL)
2140 return 0;
2141 }
2142 for (i = 0; i < n; ++i) {
2143 if (!compiler_jump_if(c, (expr_ty)asdl_seq_GET(s, i), next2, cond2))
2144 return 0;
2145 }
2146 if (!compiler_jump_if(c, (expr_ty)asdl_seq_GET(s, n), next, cond))
2147 return 0;
2148 if (next2 != next)
2149 compiler_use_next_block(c, next2);
2150 return 1;
2151 }
2152 case IfExp_kind: {
2153 basicblock *end, *next2;
2154 end = compiler_new_block(c);
2155 if (end == NULL)
2156 return 0;
2157 next2 = compiler_new_block(c);
2158 if (next2 == NULL)
2159 return 0;
2160 if (!compiler_jump_if(c, e->v.IfExp.test, next2, 0))
2161 return 0;
2162 if (!compiler_jump_if(c, e->v.IfExp.body, next, cond))
2163 return 0;
2164 ADDOP_JREL(c, JUMP_FORWARD, end);
2165 compiler_use_next_block(c, next2);
2166 if (!compiler_jump_if(c, e->v.IfExp.orelse, next, cond))
2167 return 0;
2168 compiler_use_next_block(c, end);
2169 return 1;
2170 }
2171 case Compare_kind: {
2172 Py_ssize_t i, n = asdl_seq_LEN(e->v.Compare.ops) - 1;
2173 if (n > 0) {
2174 basicblock *cleanup = compiler_new_block(c);
2175 if (cleanup == NULL)
2176 return 0;
2177 VISIT(c, expr, e->v.Compare.left);
2178 for (i = 0; i < n; i++) {
2179 VISIT(c, expr,
2180 (expr_ty)asdl_seq_GET(e->v.Compare.comparators, i));
2181 ADDOP(c, DUP_TOP);
2182 ADDOP(c, ROT_THREE);
2183 ADDOP_I(c, COMPARE_OP,
2184 cmpop((cmpop_ty)(asdl_seq_GET(e->v.Compare.ops, i))));
2185 ADDOP_JABS(c, POP_JUMP_IF_FALSE, cleanup);
2186 NEXT_BLOCK(c);
2187 }
2188 VISIT(c, expr, (expr_ty)asdl_seq_GET(e->v.Compare.comparators, n));
2189 ADDOP_I(c, COMPARE_OP,
2190 cmpop((cmpop_ty)(asdl_seq_GET(e->v.Compare.ops, n))));
2191 ADDOP_JABS(c, cond ? POP_JUMP_IF_TRUE : POP_JUMP_IF_FALSE, next);
2192 basicblock *end = compiler_new_block(c);
2193 if (end == NULL)
2194 return 0;
2195 ADDOP_JREL(c, JUMP_FORWARD, end);
2196 compiler_use_next_block(c, cleanup);
2197 ADDOP(c, POP_TOP);
2198 if (!cond) {
2199 ADDOP_JREL(c, JUMP_FORWARD, next);
2200 }
2201 compiler_use_next_block(c, end);
2202 return 1;
2203 }
2204 /* fallback to general implementation */
2205 break;
2206 }
2207 default:
2208 /* fallback to general implementation */
2209 break;
2210 }
2211
2212 /* general implementation */
2213 VISIT(c, expr, e);
2214 ADDOP_JABS(c, cond ? POP_JUMP_IF_TRUE : POP_JUMP_IF_FALSE, next);
2215 return 1;
2216 }
2217
2218 static int
compiler_ifexp(struct compiler * c,expr_ty e)2219 compiler_ifexp(struct compiler *c, expr_ty e)
2220 {
2221 basicblock *end, *next;
2222
2223 assert(e->kind == IfExp_kind);
2224 end = compiler_new_block(c);
2225 if (end == NULL)
2226 return 0;
2227 next = compiler_new_block(c);
2228 if (next == NULL)
2229 return 0;
2230 if (!compiler_jump_if(c, e->v.IfExp.test, next, 0))
2231 return 0;
2232 VISIT(c, expr, e->v.IfExp.body);
2233 ADDOP_JREL(c, JUMP_FORWARD, end);
2234 compiler_use_next_block(c, next);
2235 VISIT(c, expr, e->v.IfExp.orelse);
2236 compiler_use_next_block(c, end);
2237 return 1;
2238 }
2239
2240 static int
compiler_lambda(struct compiler * c,expr_ty e)2241 compiler_lambda(struct compiler *c, expr_ty e)
2242 {
2243 PyCodeObject *co;
2244 PyObject *qualname;
2245 static identifier name;
2246 Py_ssize_t funcflags;
2247 arguments_ty args = e->v.Lambda.args;
2248 assert(e->kind == Lambda_kind);
2249
2250 if (!name) {
2251 name = PyUnicode_InternFromString("<lambda>");
2252 if (!name)
2253 return 0;
2254 }
2255
2256 funcflags = compiler_default_arguments(c, args);
2257 if (funcflags == -1) {
2258 return 0;
2259 }
2260
2261 if (!compiler_enter_scope(c, name, COMPILER_SCOPE_LAMBDA,
2262 (void *)e, e->lineno))
2263 return 0;
2264
2265 /* Make None the first constant, so the lambda can't have a
2266 docstring. */
2267 if (compiler_add_o(c, c->u->u_consts, Py_None) < 0)
2268 return 0;
2269
2270 c->u->u_argcount = asdl_seq_LEN(args->args);
2271 c->u->u_kwonlyargcount = asdl_seq_LEN(args->kwonlyargs);
2272 VISIT_IN_SCOPE(c, expr, e->v.Lambda.body);
2273 if (c->u->u_ste->ste_generator) {
2274 co = assemble(c, 0);
2275 }
2276 else {
2277 ADDOP_IN_SCOPE(c, RETURN_VALUE);
2278 co = assemble(c, 1);
2279 }
2280 qualname = c->u->u_qualname;
2281 Py_INCREF(qualname);
2282 compiler_exit_scope(c);
2283 if (co == NULL)
2284 return 0;
2285
2286 compiler_make_closure(c, co, funcflags, qualname);
2287 Py_DECREF(qualname);
2288 Py_DECREF(co);
2289
2290 return 1;
2291 }
2292
2293 static int
compiler_if(struct compiler * c,stmt_ty s)2294 compiler_if(struct compiler *c, stmt_ty s)
2295 {
2296 basicblock *end, *next;
2297 int constant;
2298 assert(s->kind == If_kind);
2299 end = compiler_new_block(c);
2300 if (end == NULL)
2301 return 0;
2302
2303 constant = expr_constant(s->v.If.test);
2304 /* constant = 0: "if 0"
2305 * constant = 1: "if 1", "if 2", ...
2306 * constant = -1: rest */
2307 if (constant == 0) {
2308 if (s->v.If.orelse)
2309 VISIT_SEQ(c, stmt, s->v.If.orelse);
2310 } else if (constant == 1) {
2311 VISIT_SEQ(c, stmt, s->v.If.body);
2312 } else {
2313 if (asdl_seq_LEN(s->v.If.orelse)) {
2314 next = compiler_new_block(c);
2315 if (next == NULL)
2316 return 0;
2317 }
2318 else
2319 next = end;
2320 if (!compiler_jump_if(c, s->v.If.test, next, 0))
2321 return 0;
2322 VISIT_SEQ(c, stmt, s->v.If.body);
2323 if (asdl_seq_LEN(s->v.If.orelse)) {
2324 ADDOP_JREL(c, JUMP_FORWARD, end);
2325 compiler_use_next_block(c, next);
2326 VISIT_SEQ(c, stmt, s->v.If.orelse);
2327 }
2328 }
2329 compiler_use_next_block(c, end);
2330 return 1;
2331 }
2332
2333 static int
compiler_for(struct compiler * c,stmt_ty s)2334 compiler_for(struct compiler *c, stmt_ty s)
2335 {
2336 basicblock *start, *cleanup, *end;
2337
2338 start = compiler_new_block(c);
2339 cleanup = compiler_new_block(c);
2340 end = compiler_new_block(c);
2341 if (start == NULL || end == NULL || cleanup == NULL)
2342 return 0;
2343 ADDOP_JREL(c, SETUP_LOOP, end);
2344 if (!compiler_push_fblock(c, LOOP, start))
2345 return 0;
2346 VISIT(c, expr, s->v.For.iter);
2347 ADDOP(c, GET_ITER);
2348 compiler_use_next_block(c, start);
2349 ADDOP_JREL(c, FOR_ITER, cleanup);
2350 VISIT(c, expr, s->v.For.target);
2351 VISIT_SEQ(c, stmt, s->v.For.body);
2352 ADDOP_JABS(c, JUMP_ABSOLUTE, start);
2353 compiler_use_next_block(c, cleanup);
2354 ADDOP(c, POP_BLOCK);
2355 compiler_pop_fblock(c, LOOP, start);
2356 VISIT_SEQ(c, stmt, s->v.For.orelse);
2357 compiler_use_next_block(c, end);
2358 return 1;
2359 }
2360
2361
2362 static int
compiler_async_for(struct compiler * c,stmt_ty s)2363 compiler_async_for(struct compiler *c, stmt_ty s)
2364 {
2365 _Py_IDENTIFIER(StopAsyncIteration);
2366
2367 basicblock *try, *except, *end, *after_try, *try_cleanup,
2368 *after_loop_else;
2369
2370 if (c->u->u_scope_type != COMPILER_SCOPE_ASYNC_FUNCTION) {
2371 return compiler_error(c, "'async for' outside async function");
2372 }
2373
2374 PyObject *stop_aiter_error = _PyUnicode_FromId(&PyId_StopAsyncIteration);
2375 if (stop_aiter_error == NULL) {
2376 return 0;
2377 }
2378
2379 try = compiler_new_block(c);
2380 except = compiler_new_block(c);
2381 end = compiler_new_block(c);
2382 after_try = compiler_new_block(c);
2383 try_cleanup = compiler_new_block(c);
2384 after_loop_else = compiler_new_block(c);
2385
2386 if (try == NULL || except == NULL || end == NULL
2387 || after_try == NULL || try_cleanup == NULL
2388 || after_loop_else == NULL)
2389 return 0;
2390
2391 ADDOP_JREL(c, SETUP_LOOP, end);
2392 if (!compiler_push_fblock(c, LOOP, try))
2393 return 0;
2394
2395 VISIT(c, expr, s->v.AsyncFor.iter);
2396 ADDOP(c, GET_AITER);
2397
2398 compiler_use_next_block(c, try);
2399
2400
2401 ADDOP_JREL(c, SETUP_EXCEPT, except);
2402 if (!compiler_push_fblock(c, EXCEPT, try))
2403 return 0;
2404
2405 ADDOP(c, GET_ANEXT);
2406 ADDOP_O(c, LOAD_CONST, Py_None, consts);
2407 ADDOP(c, YIELD_FROM);
2408 VISIT(c, expr, s->v.AsyncFor.target);
2409 ADDOP(c, POP_BLOCK);
2410 compiler_pop_fblock(c, EXCEPT, try);
2411 ADDOP_JREL(c, JUMP_FORWARD, after_try);
2412
2413
2414 compiler_use_next_block(c, except);
2415 ADDOP(c, DUP_TOP);
2416 ADDOP_O(c, LOAD_GLOBAL, stop_aiter_error, names);
2417 ADDOP_I(c, COMPARE_OP, PyCmp_EXC_MATCH);
2418 ADDOP_JABS(c, POP_JUMP_IF_TRUE, try_cleanup);
2419 ADDOP(c, END_FINALLY);
2420
2421 compiler_use_next_block(c, after_try);
2422 VISIT_SEQ(c, stmt, s->v.AsyncFor.body);
2423 ADDOP_JABS(c, JUMP_ABSOLUTE, try);
2424
2425 compiler_use_next_block(c, try_cleanup);
2426 ADDOP(c, POP_TOP);
2427 ADDOP(c, POP_TOP);
2428 ADDOP(c, POP_TOP);
2429 ADDOP(c, POP_EXCEPT); /* for SETUP_EXCEPT */
2430 ADDOP(c, POP_TOP); /* for correct calculation of stack effect */
2431 ADDOP(c, POP_BLOCK); /* for SETUP_LOOP */
2432 compiler_pop_fblock(c, LOOP, try);
2433
2434 compiler_use_next_block(c, after_loop_else);
2435 VISIT_SEQ(c, stmt, s->v.For.orelse);
2436
2437 compiler_use_next_block(c, end);
2438
2439 return 1;
2440 }
2441
2442 static int
compiler_while(struct compiler * c,stmt_ty s)2443 compiler_while(struct compiler *c, stmt_ty s)
2444 {
2445 basicblock *loop, *orelse, *end, *anchor = NULL;
2446 int constant = expr_constant(s->v.While.test);
2447
2448 if (constant == 0) {
2449 if (s->v.While.orelse)
2450 VISIT_SEQ(c, stmt, s->v.While.orelse);
2451 return 1;
2452 }
2453 loop = compiler_new_block(c);
2454 end = compiler_new_block(c);
2455 if (constant == -1) {
2456 anchor = compiler_new_block(c);
2457 if (anchor == NULL)
2458 return 0;
2459 }
2460 if (loop == NULL || end == NULL)
2461 return 0;
2462 if (s->v.While.orelse) {
2463 orelse = compiler_new_block(c);
2464 if (orelse == NULL)
2465 return 0;
2466 }
2467 else
2468 orelse = NULL;
2469
2470 ADDOP_JREL(c, SETUP_LOOP, end);
2471 compiler_use_next_block(c, loop);
2472 if (!compiler_push_fblock(c, LOOP, loop))
2473 return 0;
2474 if (constant == -1) {
2475 if (!compiler_jump_if(c, s->v.While.test, anchor, 0))
2476 return 0;
2477 }
2478 VISIT_SEQ(c, stmt, s->v.While.body);
2479 ADDOP_JABS(c, JUMP_ABSOLUTE, loop);
2480
2481 /* XXX should the two POP instructions be in a separate block
2482 if there is no else clause ?
2483 */
2484
2485 if (constant == -1)
2486 compiler_use_next_block(c, anchor);
2487 ADDOP(c, POP_BLOCK);
2488 compiler_pop_fblock(c, LOOP, loop);
2489 if (orelse != NULL) /* what if orelse is just pass? */
2490 VISIT_SEQ(c, stmt, s->v.While.orelse);
2491 compiler_use_next_block(c, end);
2492
2493 return 1;
2494 }
2495
2496 static int
compiler_continue(struct compiler * c)2497 compiler_continue(struct compiler *c)
2498 {
2499 static const char LOOP_ERROR_MSG[] = "'continue' not properly in loop";
2500 static const char IN_FINALLY_ERROR_MSG[] =
2501 "'continue' not supported inside 'finally' clause";
2502 int i;
2503
2504 if (!c->u->u_nfblocks)
2505 return compiler_error(c, LOOP_ERROR_MSG);
2506 i = c->u->u_nfblocks - 1;
2507 switch (c->u->u_fblock[i].fb_type) {
2508 case LOOP:
2509 ADDOP_JABS(c, JUMP_ABSOLUTE, c->u->u_fblock[i].fb_block);
2510 break;
2511 case EXCEPT:
2512 case FINALLY_TRY:
2513 while (--i >= 0 && c->u->u_fblock[i].fb_type != LOOP) {
2514 /* Prevent continue anywhere under a finally
2515 even if hidden in a sub-try or except. */
2516 if (c->u->u_fblock[i].fb_type == FINALLY_END)
2517 return compiler_error(c, IN_FINALLY_ERROR_MSG);
2518 }
2519 if (i == -1)
2520 return compiler_error(c, LOOP_ERROR_MSG);
2521 ADDOP_JABS(c, CONTINUE_LOOP, c->u->u_fblock[i].fb_block);
2522 break;
2523 case FINALLY_END:
2524 return compiler_error(c, IN_FINALLY_ERROR_MSG);
2525 }
2526
2527 return 1;
2528 }
2529
2530 /* Code generated for "try: <body> finally: <finalbody>" is as follows:
2531
2532 SETUP_FINALLY L
2533 <code for body>
2534 POP_BLOCK
2535 LOAD_CONST <None>
2536 L: <code for finalbody>
2537 END_FINALLY
2538
2539 The special instructions use the block stack. Each block
2540 stack entry contains the instruction that created it (here
2541 SETUP_FINALLY), the level of the value stack at the time the
2542 block stack entry was created, and a label (here L).
2543
2544 SETUP_FINALLY:
2545 Pushes the current value stack level and the label
2546 onto the block stack.
2547 POP_BLOCK:
2548 Pops en entry from the block stack, and pops the value
2549 stack until its level is the same as indicated on the
2550 block stack. (The label is ignored.)
2551 END_FINALLY:
2552 Pops a variable number of entries from the *value* stack
2553 and re-raises the exception they specify. The number of
2554 entries popped depends on the (pseudo) exception type.
2555
2556 The block stack is unwound when an exception is raised:
2557 when a SETUP_FINALLY entry is found, the exception is pushed
2558 onto the value stack (and the exception condition is cleared),
2559 and the interpreter jumps to the label gotten from the block
2560 stack.
2561 */
2562
2563 static int
compiler_try_finally(struct compiler * c,stmt_ty s)2564 compiler_try_finally(struct compiler *c, stmt_ty s)
2565 {
2566 basicblock *body, *end;
2567 body = compiler_new_block(c);
2568 end = compiler_new_block(c);
2569 if (body == NULL || end == NULL)
2570 return 0;
2571
2572 ADDOP_JREL(c, SETUP_FINALLY, end);
2573 compiler_use_next_block(c, body);
2574 if (!compiler_push_fblock(c, FINALLY_TRY, body))
2575 return 0;
2576 if (s->v.Try.handlers && asdl_seq_LEN(s->v.Try.handlers)) {
2577 if (!compiler_try_except(c, s))
2578 return 0;
2579 }
2580 else {
2581 VISIT_SEQ(c, stmt, s->v.Try.body);
2582 }
2583 ADDOP(c, POP_BLOCK);
2584 compiler_pop_fblock(c, FINALLY_TRY, body);
2585
2586 ADDOP_O(c, LOAD_CONST, Py_None, consts);
2587 compiler_use_next_block(c, end);
2588 if (!compiler_push_fblock(c, FINALLY_END, end))
2589 return 0;
2590 VISIT_SEQ(c, stmt, s->v.Try.finalbody);
2591 ADDOP(c, END_FINALLY);
2592 compiler_pop_fblock(c, FINALLY_END, end);
2593
2594 return 1;
2595 }
2596
2597 /*
2598 Code generated for "try: S except E1 as V1: S1 except E2 as V2: S2 ...":
2599 (The contents of the value stack is shown in [], with the top
2600 at the right; 'tb' is trace-back info, 'val' the exception's
2601 associated value, and 'exc' the exception.)
2602
2603 Value stack Label Instruction Argument
2604 [] SETUP_EXCEPT L1
2605 [] <code for S>
2606 [] POP_BLOCK
2607 [] JUMP_FORWARD L0
2608
2609 [tb, val, exc] L1: DUP )
2610 [tb, val, exc, exc] <evaluate E1> )
2611 [tb, val, exc, exc, E1] COMPARE_OP EXC_MATCH ) only if E1
2612 [tb, val, exc, 1-or-0] POP_JUMP_IF_FALSE L2 )
2613 [tb, val, exc] POP
2614 [tb, val] <assign to V1> (or POP if no V1)
2615 [tb] POP
2616 [] <code for S1>
2617 JUMP_FORWARD L0
2618
2619 [tb, val, exc] L2: DUP
2620 .............................etc.......................
2621
2622 [tb, val, exc] Ln+1: END_FINALLY # re-raise exception
2623
2624 [] L0: <next statement>
2625
2626 Of course, parts are not generated if Vi or Ei is not present.
2627 */
2628 static int
compiler_try_except(struct compiler * c,stmt_ty s)2629 compiler_try_except(struct compiler *c, stmt_ty s)
2630 {
2631 basicblock *body, *orelse, *except, *end;
2632 Py_ssize_t i, n;
2633
2634 body = compiler_new_block(c);
2635 except = compiler_new_block(c);
2636 orelse = compiler_new_block(c);
2637 end = compiler_new_block(c);
2638 if (body == NULL || except == NULL || orelse == NULL || end == NULL)
2639 return 0;
2640 ADDOP_JREL(c, SETUP_EXCEPT, except);
2641 compiler_use_next_block(c, body);
2642 if (!compiler_push_fblock(c, EXCEPT, body))
2643 return 0;
2644 VISIT_SEQ(c, stmt, s->v.Try.body);
2645 ADDOP(c, POP_BLOCK);
2646 compiler_pop_fblock(c, EXCEPT, body);
2647 ADDOP_JREL(c, JUMP_FORWARD, orelse);
2648 n = asdl_seq_LEN(s->v.Try.handlers);
2649 compiler_use_next_block(c, except);
2650 for (i = 0; i < n; i++) {
2651 excepthandler_ty handler = (excepthandler_ty)asdl_seq_GET(
2652 s->v.Try.handlers, i);
2653 if (!handler->v.ExceptHandler.type && i < n-1)
2654 return compiler_error(c, "default 'except:' must be last");
2655 c->u->u_lineno_set = 0;
2656 c->u->u_lineno = handler->lineno;
2657 c->u->u_col_offset = handler->col_offset;
2658 except = compiler_new_block(c);
2659 if (except == NULL)
2660 return 0;
2661 if (handler->v.ExceptHandler.type) {
2662 ADDOP(c, DUP_TOP);
2663 VISIT(c, expr, handler->v.ExceptHandler.type);
2664 ADDOP_I(c, COMPARE_OP, PyCmp_EXC_MATCH);
2665 ADDOP_JABS(c, POP_JUMP_IF_FALSE, except);
2666 }
2667 ADDOP(c, POP_TOP);
2668 if (handler->v.ExceptHandler.name) {
2669 basicblock *cleanup_end, *cleanup_body;
2670
2671 cleanup_end = compiler_new_block(c);
2672 cleanup_body = compiler_new_block(c);
2673 if (cleanup_end == NULL || cleanup_body == NULL) {
2674 return 0;
2675 }
2676
2677 compiler_nameop(c, handler->v.ExceptHandler.name, Store);
2678 ADDOP(c, POP_TOP);
2679
2680 /*
2681 try:
2682 # body
2683 except type as name:
2684 try:
2685 # body
2686 finally:
2687 name = None
2688 del name
2689 */
2690
2691 /* second try: */
2692 ADDOP_JREL(c, SETUP_FINALLY, cleanup_end);
2693 compiler_use_next_block(c, cleanup_body);
2694 if (!compiler_push_fblock(c, FINALLY_TRY, cleanup_body))
2695 return 0;
2696
2697 /* second # body */
2698 VISIT_SEQ(c, stmt, handler->v.ExceptHandler.body);
2699 ADDOP(c, POP_BLOCK);
2700 compiler_pop_fblock(c, FINALLY_TRY, cleanup_body);
2701
2702 /* finally: */
2703 ADDOP_O(c, LOAD_CONST, Py_None, consts);
2704 compiler_use_next_block(c, cleanup_end);
2705 if (!compiler_push_fblock(c, FINALLY_END, cleanup_end))
2706 return 0;
2707
2708 /* name = None */
2709 ADDOP_O(c, LOAD_CONST, Py_None, consts);
2710 compiler_nameop(c, handler->v.ExceptHandler.name, Store);
2711
2712 /* del name */
2713 compiler_nameop(c, handler->v.ExceptHandler.name, Del);
2714
2715 ADDOP(c, END_FINALLY);
2716 ADDOP(c, POP_EXCEPT);
2717 compiler_pop_fblock(c, FINALLY_END, cleanup_end);
2718 }
2719 else {
2720 basicblock *cleanup_body;
2721
2722 cleanup_body = compiler_new_block(c);
2723 if (!cleanup_body)
2724 return 0;
2725
2726 ADDOP(c, POP_TOP);
2727 ADDOP(c, POP_TOP);
2728 compiler_use_next_block(c, cleanup_body);
2729 if (!compiler_push_fblock(c, FINALLY_TRY, cleanup_body))
2730 return 0;
2731 VISIT_SEQ(c, stmt, handler->v.ExceptHandler.body);
2732 ADDOP(c, POP_EXCEPT);
2733 compiler_pop_fblock(c, FINALLY_TRY, cleanup_body);
2734 }
2735 ADDOP_JREL(c, JUMP_FORWARD, end);
2736 compiler_use_next_block(c, except);
2737 }
2738 ADDOP(c, END_FINALLY);
2739 compiler_use_next_block(c, orelse);
2740 VISIT_SEQ(c, stmt, s->v.Try.orelse);
2741 compiler_use_next_block(c, end);
2742 return 1;
2743 }
2744
2745 static int
compiler_try(struct compiler * c,stmt_ty s)2746 compiler_try(struct compiler *c, stmt_ty s) {
2747 if (s->v.Try.finalbody && asdl_seq_LEN(s->v.Try.finalbody))
2748 return compiler_try_finally(c, s);
2749 else
2750 return compiler_try_except(c, s);
2751 }
2752
2753
2754 static int
compiler_import_as(struct compiler * c,identifier name,identifier asname)2755 compiler_import_as(struct compiler *c, identifier name, identifier asname)
2756 {
2757 /* The IMPORT_NAME opcode was already generated. This function
2758 merely needs to bind the result to a name.
2759
2760 If there is a dot in name, we need to split it and emit a
2761 IMPORT_FROM for each name.
2762 */
2763 Py_ssize_t len = PyUnicode_GET_LENGTH(name);
2764 Py_ssize_t dot = PyUnicode_FindChar(name, '.', 0, len, 1);
2765 if (dot == -2)
2766 return 0;
2767 if (dot != -1) {
2768 /* Consume the base module name to get the first attribute */
2769 while (1) {
2770 Py_ssize_t pos = dot + 1;
2771 PyObject *attr;
2772 dot = PyUnicode_FindChar(name, '.', pos, len, 1);
2773 if (dot == -2)
2774 return 0;
2775 attr = PyUnicode_Substring(name, pos, (dot != -1) ? dot : len);
2776 if (!attr)
2777 return 0;
2778 ADDOP_N(c, IMPORT_FROM, attr, names);
2779 if (dot == -1) {
2780 break;
2781 }
2782 ADDOP(c, ROT_TWO);
2783 ADDOP(c, POP_TOP);
2784 }
2785 if (!compiler_nameop(c, asname, Store)) {
2786 return 0;
2787 }
2788 ADDOP(c, POP_TOP);
2789 return 1;
2790 }
2791 return compiler_nameop(c, asname, Store);
2792 }
2793
2794 static int
compiler_import(struct compiler * c,stmt_ty s)2795 compiler_import(struct compiler *c, stmt_ty s)
2796 {
2797 /* The Import node stores a module name like a.b.c as a single
2798 string. This is convenient for all cases except
2799 import a.b.c as d
2800 where we need to parse that string to extract the individual
2801 module names.
2802 XXX Perhaps change the representation to make this case simpler?
2803 */
2804 Py_ssize_t i, n = asdl_seq_LEN(s->v.Import.names);
2805
2806 for (i = 0; i < n; i++) {
2807 alias_ty alias = (alias_ty)asdl_seq_GET(s->v.Import.names, i);
2808 int r;
2809
2810 ADDOP_O(c, LOAD_CONST, _PyLong_Zero, consts);
2811 ADDOP_O(c, LOAD_CONST, Py_None, consts);
2812 ADDOP_NAME(c, IMPORT_NAME, alias->name, names);
2813
2814 if (alias->asname) {
2815 r = compiler_import_as(c, alias->name, alias->asname);
2816 if (!r)
2817 return r;
2818 }
2819 else {
2820 identifier tmp = alias->name;
2821 Py_ssize_t dot = PyUnicode_FindChar(
2822 alias->name, '.', 0, PyUnicode_GET_LENGTH(alias->name), 1);
2823 if (dot != -1) {
2824 tmp = PyUnicode_Substring(alias->name, 0, dot);
2825 if (tmp == NULL)
2826 return 0;
2827 }
2828 r = compiler_nameop(c, tmp, Store);
2829 if (dot != -1) {
2830 Py_DECREF(tmp);
2831 }
2832 if (!r)
2833 return r;
2834 }
2835 }
2836 return 1;
2837 }
2838
2839 static int
compiler_from_import(struct compiler * c,stmt_ty s)2840 compiler_from_import(struct compiler *c, stmt_ty s)
2841 {
2842 Py_ssize_t i, n = asdl_seq_LEN(s->v.ImportFrom.names);
2843 PyObject *level, *names;
2844 static PyObject *empty_string;
2845
2846 if (!empty_string) {
2847 empty_string = PyUnicode_FromString("");
2848 if (!empty_string)
2849 return 0;
2850 }
2851
2852 level = PyLong_FromLong(s->v.ImportFrom.level);
2853 if (!level) {
2854 return 0;
2855 }
2856 ADDOP_N(c, LOAD_CONST, level, consts);
2857
2858 names = PyTuple_New(n);
2859 if (!names)
2860 return 0;
2861
2862 /* build up the names */
2863 for (i = 0; i < n; i++) {
2864 alias_ty alias = (alias_ty)asdl_seq_GET(s->v.ImportFrom.names, i);
2865 Py_INCREF(alias->name);
2866 PyTuple_SET_ITEM(names, i, alias->name);
2867 }
2868
2869 if (s->lineno > c->c_future->ff_lineno && s->v.ImportFrom.module &&
2870 _PyUnicode_EqualToASCIIString(s->v.ImportFrom.module, "__future__")) {
2871 Py_DECREF(names);
2872 return compiler_error(c, "from __future__ imports must occur "
2873 "at the beginning of the file");
2874 }
2875 ADDOP_N(c, LOAD_CONST, names, consts);
2876
2877 if (s->v.ImportFrom.module) {
2878 ADDOP_NAME(c, IMPORT_NAME, s->v.ImportFrom.module, names);
2879 }
2880 else {
2881 ADDOP_NAME(c, IMPORT_NAME, empty_string, names);
2882 }
2883 for (i = 0; i < n; i++) {
2884 alias_ty alias = (alias_ty)asdl_seq_GET(s->v.ImportFrom.names, i);
2885 identifier store_name;
2886
2887 if (i == 0 && PyUnicode_READ_CHAR(alias->name, 0) == '*') {
2888 assert(n == 1);
2889 ADDOP(c, IMPORT_STAR);
2890 return 1;
2891 }
2892
2893 ADDOP_NAME(c, IMPORT_FROM, alias->name, names);
2894 store_name = alias->name;
2895 if (alias->asname)
2896 store_name = alias->asname;
2897
2898 if (!compiler_nameop(c, store_name, Store)) {
2899 return 0;
2900 }
2901 }
2902 /* remove imported module */
2903 ADDOP(c, POP_TOP);
2904 return 1;
2905 }
2906
2907 static int
compiler_assert(struct compiler * c,stmt_ty s)2908 compiler_assert(struct compiler *c, stmt_ty s)
2909 {
2910 static PyObject *assertion_error = NULL;
2911 basicblock *end;
2912 PyObject* msg;
2913
2914 if (c->c_optimize)
2915 return 1;
2916 if (assertion_error == NULL) {
2917 assertion_error = PyUnicode_InternFromString("AssertionError");
2918 if (assertion_error == NULL)
2919 return 0;
2920 }
2921 if (s->v.Assert.test->kind == Tuple_kind &&
2922 asdl_seq_LEN(s->v.Assert.test->v.Tuple.elts) > 0) {
2923 msg = PyUnicode_FromString("assertion is always true, "
2924 "perhaps remove parentheses?");
2925 if (msg == NULL)
2926 return 0;
2927 if (PyErr_WarnExplicitObject(PyExc_SyntaxWarning, msg,
2928 c->c_filename, c->u->u_lineno,
2929 NULL, NULL) == -1) {
2930 Py_DECREF(msg);
2931 return 0;
2932 }
2933 Py_DECREF(msg);
2934 }
2935 end = compiler_new_block(c);
2936 if (end == NULL)
2937 return 0;
2938 if (!compiler_jump_if(c, s->v.Assert.test, end, 1))
2939 return 0;
2940 ADDOP_O(c, LOAD_GLOBAL, assertion_error, names);
2941 if (s->v.Assert.msg) {
2942 VISIT(c, expr, s->v.Assert.msg);
2943 ADDOP_I(c, CALL_FUNCTION, 1);
2944 }
2945 ADDOP_I(c, RAISE_VARARGS, 1);
2946 compiler_use_next_block(c, end);
2947 return 1;
2948 }
2949
2950 static int
compiler_visit_stmt_expr(struct compiler * c,expr_ty value)2951 compiler_visit_stmt_expr(struct compiler *c, expr_ty value)
2952 {
2953 if (c->c_interactive && c->c_nestlevel <= 1) {
2954 VISIT(c, expr, value);
2955 ADDOP(c, PRINT_EXPR);
2956 return 1;
2957 }
2958
2959 if (is_const(value)) {
2960 /* ignore constant statement */
2961 return 1;
2962 }
2963
2964 VISIT(c, expr, value);
2965 ADDOP(c, POP_TOP);
2966 return 1;
2967 }
2968
2969 static int
compiler_visit_stmt(struct compiler * c,stmt_ty s)2970 compiler_visit_stmt(struct compiler *c, stmt_ty s)
2971 {
2972 Py_ssize_t i, n;
2973
2974 /* Always assign a lineno to the next instruction for a stmt. */
2975 c->u->u_lineno = s->lineno;
2976 c->u->u_col_offset = s->col_offset;
2977 c->u->u_lineno_set = 0;
2978
2979 switch (s->kind) {
2980 case FunctionDef_kind:
2981 return compiler_function(c, s, 0);
2982 case ClassDef_kind:
2983 return compiler_class(c, s);
2984 case Return_kind:
2985 if (c->u->u_ste->ste_type != FunctionBlock)
2986 return compiler_error(c, "'return' outside function");
2987 if (s->v.Return.value) {
2988 if (c->u->u_ste->ste_coroutine && c->u->u_ste->ste_generator)
2989 return compiler_error(
2990 c, "'return' with value in async generator");
2991 VISIT(c, expr, s->v.Return.value);
2992 }
2993 else
2994 ADDOP_O(c, LOAD_CONST, Py_None, consts);
2995 ADDOP(c, RETURN_VALUE);
2996 break;
2997 case Delete_kind:
2998 VISIT_SEQ(c, expr, s->v.Delete.targets)
2999 break;
3000 case Assign_kind:
3001 n = asdl_seq_LEN(s->v.Assign.targets);
3002 VISIT(c, expr, s->v.Assign.value);
3003 for (i = 0; i < n; i++) {
3004 if (i < n - 1)
3005 ADDOP(c, DUP_TOP);
3006 VISIT(c, expr,
3007 (expr_ty)asdl_seq_GET(s->v.Assign.targets, i));
3008 }
3009 break;
3010 case AugAssign_kind:
3011 return compiler_augassign(c, s);
3012 case AnnAssign_kind:
3013 return compiler_annassign(c, s);
3014 case For_kind:
3015 return compiler_for(c, s);
3016 case While_kind:
3017 return compiler_while(c, s);
3018 case If_kind:
3019 return compiler_if(c, s);
3020 case Raise_kind:
3021 n = 0;
3022 if (s->v.Raise.exc) {
3023 VISIT(c, expr, s->v.Raise.exc);
3024 n++;
3025 if (s->v.Raise.cause) {
3026 VISIT(c, expr, s->v.Raise.cause);
3027 n++;
3028 }
3029 }
3030 ADDOP_I(c, RAISE_VARARGS, (int)n);
3031 break;
3032 case Try_kind:
3033 return compiler_try(c, s);
3034 case Assert_kind:
3035 return compiler_assert(c, s);
3036 case Import_kind:
3037 return compiler_import(c, s);
3038 case ImportFrom_kind:
3039 return compiler_from_import(c, s);
3040 case Global_kind:
3041 case Nonlocal_kind:
3042 break;
3043 case Expr_kind:
3044 return compiler_visit_stmt_expr(c, s->v.Expr.value);
3045 case Pass_kind:
3046 break;
3047 case Break_kind:
3048 if (!compiler_in_loop(c))
3049 return compiler_error(c, "'break' outside loop");
3050 ADDOP(c, BREAK_LOOP);
3051 break;
3052 case Continue_kind:
3053 return compiler_continue(c);
3054 case With_kind:
3055 return compiler_with(c, s, 0);
3056 case AsyncFunctionDef_kind:
3057 return compiler_function(c, s, 1);
3058 case AsyncWith_kind:
3059 return compiler_async_with(c, s, 0);
3060 case AsyncFor_kind:
3061 return compiler_async_for(c, s);
3062 }
3063
3064 return 1;
3065 }
3066
3067 static int
unaryop(unaryop_ty op)3068 unaryop(unaryop_ty op)
3069 {
3070 switch (op) {
3071 case Invert:
3072 return UNARY_INVERT;
3073 case Not:
3074 return UNARY_NOT;
3075 case UAdd:
3076 return UNARY_POSITIVE;
3077 case USub:
3078 return UNARY_NEGATIVE;
3079 default:
3080 PyErr_Format(PyExc_SystemError,
3081 "unary op %d should not be possible", op);
3082 return 0;
3083 }
3084 }
3085
3086 static int
binop(struct compiler * c,operator_ty op)3087 binop(struct compiler *c, operator_ty op)
3088 {
3089 switch (op) {
3090 case Add:
3091 return BINARY_ADD;
3092 case Sub:
3093 return BINARY_SUBTRACT;
3094 case Mult:
3095 return BINARY_MULTIPLY;
3096 case MatMult:
3097 return BINARY_MATRIX_MULTIPLY;
3098 case Div:
3099 return BINARY_TRUE_DIVIDE;
3100 case Mod:
3101 return BINARY_MODULO;
3102 case Pow:
3103 return BINARY_POWER;
3104 case LShift:
3105 return BINARY_LSHIFT;
3106 case RShift:
3107 return BINARY_RSHIFT;
3108 case BitOr:
3109 return BINARY_OR;
3110 case BitXor:
3111 return BINARY_XOR;
3112 case BitAnd:
3113 return BINARY_AND;
3114 case FloorDiv:
3115 return BINARY_FLOOR_DIVIDE;
3116 default:
3117 PyErr_Format(PyExc_SystemError,
3118 "binary op %d should not be possible", op);
3119 return 0;
3120 }
3121 }
3122
3123 static int
inplace_binop(struct compiler * c,operator_ty op)3124 inplace_binop(struct compiler *c, operator_ty op)
3125 {
3126 switch (op) {
3127 case Add:
3128 return INPLACE_ADD;
3129 case Sub:
3130 return INPLACE_SUBTRACT;
3131 case Mult:
3132 return INPLACE_MULTIPLY;
3133 case MatMult:
3134 return INPLACE_MATRIX_MULTIPLY;
3135 case Div:
3136 return INPLACE_TRUE_DIVIDE;
3137 case Mod:
3138 return INPLACE_MODULO;
3139 case Pow:
3140 return INPLACE_POWER;
3141 case LShift:
3142 return INPLACE_LSHIFT;
3143 case RShift:
3144 return INPLACE_RSHIFT;
3145 case BitOr:
3146 return INPLACE_OR;
3147 case BitXor:
3148 return INPLACE_XOR;
3149 case BitAnd:
3150 return INPLACE_AND;
3151 case FloorDiv:
3152 return INPLACE_FLOOR_DIVIDE;
3153 default:
3154 PyErr_Format(PyExc_SystemError,
3155 "inplace binary op %d should not be possible", op);
3156 return 0;
3157 }
3158 }
3159
3160 static int
compiler_nameop(struct compiler * c,identifier name,expr_context_ty ctx)3161 compiler_nameop(struct compiler *c, identifier name, expr_context_ty ctx)
3162 {
3163 int op, scope;
3164 Py_ssize_t arg;
3165 enum { OP_FAST, OP_GLOBAL, OP_DEREF, OP_NAME } optype;
3166
3167 PyObject *dict = c->u->u_names;
3168 PyObject *mangled;
3169 /* XXX AugStore isn't used anywhere! */
3170
3171 assert(!_PyUnicode_EqualToASCIIString(name, "None") &&
3172 !_PyUnicode_EqualToASCIIString(name, "True") &&
3173 !_PyUnicode_EqualToASCIIString(name, "False"));
3174
3175 mangled = _Py_Mangle(c->u->u_private, name);
3176 if (!mangled)
3177 return 0;
3178
3179 op = 0;
3180 optype = OP_NAME;
3181 scope = PyST_GetScope(c->u->u_ste, mangled);
3182 switch (scope) {
3183 case FREE:
3184 dict = c->u->u_freevars;
3185 optype = OP_DEREF;
3186 break;
3187 case CELL:
3188 dict = c->u->u_cellvars;
3189 optype = OP_DEREF;
3190 break;
3191 case LOCAL:
3192 if (c->u->u_ste->ste_type == FunctionBlock)
3193 optype = OP_FAST;
3194 break;
3195 case GLOBAL_IMPLICIT:
3196 if (c->u->u_ste->ste_type == FunctionBlock)
3197 optype = OP_GLOBAL;
3198 break;
3199 case GLOBAL_EXPLICIT:
3200 optype = OP_GLOBAL;
3201 break;
3202 default:
3203 /* scope can be 0 */
3204 break;
3205 }
3206
3207 /* XXX Leave assert here, but handle __doc__ and the like better */
3208 assert(scope || PyUnicode_READ_CHAR(name, 0) == '_');
3209
3210 switch (optype) {
3211 case OP_DEREF:
3212 switch (ctx) {
3213 case Load:
3214 op = (c->u->u_ste->ste_type == ClassBlock) ? LOAD_CLASSDEREF : LOAD_DEREF;
3215 break;
3216 case Store: op = STORE_DEREF; break;
3217 case AugLoad:
3218 case AugStore:
3219 break;
3220 case Del: op = DELETE_DEREF; break;
3221 case Param:
3222 default:
3223 PyErr_SetString(PyExc_SystemError,
3224 "param invalid for deref variable");
3225 return 0;
3226 }
3227 break;
3228 case OP_FAST:
3229 switch (ctx) {
3230 case Load: op = LOAD_FAST; break;
3231 case Store: op = STORE_FAST; break;
3232 case Del: op = DELETE_FAST; break;
3233 case AugLoad:
3234 case AugStore:
3235 break;
3236 case Param:
3237 default:
3238 PyErr_SetString(PyExc_SystemError,
3239 "param invalid for local variable");
3240 return 0;
3241 }
3242 ADDOP_N(c, op, mangled, varnames);
3243 return 1;
3244 case OP_GLOBAL:
3245 switch (ctx) {
3246 case Load: op = LOAD_GLOBAL; break;
3247 case Store: op = STORE_GLOBAL; break;
3248 case Del: op = DELETE_GLOBAL; break;
3249 case AugLoad:
3250 case AugStore:
3251 break;
3252 case Param:
3253 default:
3254 PyErr_SetString(PyExc_SystemError,
3255 "param invalid for global variable");
3256 return 0;
3257 }
3258 break;
3259 case OP_NAME:
3260 switch (ctx) {
3261 case Load: op = LOAD_NAME; break;
3262 case Store: op = STORE_NAME; break;
3263 case Del: op = DELETE_NAME; break;
3264 case AugLoad:
3265 case AugStore:
3266 break;
3267 case Param:
3268 default:
3269 PyErr_SetString(PyExc_SystemError,
3270 "param invalid for name variable");
3271 return 0;
3272 }
3273 break;
3274 }
3275
3276 assert(op);
3277 arg = compiler_add_o(c, dict, mangled);
3278 Py_DECREF(mangled);
3279 if (arg < 0)
3280 return 0;
3281 return compiler_addop_i(c, op, arg);
3282 }
3283
3284 static int
compiler_boolop(struct compiler * c,expr_ty e)3285 compiler_boolop(struct compiler *c, expr_ty e)
3286 {
3287 basicblock *end;
3288 int jumpi;
3289 Py_ssize_t i, n;
3290 asdl_seq *s;
3291
3292 assert(e->kind == BoolOp_kind);
3293 if (e->v.BoolOp.op == And)
3294 jumpi = JUMP_IF_FALSE_OR_POP;
3295 else
3296 jumpi = JUMP_IF_TRUE_OR_POP;
3297 end = compiler_new_block(c);
3298 if (end == NULL)
3299 return 0;
3300 s = e->v.BoolOp.values;
3301 n = asdl_seq_LEN(s) - 1;
3302 assert(n >= 0);
3303 for (i = 0; i < n; ++i) {
3304 VISIT(c, expr, (expr_ty)asdl_seq_GET(s, i));
3305 ADDOP_JABS(c, jumpi, end);
3306 }
3307 VISIT(c, expr, (expr_ty)asdl_seq_GET(s, n));
3308 compiler_use_next_block(c, end);
3309 return 1;
3310 }
3311
3312 static int
starunpack_helper(struct compiler * c,asdl_seq * elts,int single_op,int inner_op,int outer_op)3313 starunpack_helper(struct compiler *c, asdl_seq *elts,
3314 int single_op, int inner_op, int outer_op)
3315 {
3316 Py_ssize_t n = asdl_seq_LEN(elts);
3317 Py_ssize_t i, nsubitems = 0, nseen = 0;
3318 for (i = 0; i < n; i++) {
3319 expr_ty elt = asdl_seq_GET(elts, i);
3320 if (elt->kind == Starred_kind) {
3321 if (nseen) {
3322 ADDOP_I(c, inner_op, nseen);
3323 nseen = 0;
3324 nsubitems++;
3325 }
3326 VISIT(c, expr, elt->v.Starred.value);
3327 nsubitems++;
3328 }
3329 else {
3330 VISIT(c, expr, elt);
3331 nseen++;
3332 }
3333 }
3334 if (nsubitems) {
3335 if (nseen) {
3336 ADDOP_I(c, inner_op, nseen);
3337 nsubitems++;
3338 }
3339 ADDOP_I(c, outer_op, nsubitems);
3340 }
3341 else
3342 ADDOP_I(c, single_op, nseen);
3343 return 1;
3344 }
3345
3346 static int
assignment_helper(struct compiler * c,asdl_seq * elts)3347 assignment_helper(struct compiler *c, asdl_seq *elts)
3348 {
3349 Py_ssize_t n = asdl_seq_LEN(elts);
3350 Py_ssize_t i;
3351 int seen_star = 0;
3352 for (i = 0; i < n; i++) {
3353 expr_ty elt = asdl_seq_GET(elts, i);
3354 if (elt->kind == Starred_kind && !seen_star) {
3355 if ((i >= (1 << 8)) ||
3356 (n-i-1 >= (INT_MAX >> 8)))
3357 return compiler_error(c,
3358 "too many expressions in "
3359 "star-unpacking assignment");
3360 ADDOP_I(c, UNPACK_EX, (i + ((n-i-1) << 8)));
3361 seen_star = 1;
3362 asdl_seq_SET(elts, i, elt->v.Starred.value);
3363 }
3364 else if (elt->kind == Starred_kind) {
3365 return compiler_error(c,
3366 "two starred expressions in assignment");
3367 }
3368 }
3369 if (!seen_star) {
3370 ADDOP_I(c, UNPACK_SEQUENCE, n);
3371 }
3372 VISIT_SEQ(c, expr, elts);
3373 return 1;
3374 }
3375
3376 static int
compiler_list(struct compiler * c,expr_ty e)3377 compiler_list(struct compiler *c, expr_ty e)
3378 {
3379 asdl_seq *elts = e->v.List.elts;
3380 if (e->v.List.ctx == Store) {
3381 return assignment_helper(c, elts);
3382 }
3383 else if (e->v.List.ctx == Load) {
3384 return starunpack_helper(c, elts,
3385 BUILD_LIST, BUILD_TUPLE, BUILD_LIST_UNPACK);
3386 }
3387 else
3388 VISIT_SEQ(c, expr, elts);
3389 return 1;
3390 }
3391
3392 static int
compiler_tuple(struct compiler * c,expr_ty e)3393 compiler_tuple(struct compiler *c, expr_ty e)
3394 {
3395 asdl_seq *elts = e->v.Tuple.elts;
3396 if (e->v.Tuple.ctx == Store) {
3397 return assignment_helper(c, elts);
3398 }
3399 else if (e->v.Tuple.ctx == Load) {
3400 return starunpack_helper(c, elts,
3401 BUILD_TUPLE, BUILD_TUPLE, BUILD_TUPLE_UNPACK);
3402 }
3403 else
3404 VISIT_SEQ(c, expr, elts);
3405 return 1;
3406 }
3407
3408 static int
compiler_set(struct compiler * c,expr_ty e)3409 compiler_set(struct compiler *c, expr_ty e)
3410 {
3411 return starunpack_helper(c, e->v.Set.elts, BUILD_SET,
3412 BUILD_SET, BUILD_SET_UNPACK);
3413 }
3414
3415 static int
are_all_items_const(asdl_seq * seq,Py_ssize_t begin,Py_ssize_t end)3416 are_all_items_const(asdl_seq *seq, Py_ssize_t begin, Py_ssize_t end)
3417 {
3418 Py_ssize_t i;
3419 for (i = begin; i < end; i++) {
3420 expr_ty key = (expr_ty)asdl_seq_GET(seq, i);
3421 if (key == NULL || !is_const(key))
3422 return 0;
3423 }
3424 return 1;
3425 }
3426
3427 static int
compiler_subdict(struct compiler * c,expr_ty e,Py_ssize_t begin,Py_ssize_t end)3428 compiler_subdict(struct compiler *c, expr_ty e, Py_ssize_t begin, Py_ssize_t end)
3429 {
3430 Py_ssize_t i, n = end - begin;
3431 PyObject *keys, *key;
3432 if (n > 1 && are_all_items_const(e->v.Dict.keys, begin, end)) {
3433 for (i = begin; i < end; i++) {
3434 VISIT(c, expr, (expr_ty)asdl_seq_GET(e->v.Dict.values, i));
3435 }
3436 keys = PyTuple_New(n);
3437 if (keys == NULL) {
3438 return 0;
3439 }
3440 for (i = begin; i < end; i++) {
3441 key = get_const_value((expr_ty)asdl_seq_GET(e->v.Dict.keys, i));
3442 Py_INCREF(key);
3443 PyTuple_SET_ITEM(keys, i - begin, key);
3444 }
3445 ADDOP_N(c, LOAD_CONST, keys, consts);
3446 ADDOP_I(c, BUILD_CONST_KEY_MAP, n);
3447 }
3448 else {
3449 for (i = begin; i < end; i++) {
3450 VISIT(c, expr, (expr_ty)asdl_seq_GET(e->v.Dict.keys, i));
3451 VISIT(c, expr, (expr_ty)asdl_seq_GET(e->v.Dict.values, i));
3452 }
3453 ADDOP_I(c, BUILD_MAP, n);
3454 }
3455 return 1;
3456 }
3457
3458 static int
compiler_dict(struct compiler * c,expr_ty e)3459 compiler_dict(struct compiler *c, expr_ty e)
3460 {
3461 Py_ssize_t i, n, elements;
3462 int containers;
3463 int is_unpacking = 0;
3464 n = asdl_seq_LEN(e->v.Dict.values);
3465 containers = 0;
3466 elements = 0;
3467 for (i = 0; i < n; i++) {
3468 is_unpacking = (expr_ty)asdl_seq_GET(e->v.Dict.keys, i) == NULL;
3469 if (elements == 0xFFFF || (elements && is_unpacking)) {
3470 if (!compiler_subdict(c, e, i - elements, i))
3471 return 0;
3472 containers++;
3473 elements = 0;
3474 }
3475 if (is_unpacking) {
3476 VISIT(c, expr, (expr_ty)asdl_seq_GET(e->v.Dict.values, i));
3477 containers++;
3478 }
3479 else {
3480 elements++;
3481 }
3482 }
3483 if (elements || containers == 0) {
3484 if (!compiler_subdict(c, e, n - elements, n))
3485 return 0;
3486 containers++;
3487 }
3488 /* If there is more than one dict, they need to be merged into a new
3489 * dict. If there is one dict and it's an unpacking, then it needs
3490 * to be copied into a new dict." */
3491 if (containers > 1 || is_unpacking) {
3492 ADDOP_I(c, BUILD_MAP_UNPACK, containers);
3493 }
3494 return 1;
3495 }
3496
3497 static int
compiler_compare(struct compiler * c,expr_ty e)3498 compiler_compare(struct compiler *c, expr_ty e)
3499 {
3500 Py_ssize_t i, n;
3501
3502 VISIT(c, expr, e->v.Compare.left);
3503 assert(asdl_seq_LEN(e->v.Compare.ops) > 0);
3504 n = asdl_seq_LEN(e->v.Compare.ops) - 1;
3505 if (n == 0) {
3506 VISIT(c, expr, (expr_ty)asdl_seq_GET(e->v.Compare.comparators, 0));
3507 ADDOP_I(c, COMPARE_OP,
3508 cmpop((cmpop_ty)(asdl_seq_GET(e->v.Compare.ops, 0))));
3509 }
3510 else {
3511 basicblock *cleanup = compiler_new_block(c);
3512 if (cleanup == NULL)
3513 return 0;
3514 for (i = 0; i < n; i++) {
3515 VISIT(c, expr,
3516 (expr_ty)asdl_seq_GET(e->v.Compare.comparators, i));
3517 ADDOP(c, DUP_TOP);
3518 ADDOP(c, ROT_THREE);
3519 ADDOP_I(c, COMPARE_OP,
3520 cmpop((cmpop_ty)(asdl_seq_GET(e->v.Compare.ops, i))));
3521 ADDOP_JABS(c, JUMP_IF_FALSE_OR_POP, cleanup);
3522 NEXT_BLOCK(c);
3523 }
3524 VISIT(c, expr, (expr_ty)asdl_seq_GET(e->v.Compare.comparators, n));
3525 ADDOP_I(c, COMPARE_OP,
3526 cmpop((cmpop_ty)(asdl_seq_GET(e->v.Compare.ops, n))));
3527 basicblock *end = compiler_new_block(c);
3528 if (end == NULL)
3529 return 0;
3530 ADDOP_JREL(c, JUMP_FORWARD, end);
3531 compiler_use_next_block(c, cleanup);
3532 ADDOP(c, ROT_TWO);
3533 ADDOP(c, POP_TOP);
3534 compiler_use_next_block(c, end);
3535 }
3536 return 1;
3537 }
3538
3539 static int
maybe_optimize_method_call(struct compiler * c,expr_ty e)3540 maybe_optimize_method_call(struct compiler *c, expr_ty e)
3541 {
3542 Py_ssize_t argsl, i;
3543 expr_ty meth = e->v.Call.func;
3544 asdl_seq *args = e->v.Call.args;
3545
3546 /* Check that the call node is an attribute access, and that
3547 the call doesn't have keyword parameters. */
3548 if (meth->kind != Attribute_kind || meth->v.Attribute.ctx != Load ||
3549 asdl_seq_LEN(e->v.Call.keywords))
3550 return -1;
3551
3552 /* Check that there are no *varargs types of arguments. */
3553 argsl = asdl_seq_LEN(args);
3554 for (i = 0; i < argsl; i++) {
3555 expr_ty elt = asdl_seq_GET(args, i);
3556 if (elt->kind == Starred_kind) {
3557 return -1;
3558 }
3559 }
3560
3561 /* Alright, we can optimize the code. */
3562 VISIT(c, expr, meth->v.Attribute.value);
3563 ADDOP_NAME(c, LOAD_METHOD, meth->v.Attribute.attr, names);
3564 VISIT_SEQ(c, expr, e->v.Call.args);
3565 ADDOP_I(c, CALL_METHOD, asdl_seq_LEN(e->v.Call.args));
3566 return 1;
3567 }
3568
3569 static int
compiler_call(struct compiler * c,expr_ty e)3570 compiler_call(struct compiler *c, expr_ty e)
3571 {
3572 if (maybe_optimize_method_call(c, e) > 0)
3573 return 1;
3574
3575 VISIT(c, expr, e->v.Call.func);
3576 return compiler_call_helper(c, 0,
3577 e->v.Call.args,
3578 e->v.Call.keywords);
3579 }
3580
3581 static int
compiler_joined_str(struct compiler * c,expr_ty e)3582 compiler_joined_str(struct compiler *c, expr_ty e)
3583 {
3584 VISIT_SEQ(c, expr, e->v.JoinedStr.values);
3585 if (asdl_seq_LEN(e->v.JoinedStr.values) != 1)
3586 ADDOP_I(c, BUILD_STRING, asdl_seq_LEN(e->v.JoinedStr.values));
3587 return 1;
3588 }
3589
3590 /* Used to implement f-strings. Format a single value. */
3591 static int
compiler_formatted_value(struct compiler * c,expr_ty e)3592 compiler_formatted_value(struct compiler *c, expr_ty e)
3593 {
3594 /* Our oparg encodes 2 pieces of information: the conversion
3595 character, and whether or not a format_spec was provided.
3596
3597 Convert the conversion char to 2 bits:
3598 None: 000 0x0 FVC_NONE
3599 !s : 001 0x1 FVC_STR
3600 !r : 010 0x2 FVC_REPR
3601 !a : 011 0x3 FVC_ASCII
3602
3603 next bit is whether or not we have a format spec:
3604 yes : 100 0x4
3605 no : 000 0x0
3606 */
3607
3608 int oparg;
3609
3610 /* Evaluate the expression to be formatted. */
3611 VISIT(c, expr, e->v.FormattedValue.value);
3612
3613 switch (e->v.FormattedValue.conversion) {
3614 case 's': oparg = FVC_STR; break;
3615 case 'r': oparg = FVC_REPR; break;
3616 case 'a': oparg = FVC_ASCII; break;
3617 case -1: oparg = FVC_NONE; break;
3618 default:
3619 PyErr_SetString(PyExc_SystemError,
3620 "Unrecognized conversion character");
3621 return 0;
3622 }
3623 if (e->v.FormattedValue.format_spec) {
3624 /* Evaluate the format spec, and update our opcode arg. */
3625 VISIT(c, expr, e->v.FormattedValue.format_spec);
3626 oparg |= FVS_HAVE_SPEC;
3627 }
3628
3629 /* And push our opcode and oparg */
3630 ADDOP_I(c, FORMAT_VALUE, oparg);
3631 return 1;
3632 }
3633
3634 static int
compiler_subkwargs(struct compiler * c,asdl_seq * keywords,Py_ssize_t begin,Py_ssize_t end)3635 compiler_subkwargs(struct compiler *c, asdl_seq *keywords, Py_ssize_t begin, Py_ssize_t end)
3636 {
3637 Py_ssize_t i, n = end - begin;
3638 keyword_ty kw;
3639 PyObject *keys, *key;
3640 assert(n > 0);
3641 if (n > 1) {
3642 for (i = begin; i < end; i++) {
3643 kw = asdl_seq_GET(keywords, i);
3644 VISIT(c, expr, kw->value);
3645 }
3646 keys = PyTuple_New(n);
3647 if (keys == NULL) {
3648 return 0;
3649 }
3650 for (i = begin; i < end; i++) {
3651 key = ((keyword_ty) asdl_seq_GET(keywords, i))->arg;
3652 Py_INCREF(key);
3653 PyTuple_SET_ITEM(keys, i - begin, key);
3654 }
3655 ADDOP_N(c, LOAD_CONST, keys, consts);
3656 ADDOP_I(c, BUILD_CONST_KEY_MAP, n);
3657 }
3658 else {
3659 /* a for loop only executes once */
3660 for (i = begin; i < end; i++) {
3661 kw = asdl_seq_GET(keywords, i);
3662 ADDOP_O(c, LOAD_CONST, kw->arg, consts);
3663 VISIT(c, expr, kw->value);
3664 }
3665 ADDOP_I(c, BUILD_MAP, n);
3666 }
3667 return 1;
3668 }
3669
3670 /* shared code between compiler_call and compiler_class */
3671 static int
compiler_call_helper(struct compiler * c,int n,asdl_seq * args,asdl_seq * keywords)3672 compiler_call_helper(struct compiler *c,
3673 int n, /* Args already pushed */
3674 asdl_seq *args,
3675 asdl_seq *keywords)
3676 {
3677 Py_ssize_t i, nseen, nelts, nkwelts;
3678 int mustdictunpack = 0;
3679
3680 /* the number of tuples and dictionaries on the stack */
3681 Py_ssize_t nsubargs = 0, nsubkwargs = 0;
3682
3683 nelts = asdl_seq_LEN(args);
3684 nkwelts = asdl_seq_LEN(keywords);
3685
3686 for (i = 0; i < nkwelts; i++) {
3687 keyword_ty kw = asdl_seq_GET(keywords, i);
3688 if (kw->arg == NULL) {
3689 mustdictunpack = 1;
3690 break;
3691 }
3692 }
3693
3694 nseen = n; /* the number of positional arguments on the stack */
3695 for (i = 0; i < nelts; i++) {
3696 expr_ty elt = asdl_seq_GET(args, i);
3697 if (elt->kind == Starred_kind) {
3698 /* A star-arg. If we've seen positional arguments,
3699 pack the positional arguments into a tuple. */
3700 if (nseen) {
3701 ADDOP_I(c, BUILD_TUPLE, nseen);
3702 nseen = 0;
3703 nsubargs++;
3704 }
3705 VISIT(c, expr, elt->v.Starred.value);
3706 nsubargs++;
3707 }
3708 else {
3709 VISIT(c, expr, elt);
3710 nseen++;
3711 }
3712 }
3713
3714 /* Same dance again for keyword arguments */
3715 if (nsubargs || mustdictunpack) {
3716 if (nseen) {
3717 /* Pack up any trailing positional arguments. */
3718 ADDOP_I(c, BUILD_TUPLE, nseen);
3719 nsubargs++;
3720 }
3721 if (nsubargs > 1) {
3722 /* If we ended up with more than one stararg, we need
3723 to concatenate them into a single sequence. */
3724 ADDOP_I(c, BUILD_TUPLE_UNPACK_WITH_CALL, nsubargs);
3725 }
3726 else if (nsubargs == 0) {
3727 ADDOP_I(c, BUILD_TUPLE, 0);
3728 }
3729 nseen = 0; /* the number of keyword arguments on the stack following */
3730 for (i = 0; i < nkwelts; i++) {
3731 keyword_ty kw = asdl_seq_GET(keywords, i);
3732 if (kw->arg == NULL) {
3733 /* A keyword argument unpacking. */
3734 if (nseen) {
3735 if (!compiler_subkwargs(c, keywords, i - nseen, i))
3736 return 0;
3737 nsubkwargs++;
3738 nseen = 0;
3739 }
3740 VISIT(c, expr, kw->value);
3741 nsubkwargs++;
3742 }
3743 else {
3744 nseen++;
3745 }
3746 }
3747 if (nseen) {
3748 /* Pack up any trailing keyword arguments. */
3749 if (!compiler_subkwargs(c, keywords, nkwelts - nseen, nkwelts))
3750 return 0;
3751 nsubkwargs++;
3752 }
3753 if (nsubkwargs > 1) {
3754 /* Pack it all up */
3755 ADDOP_I(c, BUILD_MAP_UNPACK_WITH_CALL, nsubkwargs);
3756 }
3757 ADDOP_I(c, CALL_FUNCTION_EX, nsubkwargs > 0);
3758 return 1;
3759 }
3760 else if (nkwelts) {
3761 PyObject *names;
3762 VISIT_SEQ(c, keyword, keywords);
3763 names = PyTuple_New(nkwelts);
3764 if (names == NULL) {
3765 return 0;
3766 }
3767 for (i = 0; i < nkwelts; i++) {
3768 keyword_ty kw = asdl_seq_GET(keywords, i);
3769 Py_INCREF(kw->arg);
3770 PyTuple_SET_ITEM(names, i, kw->arg);
3771 }
3772 ADDOP_N(c, LOAD_CONST, names, consts);
3773 ADDOP_I(c, CALL_FUNCTION_KW, n + nelts + nkwelts);
3774 return 1;
3775 }
3776 else {
3777 ADDOP_I(c, CALL_FUNCTION, n + nelts);
3778 return 1;
3779 }
3780 }
3781
3782
3783 /* List and set comprehensions and generator expressions work by creating a
3784 nested function to perform the actual iteration. This means that the
3785 iteration variables don't leak into the current scope.
3786 The defined function is called immediately following its definition, with the
3787 result of that call being the result of the expression.
3788 The LC/SC version returns the populated container, while the GE version is
3789 flagged in symtable.c as a generator, so it returns the generator object
3790 when the function is called.
3791 This code *knows* that the loop cannot contain break, continue, or return,
3792 so it cheats and skips the SETUP_LOOP/POP_BLOCK steps used in normal loops.
3793
3794 Possible cleanups:
3795 - iterate over the generator sequence instead of using recursion
3796 */
3797
3798
3799 static int
compiler_comprehension_generator(struct compiler * c,asdl_seq * generators,int gen_index,expr_ty elt,expr_ty val,int type)3800 compiler_comprehension_generator(struct compiler *c,
3801 asdl_seq *generators, int gen_index,
3802 expr_ty elt, expr_ty val, int type)
3803 {
3804 comprehension_ty gen;
3805 gen = (comprehension_ty)asdl_seq_GET(generators, gen_index);
3806 if (gen->is_async) {
3807 return compiler_async_comprehension_generator(
3808 c, generators, gen_index, elt, val, type);
3809 } else {
3810 return compiler_sync_comprehension_generator(
3811 c, generators, gen_index, elt, val, type);
3812 }
3813 }
3814
3815 static int
compiler_sync_comprehension_generator(struct compiler * c,asdl_seq * generators,int gen_index,expr_ty elt,expr_ty val,int type)3816 compiler_sync_comprehension_generator(struct compiler *c,
3817 asdl_seq *generators, int gen_index,
3818 expr_ty elt, expr_ty val, int type)
3819 {
3820 /* generate code for the iterator, then each of the ifs,
3821 and then write to the element */
3822
3823 comprehension_ty gen;
3824 basicblock *start, *anchor, *skip, *if_cleanup;
3825 Py_ssize_t i, n;
3826
3827 start = compiler_new_block(c);
3828 skip = compiler_new_block(c);
3829 if_cleanup = compiler_new_block(c);
3830 anchor = compiler_new_block(c);
3831
3832 if (start == NULL || skip == NULL || if_cleanup == NULL ||
3833 anchor == NULL)
3834 return 0;
3835
3836 gen = (comprehension_ty)asdl_seq_GET(generators, gen_index);
3837
3838 if (gen_index == 0) {
3839 /* Receive outermost iter as an implicit argument */
3840 c->u->u_argcount = 1;
3841 ADDOP_I(c, LOAD_FAST, 0);
3842 }
3843 else {
3844 /* Sub-iter - calculate on the fly */
3845 VISIT(c, expr, gen->iter);
3846 ADDOP(c, GET_ITER);
3847 }
3848 compiler_use_next_block(c, start);
3849 ADDOP_JREL(c, FOR_ITER, anchor);
3850 NEXT_BLOCK(c);
3851 VISIT(c, expr, gen->target);
3852
3853 /* XXX this needs to be cleaned up...a lot! */
3854 n = asdl_seq_LEN(gen->ifs);
3855 for (i = 0; i < n; i++) {
3856 expr_ty e = (expr_ty)asdl_seq_GET(gen->ifs, i);
3857 if (!compiler_jump_if(c, e, if_cleanup, 0))
3858 return 0;
3859 NEXT_BLOCK(c);
3860 }
3861
3862 if (++gen_index < asdl_seq_LEN(generators))
3863 if (!compiler_comprehension_generator(c,
3864 generators, gen_index,
3865 elt, val, type))
3866 return 0;
3867
3868 /* only append after the last for generator */
3869 if (gen_index >= asdl_seq_LEN(generators)) {
3870 /* comprehension specific code */
3871 switch (type) {
3872 case COMP_GENEXP:
3873 VISIT(c, expr, elt);
3874 ADDOP(c, YIELD_VALUE);
3875 ADDOP(c, POP_TOP);
3876 break;
3877 case COMP_LISTCOMP:
3878 VISIT(c, expr, elt);
3879 ADDOP_I(c, LIST_APPEND, gen_index + 1);
3880 break;
3881 case COMP_SETCOMP:
3882 VISIT(c, expr, elt);
3883 ADDOP_I(c, SET_ADD, gen_index + 1);
3884 break;
3885 case COMP_DICTCOMP:
3886 /* With 'd[k] = v', v is evaluated before k, so we do
3887 the same. */
3888 VISIT(c, expr, val);
3889 VISIT(c, expr, elt);
3890 ADDOP_I(c, MAP_ADD, gen_index + 1);
3891 break;
3892 default:
3893 return 0;
3894 }
3895
3896 compiler_use_next_block(c, skip);
3897 }
3898 compiler_use_next_block(c, if_cleanup);
3899 ADDOP_JABS(c, JUMP_ABSOLUTE, start);
3900 compiler_use_next_block(c, anchor);
3901
3902 return 1;
3903 }
3904
3905 static int
compiler_async_comprehension_generator(struct compiler * c,asdl_seq * generators,int gen_index,expr_ty elt,expr_ty val,int type)3906 compiler_async_comprehension_generator(struct compiler *c,
3907 asdl_seq *generators, int gen_index,
3908 expr_ty elt, expr_ty val, int type)
3909 {
3910 _Py_IDENTIFIER(StopAsyncIteration);
3911
3912 comprehension_ty gen;
3913 basicblock *if_cleanup, *try,
3914 *after_try, *except, *try_cleanup;
3915 Py_ssize_t i, n;
3916
3917 PyObject *stop_aiter_error = _PyUnicode_FromId(&PyId_StopAsyncIteration);
3918 if (stop_aiter_error == NULL) {
3919 return 0;
3920 }
3921
3922 try = compiler_new_block(c);
3923 after_try = compiler_new_block(c);
3924 except = compiler_new_block(c);
3925 if_cleanup = compiler_new_block(c);
3926 try_cleanup = compiler_new_block(c);
3927
3928 if (if_cleanup == NULL ||
3929 try == NULL || after_try == NULL ||
3930 except == NULL || try_cleanup == NULL) {
3931 return 0;
3932 }
3933
3934 gen = (comprehension_ty)asdl_seq_GET(generators, gen_index);
3935
3936 if (gen_index == 0) {
3937 /* Receive outermost iter as an implicit argument */
3938 c->u->u_argcount = 1;
3939 ADDOP_I(c, LOAD_FAST, 0);
3940 }
3941 else {
3942 /* Sub-iter - calculate on the fly */
3943 VISIT(c, expr, gen->iter);
3944 ADDOP(c, GET_AITER);
3945 }
3946
3947 compiler_use_next_block(c, try);
3948
3949
3950 ADDOP_JREL(c, SETUP_EXCEPT, except);
3951 if (!compiler_push_fblock(c, EXCEPT, try))
3952 return 0;
3953
3954 ADDOP(c, GET_ANEXT);
3955 ADDOP_O(c, LOAD_CONST, Py_None, consts);
3956 ADDOP(c, YIELD_FROM);
3957 VISIT(c, expr, gen->target);
3958 ADDOP(c, POP_BLOCK);
3959 compiler_pop_fblock(c, EXCEPT, try);
3960 ADDOP_JREL(c, JUMP_FORWARD, after_try);
3961
3962
3963 compiler_use_next_block(c, except);
3964 ADDOP(c, DUP_TOP);
3965 ADDOP_O(c, LOAD_GLOBAL, stop_aiter_error, names);
3966 ADDOP_I(c, COMPARE_OP, PyCmp_EXC_MATCH);
3967 ADDOP_JABS(c, POP_JUMP_IF_TRUE, try_cleanup);
3968 ADDOP(c, END_FINALLY);
3969
3970 compiler_use_next_block(c, after_try);
3971
3972 n = asdl_seq_LEN(gen->ifs);
3973 for (i = 0; i < n; i++) {
3974 expr_ty e = (expr_ty)asdl_seq_GET(gen->ifs, i);
3975 if (!compiler_jump_if(c, e, if_cleanup, 0))
3976 return 0;
3977 NEXT_BLOCK(c);
3978 }
3979
3980 if (++gen_index < asdl_seq_LEN(generators))
3981 if (!compiler_comprehension_generator(c,
3982 generators, gen_index,
3983 elt, val, type))
3984 return 0;
3985
3986 /* only append after the last for generator */
3987 if (gen_index >= asdl_seq_LEN(generators)) {
3988 /* comprehension specific code */
3989 switch (type) {
3990 case COMP_GENEXP:
3991 VISIT(c, expr, elt);
3992 ADDOP(c, YIELD_VALUE);
3993 ADDOP(c, POP_TOP);
3994 break;
3995 case COMP_LISTCOMP:
3996 VISIT(c, expr, elt);
3997 ADDOP_I(c, LIST_APPEND, gen_index + 1);
3998 break;
3999 case COMP_SETCOMP:
4000 VISIT(c, expr, elt);
4001 ADDOP_I(c, SET_ADD, gen_index + 1);
4002 break;
4003 case COMP_DICTCOMP:
4004 /* With 'd[k] = v', v is evaluated before k, so we do
4005 the same. */
4006 VISIT(c, expr, val);
4007 VISIT(c, expr, elt);
4008 ADDOP_I(c, MAP_ADD, gen_index + 1);
4009 break;
4010 default:
4011 return 0;
4012 }
4013 }
4014 compiler_use_next_block(c, if_cleanup);
4015 ADDOP_JABS(c, JUMP_ABSOLUTE, try);
4016
4017 compiler_use_next_block(c, try_cleanup);
4018 ADDOP(c, POP_TOP);
4019 ADDOP(c, POP_TOP);
4020 ADDOP(c, POP_TOP);
4021 ADDOP(c, POP_EXCEPT); /* for SETUP_EXCEPT */
4022 ADDOP(c, POP_TOP);
4023
4024 return 1;
4025 }
4026
4027 static int
compiler_comprehension(struct compiler * c,expr_ty e,int type,identifier name,asdl_seq * generators,expr_ty elt,expr_ty val)4028 compiler_comprehension(struct compiler *c, expr_ty e, int type,
4029 identifier name, asdl_seq *generators, expr_ty elt,
4030 expr_ty val)
4031 {
4032 PyCodeObject *co = NULL;
4033 comprehension_ty outermost;
4034 PyObject *qualname = NULL;
4035 int is_async_function = c->u->u_ste->ste_coroutine;
4036 int is_async_generator = 0;
4037
4038 outermost = (comprehension_ty) asdl_seq_GET(generators, 0);
4039
4040 if (!compiler_enter_scope(c, name, COMPILER_SCOPE_COMPREHENSION,
4041 (void *)e, e->lineno))
4042 {
4043 goto error;
4044 }
4045
4046 is_async_generator = c->u->u_ste->ste_coroutine;
4047
4048 if (is_async_generator && !is_async_function && type != COMP_GENEXP) {
4049 if (e->lineno > c->u->u_lineno) {
4050 c->u->u_lineno = e->lineno;
4051 c->u->u_lineno_set = 0;
4052 }
4053 compiler_error(c, "asynchronous comprehension outside of "
4054 "an asynchronous function");
4055 goto error_in_scope;
4056 }
4057
4058 if (type != COMP_GENEXP) {
4059 int op;
4060 switch (type) {
4061 case COMP_LISTCOMP:
4062 op = BUILD_LIST;
4063 break;
4064 case COMP_SETCOMP:
4065 op = BUILD_SET;
4066 break;
4067 case COMP_DICTCOMP:
4068 op = BUILD_MAP;
4069 break;
4070 default:
4071 PyErr_Format(PyExc_SystemError,
4072 "unknown comprehension type %d", type);
4073 goto error_in_scope;
4074 }
4075
4076 ADDOP_I(c, op, 0);
4077 }
4078
4079 if (!compiler_comprehension_generator(c, generators, 0, elt,
4080 val, type))
4081 goto error_in_scope;
4082
4083 if (type != COMP_GENEXP) {
4084 ADDOP(c, RETURN_VALUE);
4085 }
4086
4087 co = assemble(c, 1);
4088 qualname = c->u->u_qualname;
4089 Py_INCREF(qualname);
4090 compiler_exit_scope(c);
4091 if (co == NULL)
4092 goto error;
4093
4094 if (!compiler_make_closure(c, co, 0, qualname))
4095 goto error;
4096 Py_DECREF(qualname);
4097 Py_DECREF(co);
4098
4099 VISIT(c, expr, outermost->iter);
4100
4101 if (outermost->is_async) {
4102 ADDOP(c, GET_AITER);
4103 } else {
4104 ADDOP(c, GET_ITER);
4105 }
4106
4107 ADDOP_I(c, CALL_FUNCTION, 1);
4108
4109 if (is_async_generator && type != COMP_GENEXP) {
4110 ADDOP(c, GET_AWAITABLE);
4111 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4112 ADDOP(c, YIELD_FROM);
4113 }
4114
4115 return 1;
4116 error_in_scope:
4117 compiler_exit_scope(c);
4118 error:
4119 Py_XDECREF(qualname);
4120 Py_XDECREF(co);
4121 return 0;
4122 }
4123
4124 static int
compiler_genexp(struct compiler * c,expr_ty e)4125 compiler_genexp(struct compiler *c, expr_ty e)
4126 {
4127 static identifier name;
4128 if (!name) {
4129 name = PyUnicode_FromString("<genexpr>");
4130 if (!name)
4131 return 0;
4132 }
4133 assert(e->kind == GeneratorExp_kind);
4134 return compiler_comprehension(c, e, COMP_GENEXP, name,
4135 e->v.GeneratorExp.generators,
4136 e->v.GeneratorExp.elt, NULL);
4137 }
4138
4139 static int
compiler_listcomp(struct compiler * c,expr_ty e)4140 compiler_listcomp(struct compiler *c, expr_ty e)
4141 {
4142 static identifier name;
4143 if (!name) {
4144 name = PyUnicode_FromString("<listcomp>");
4145 if (!name)
4146 return 0;
4147 }
4148 assert(e->kind == ListComp_kind);
4149 return compiler_comprehension(c, e, COMP_LISTCOMP, name,
4150 e->v.ListComp.generators,
4151 e->v.ListComp.elt, NULL);
4152 }
4153
4154 static int
compiler_setcomp(struct compiler * c,expr_ty e)4155 compiler_setcomp(struct compiler *c, expr_ty e)
4156 {
4157 static identifier name;
4158 if (!name) {
4159 name = PyUnicode_FromString("<setcomp>");
4160 if (!name)
4161 return 0;
4162 }
4163 assert(e->kind == SetComp_kind);
4164 return compiler_comprehension(c, e, COMP_SETCOMP, name,
4165 e->v.SetComp.generators,
4166 e->v.SetComp.elt, NULL);
4167 }
4168
4169
4170 static int
compiler_dictcomp(struct compiler * c,expr_ty e)4171 compiler_dictcomp(struct compiler *c, expr_ty e)
4172 {
4173 static identifier name;
4174 if (!name) {
4175 name = PyUnicode_FromString("<dictcomp>");
4176 if (!name)
4177 return 0;
4178 }
4179 assert(e->kind == DictComp_kind);
4180 return compiler_comprehension(c, e, COMP_DICTCOMP, name,
4181 e->v.DictComp.generators,
4182 e->v.DictComp.key, e->v.DictComp.value);
4183 }
4184
4185
4186 static int
compiler_visit_keyword(struct compiler * c,keyword_ty k)4187 compiler_visit_keyword(struct compiler *c, keyword_ty k)
4188 {
4189 VISIT(c, expr, k->value);
4190 return 1;
4191 }
4192
4193 /* Test whether expression is constant. For constants, report
4194 whether they are true or false.
4195
4196 Return values: 1 for true, 0 for false, -1 for non-constant.
4197 */
4198
4199 static int
expr_constant(expr_ty e)4200 expr_constant(expr_ty e)
4201 {
4202 if (is_const(e)) {
4203 return PyObject_IsTrue(get_const_value(e));
4204 }
4205 return -1;
4206 }
4207
4208
4209 /*
4210 Implements the async with statement.
4211
4212 The semantics outlined in that PEP are as follows:
4213
4214 async with EXPR as VAR:
4215 BLOCK
4216
4217 It is implemented roughly as:
4218
4219 context = EXPR
4220 exit = context.__aexit__ # not calling it
4221 value = await context.__aenter__()
4222 try:
4223 VAR = value # if VAR present in the syntax
4224 BLOCK
4225 finally:
4226 if an exception was raised:
4227 exc = copy of (exception, instance, traceback)
4228 else:
4229 exc = (None, None, None)
4230 if not (await exit(*exc)):
4231 raise
4232 */
4233 static int
compiler_async_with(struct compiler * c,stmt_ty s,int pos)4234 compiler_async_with(struct compiler *c, stmt_ty s, int pos)
4235 {
4236 basicblock *block, *finally;
4237 withitem_ty item = asdl_seq_GET(s->v.AsyncWith.items, pos);
4238
4239 assert(s->kind == AsyncWith_kind);
4240 if (c->u->u_scope_type != COMPILER_SCOPE_ASYNC_FUNCTION) {
4241 return compiler_error(c, "'async with' outside async function");
4242 }
4243
4244 block = compiler_new_block(c);
4245 finally = compiler_new_block(c);
4246 if (!block || !finally)
4247 return 0;
4248
4249 /* Evaluate EXPR */
4250 VISIT(c, expr, item->context_expr);
4251
4252 ADDOP(c, BEFORE_ASYNC_WITH);
4253 ADDOP(c, GET_AWAITABLE);
4254 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4255 ADDOP(c, YIELD_FROM);
4256
4257 ADDOP_JREL(c, SETUP_ASYNC_WITH, finally);
4258
4259 /* SETUP_ASYNC_WITH pushes a finally block. */
4260 compiler_use_next_block(c, block);
4261 if (!compiler_push_fblock(c, FINALLY_TRY, block)) {
4262 return 0;
4263 }
4264
4265 if (item->optional_vars) {
4266 VISIT(c, expr, item->optional_vars);
4267 }
4268 else {
4269 /* Discard result from context.__aenter__() */
4270 ADDOP(c, POP_TOP);
4271 }
4272
4273 pos++;
4274 if (pos == asdl_seq_LEN(s->v.AsyncWith.items))
4275 /* BLOCK code */
4276 VISIT_SEQ(c, stmt, s->v.AsyncWith.body)
4277 else if (!compiler_async_with(c, s, pos))
4278 return 0;
4279
4280 /* End of try block; start the finally block */
4281 ADDOP(c, POP_BLOCK);
4282 compiler_pop_fblock(c, FINALLY_TRY, block);
4283
4284 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4285 compiler_use_next_block(c, finally);
4286 if (!compiler_push_fblock(c, FINALLY_END, finally))
4287 return 0;
4288
4289 /* Finally block starts; context.__exit__ is on the stack under
4290 the exception or return information. Just issue our magic
4291 opcode. */
4292 ADDOP(c, WITH_CLEANUP_START);
4293
4294 ADDOP(c, GET_AWAITABLE);
4295 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4296 ADDOP(c, YIELD_FROM);
4297
4298 ADDOP(c, WITH_CLEANUP_FINISH);
4299
4300 /* Finally block ends. */
4301 ADDOP(c, END_FINALLY);
4302 compiler_pop_fblock(c, FINALLY_END, finally);
4303 return 1;
4304 }
4305
4306
4307 /*
4308 Implements the with statement from PEP 343.
4309
4310 The semantics outlined in that PEP are as follows:
4311
4312 with EXPR as VAR:
4313 BLOCK
4314
4315 It is implemented roughly as:
4316
4317 context = EXPR
4318 exit = context.__exit__ # not calling it
4319 value = context.__enter__()
4320 try:
4321 VAR = value # if VAR present in the syntax
4322 BLOCK
4323 finally:
4324 if an exception was raised:
4325 exc = copy of (exception, instance, traceback)
4326 else:
4327 exc = (None, None, None)
4328 exit(*exc)
4329 */
4330 static int
compiler_with(struct compiler * c,stmt_ty s,int pos)4331 compiler_with(struct compiler *c, stmt_ty s, int pos)
4332 {
4333 basicblock *block, *finally;
4334 withitem_ty item = asdl_seq_GET(s->v.With.items, pos);
4335
4336 assert(s->kind == With_kind);
4337
4338 block = compiler_new_block(c);
4339 finally = compiler_new_block(c);
4340 if (!block || !finally)
4341 return 0;
4342
4343 /* Evaluate EXPR */
4344 VISIT(c, expr, item->context_expr);
4345 ADDOP_JREL(c, SETUP_WITH, finally);
4346
4347 /* SETUP_WITH pushes a finally block. */
4348 compiler_use_next_block(c, block);
4349 if (!compiler_push_fblock(c, FINALLY_TRY, block)) {
4350 return 0;
4351 }
4352
4353 if (item->optional_vars) {
4354 VISIT(c, expr, item->optional_vars);
4355 }
4356 else {
4357 /* Discard result from context.__enter__() */
4358 ADDOP(c, POP_TOP);
4359 }
4360
4361 pos++;
4362 if (pos == asdl_seq_LEN(s->v.With.items))
4363 /* BLOCK code */
4364 VISIT_SEQ(c, stmt, s->v.With.body)
4365 else if (!compiler_with(c, s, pos))
4366 return 0;
4367
4368 /* End of try block; start the finally block */
4369 ADDOP(c, POP_BLOCK);
4370 compiler_pop_fblock(c, FINALLY_TRY, block);
4371
4372 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4373 compiler_use_next_block(c, finally);
4374 if (!compiler_push_fblock(c, FINALLY_END, finally))
4375 return 0;
4376
4377 /* Finally block starts; context.__exit__ is on the stack under
4378 the exception or return information. Just issue our magic
4379 opcode. */
4380 ADDOP(c, WITH_CLEANUP_START);
4381 ADDOP(c, WITH_CLEANUP_FINISH);
4382
4383 /* Finally block ends. */
4384 ADDOP(c, END_FINALLY);
4385 compiler_pop_fblock(c, FINALLY_END, finally);
4386 return 1;
4387 }
4388
4389 static int
compiler_visit_expr(struct compiler * c,expr_ty e)4390 compiler_visit_expr(struct compiler *c, expr_ty e)
4391 {
4392 /* If expr e has a different line number than the last expr/stmt,
4393 set a new line number for the next instruction.
4394 */
4395 if (e->lineno > c->u->u_lineno) {
4396 c->u->u_lineno = e->lineno;
4397 c->u->u_lineno_set = 0;
4398 }
4399 /* Updating the column offset is always harmless. */
4400 c->u->u_col_offset = e->col_offset;
4401 switch (e->kind) {
4402 case BoolOp_kind:
4403 return compiler_boolop(c, e);
4404 case BinOp_kind:
4405 VISIT(c, expr, e->v.BinOp.left);
4406 VISIT(c, expr, e->v.BinOp.right);
4407 ADDOP(c, binop(c, e->v.BinOp.op));
4408 break;
4409 case UnaryOp_kind:
4410 VISIT(c, expr, e->v.UnaryOp.operand);
4411 ADDOP(c, unaryop(e->v.UnaryOp.op));
4412 break;
4413 case Lambda_kind:
4414 return compiler_lambda(c, e);
4415 case IfExp_kind:
4416 return compiler_ifexp(c, e);
4417 case Dict_kind:
4418 return compiler_dict(c, e);
4419 case Set_kind:
4420 return compiler_set(c, e);
4421 case GeneratorExp_kind:
4422 return compiler_genexp(c, e);
4423 case ListComp_kind:
4424 return compiler_listcomp(c, e);
4425 case SetComp_kind:
4426 return compiler_setcomp(c, e);
4427 case DictComp_kind:
4428 return compiler_dictcomp(c, e);
4429 case Yield_kind:
4430 if (c->u->u_ste->ste_type != FunctionBlock)
4431 return compiler_error(c, "'yield' outside function");
4432 if (e->v.Yield.value) {
4433 VISIT(c, expr, e->v.Yield.value);
4434 }
4435 else {
4436 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4437 }
4438 ADDOP(c, YIELD_VALUE);
4439 break;
4440 case YieldFrom_kind:
4441 if (c->u->u_ste->ste_type != FunctionBlock)
4442 return compiler_error(c, "'yield' outside function");
4443
4444 if (c->u->u_scope_type == COMPILER_SCOPE_ASYNC_FUNCTION)
4445 return compiler_error(c, "'yield from' inside async function");
4446
4447 VISIT(c, expr, e->v.YieldFrom.value);
4448 ADDOP(c, GET_YIELD_FROM_ITER);
4449 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4450 ADDOP(c, YIELD_FROM);
4451 break;
4452 case Await_kind:
4453 if (c->u->u_ste->ste_type != FunctionBlock)
4454 return compiler_error(c, "'await' outside function");
4455
4456 if (c->u->u_scope_type != COMPILER_SCOPE_ASYNC_FUNCTION &&
4457 c->u->u_scope_type != COMPILER_SCOPE_COMPREHENSION)
4458 return compiler_error(c, "'await' outside async function");
4459
4460 VISIT(c, expr, e->v.Await.value);
4461 ADDOP(c, GET_AWAITABLE);
4462 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4463 ADDOP(c, YIELD_FROM);
4464 break;
4465 case Compare_kind:
4466 return compiler_compare(c, e);
4467 case Call_kind:
4468 return compiler_call(c, e);
4469 case Constant_kind:
4470 ADDOP_O(c, LOAD_CONST, e->v.Constant.value, consts);
4471 break;
4472 case Num_kind:
4473 ADDOP_O(c, LOAD_CONST, e->v.Num.n, consts);
4474 break;
4475 case Str_kind:
4476 ADDOP_O(c, LOAD_CONST, e->v.Str.s, consts);
4477 break;
4478 case JoinedStr_kind:
4479 return compiler_joined_str(c, e);
4480 case FormattedValue_kind:
4481 return compiler_formatted_value(c, e);
4482 case Bytes_kind:
4483 ADDOP_O(c, LOAD_CONST, e->v.Bytes.s, consts);
4484 break;
4485 case Ellipsis_kind:
4486 ADDOP_O(c, LOAD_CONST, Py_Ellipsis, consts);
4487 break;
4488 case NameConstant_kind:
4489 ADDOP_O(c, LOAD_CONST, e->v.NameConstant.value, consts);
4490 break;
4491 /* The following exprs can be assignment targets. */
4492 case Attribute_kind:
4493 if (e->v.Attribute.ctx != AugStore)
4494 VISIT(c, expr, e->v.Attribute.value);
4495 switch (e->v.Attribute.ctx) {
4496 case AugLoad:
4497 ADDOP(c, DUP_TOP);
4498 /* Fall through */
4499 case Load:
4500 ADDOP_NAME(c, LOAD_ATTR, e->v.Attribute.attr, names);
4501 break;
4502 case AugStore:
4503 ADDOP(c, ROT_TWO);
4504 /* Fall through */
4505 case Store:
4506 ADDOP_NAME(c, STORE_ATTR, e->v.Attribute.attr, names);
4507 break;
4508 case Del:
4509 ADDOP_NAME(c, DELETE_ATTR, e->v.Attribute.attr, names);
4510 break;
4511 case Param:
4512 default:
4513 PyErr_SetString(PyExc_SystemError,
4514 "param invalid in attribute expression");
4515 return 0;
4516 }
4517 break;
4518 case Subscript_kind:
4519 switch (e->v.Subscript.ctx) {
4520 case AugLoad:
4521 VISIT(c, expr, e->v.Subscript.value);
4522 VISIT_SLICE(c, e->v.Subscript.slice, AugLoad);
4523 break;
4524 case Load:
4525 VISIT(c, expr, e->v.Subscript.value);
4526 VISIT_SLICE(c, e->v.Subscript.slice, Load);
4527 break;
4528 case AugStore:
4529 VISIT_SLICE(c, e->v.Subscript.slice, AugStore);
4530 break;
4531 case Store:
4532 VISIT(c, expr, e->v.Subscript.value);
4533 VISIT_SLICE(c, e->v.Subscript.slice, Store);
4534 break;
4535 case Del:
4536 VISIT(c, expr, e->v.Subscript.value);
4537 VISIT_SLICE(c, e->v.Subscript.slice, Del);
4538 break;
4539 case Param:
4540 default:
4541 PyErr_SetString(PyExc_SystemError,
4542 "param invalid in subscript expression");
4543 return 0;
4544 }
4545 break;
4546 case Starred_kind:
4547 switch (e->v.Starred.ctx) {
4548 case Store:
4549 /* In all legitimate cases, the Starred node was already replaced
4550 * by compiler_list/compiler_tuple. XXX: is that okay? */
4551 return compiler_error(c,
4552 "starred assignment target must be in a list or tuple");
4553 default:
4554 return compiler_error(c,
4555 "can't use starred expression here");
4556 }
4557 break;
4558 case Name_kind:
4559 return compiler_nameop(c, e->v.Name.id, e->v.Name.ctx);
4560 /* child nodes of List and Tuple will have expr_context set */
4561 case List_kind:
4562 return compiler_list(c, e);
4563 case Tuple_kind:
4564 return compiler_tuple(c, e);
4565 }
4566 return 1;
4567 }
4568
4569 static int
compiler_augassign(struct compiler * c,stmt_ty s)4570 compiler_augassign(struct compiler *c, stmt_ty s)
4571 {
4572 expr_ty e = s->v.AugAssign.target;
4573 expr_ty auge;
4574
4575 assert(s->kind == AugAssign_kind);
4576
4577 switch (e->kind) {
4578 case Attribute_kind:
4579 auge = Attribute(e->v.Attribute.value, e->v.Attribute.attr,
4580 AugLoad, e->lineno, e->col_offset, c->c_arena);
4581 if (auge == NULL)
4582 return 0;
4583 VISIT(c, expr, auge);
4584 VISIT(c, expr, s->v.AugAssign.value);
4585 ADDOP(c, inplace_binop(c, s->v.AugAssign.op));
4586 auge->v.Attribute.ctx = AugStore;
4587 VISIT(c, expr, auge);
4588 break;
4589 case Subscript_kind:
4590 auge = Subscript(e->v.Subscript.value, e->v.Subscript.slice,
4591 AugLoad, e->lineno, e->col_offset, c->c_arena);
4592 if (auge == NULL)
4593 return 0;
4594 VISIT(c, expr, auge);
4595 VISIT(c, expr, s->v.AugAssign.value);
4596 ADDOP(c, inplace_binop(c, s->v.AugAssign.op));
4597 auge->v.Subscript.ctx = AugStore;
4598 VISIT(c, expr, auge);
4599 break;
4600 case Name_kind:
4601 if (!compiler_nameop(c, e->v.Name.id, Load))
4602 return 0;
4603 VISIT(c, expr, s->v.AugAssign.value);
4604 ADDOP(c, inplace_binop(c, s->v.AugAssign.op));
4605 return compiler_nameop(c, e->v.Name.id, Store);
4606 default:
4607 PyErr_Format(PyExc_SystemError,
4608 "invalid node type (%d) for augmented assignment",
4609 e->kind);
4610 return 0;
4611 }
4612 return 1;
4613 }
4614
4615 static int
check_ann_expr(struct compiler * c,expr_ty e)4616 check_ann_expr(struct compiler *c, expr_ty e)
4617 {
4618 VISIT(c, expr, e);
4619 ADDOP(c, POP_TOP);
4620 return 1;
4621 }
4622
4623 static int
check_annotation(struct compiler * c,stmt_ty s)4624 check_annotation(struct compiler *c, stmt_ty s)
4625 {
4626 /* Annotations are only evaluated in a module or class. */
4627 if (c->u->u_scope_type == COMPILER_SCOPE_MODULE ||
4628 c->u->u_scope_type == COMPILER_SCOPE_CLASS) {
4629 return check_ann_expr(c, s->v.AnnAssign.annotation);
4630 }
4631 return 1;
4632 }
4633
4634 static int
check_ann_slice(struct compiler * c,slice_ty sl)4635 check_ann_slice(struct compiler *c, slice_ty sl)
4636 {
4637 switch(sl->kind) {
4638 case Index_kind:
4639 return check_ann_expr(c, sl->v.Index.value);
4640 case Slice_kind:
4641 if (sl->v.Slice.lower && !check_ann_expr(c, sl->v.Slice.lower)) {
4642 return 0;
4643 }
4644 if (sl->v.Slice.upper && !check_ann_expr(c, sl->v.Slice.upper)) {
4645 return 0;
4646 }
4647 if (sl->v.Slice.step && !check_ann_expr(c, sl->v.Slice.step)) {
4648 return 0;
4649 }
4650 break;
4651 default:
4652 PyErr_SetString(PyExc_SystemError,
4653 "unexpected slice kind");
4654 return 0;
4655 }
4656 return 1;
4657 }
4658
4659 static int
check_ann_subscr(struct compiler * c,slice_ty sl)4660 check_ann_subscr(struct compiler *c, slice_ty sl)
4661 {
4662 /* We check that everything in a subscript is defined at runtime. */
4663 Py_ssize_t i, n;
4664
4665 switch (sl->kind) {
4666 case Index_kind:
4667 case Slice_kind:
4668 if (!check_ann_slice(c, sl)) {
4669 return 0;
4670 }
4671 break;
4672 case ExtSlice_kind:
4673 n = asdl_seq_LEN(sl->v.ExtSlice.dims);
4674 for (i = 0; i < n; i++) {
4675 slice_ty subsl = (slice_ty)asdl_seq_GET(sl->v.ExtSlice.dims, i);
4676 switch (subsl->kind) {
4677 case Index_kind:
4678 case Slice_kind:
4679 if (!check_ann_slice(c, subsl)) {
4680 return 0;
4681 }
4682 break;
4683 case ExtSlice_kind:
4684 default:
4685 PyErr_SetString(PyExc_SystemError,
4686 "extended slice invalid in nested slice");
4687 return 0;
4688 }
4689 }
4690 break;
4691 default:
4692 PyErr_Format(PyExc_SystemError,
4693 "invalid subscript kind %d", sl->kind);
4694 return 0;
4695 }
4696 return 1;
4697 }
4698
4699 static int
compiler_annassign(struct compiler * c,stmt_ty s)4700 compiler_annassign(struct compiler *c, stmt_ty s)
4701 {
4702 expr_ty targ = s->v.AnnAssign.target;
4703 PyObject* mangled;
4704
4705 assert(s->kind == AnnAssign_kind);
4706
4707 /* We perform the actual assignment first. */
4708 if (s->v.AnnAssign.value) {
4709 VISIT(c, expr, s->v.AnnAssign.value);
4710 VISIT(c, expr, targ);
4711 }
4712 switch (targ->kind) {
4713 case Name_kind:
4714 /* If we have a simple name in a module or class, store annotation. */
4715 if (s->v.AnnAssign.simple &&
4716 (c->u->u_scope_type == COMPILER_SCOPE_MODULE ||
4717 c->u->u_scope_type == COMPILER_SCOPE_CLASS)) {
4718 if (c->c_future->ff_features & CO_FUTURE_ANNOTATIONS) {
4719 VISIT(c, annexpr, s->v.AnnAssign.annotation)
4720 }
4721 else {
4722 VISIT(c, expr, s->v.AnnAssign.annotation);
4723 }
4724 ADDOP_NAME(c, LOAD_NAME, __annotations__, names);
4725 mangled = _Py_Mangle(c->u->u_private, targ->v.Name.id);
4726 if (!mangled) {
4727 return 0;
4728 }
4729 ADDOP_N(c, LOAD_CONST, mangled, consts);
4730 ADDOP(c, STORE_SUBSCR);
4731 }
4732 break;
4733 case Attribute_kind:
4734 if (!s->v.AnnAssign.value &&
4735 !check_ann_expr(c, targ->v.Attribute.value)) {
4736 return 0;
4737 }
4738 break;
4739 case Subscript_kind:
4740 if (!s->v.AnnAssign.value &&
4741 (!check_ann_expr(c, targ->v.Subscript.value) ||
4742 !check_ann_subscr(c, targ->v.Subscript.slice))) {
4743 return 0;
4744 }
4745 break;
4746 default:
4747 PyErr_Format(PyExc_SystemError,
4748 "invalid node type (%d) for annotated assignment",
4749 targ->kind);
4750 return 0;
4751 }
4752 /* Annotation is evaluated last. */
4753 if (!s->v.AnnAssign.simple && !check_annotation(c, s)) {
4754 return 0;
4755 }
4756 return 1;
4757 }
4758
4759 static int
compiler_push_fblock(struct compiler * c,enum fblocktype t,basicblock * b)4760 compiler_push_fblock(struct compiler *c, enum fblocktype t, basicblock *b)
4761 {
4762 struct fblockinfo *f;
4763 if (c->u->u_nfblocks >= CO_MAXBLOCKS) {
4764 PyErr_SetString(PyExc_SyntaxError,
4765 "too many statically nested blocks");
4766 return 0;
4767 }
4768 f = &c->u->u_fblock[c->u->u_nfblocks++];
4769 f->fb_type = t;
4770 f->fb_block = b;
4771 return 1;
4772 }
4773
4774 static void
compiler_pop_fblock(struct compiler * c,enum fblocktype t,basicblock * b)4775 compiler_pop_fblock(struct compiler *c, enum fblocktype t, basicblock *b)
4776 {
4777 struct compiler_unit *u = c->u;
4778 assert(u->u_nfblocks > 0);
4779 u->u_nfblocks--;
4780 assert(u->u_fblock[u->u_nfblocks].fb_type == t);
4781 assert(u->u_fblock[u->u_nfblocks].fb_block == b);
4782 }
4783
4784 static int
compiler_in_loop(struct compiler * c)4785 compiler_in_loop(struct compiler *c) {
4786 int i;
4787 struct compiler_unit *u = c->u;
4788 for (i = 0; i < u->u_nfblocks; ++i) {
4789 if (u->u_fblock[i].fb_type == LOOP)
4790 return 1;
4791 }
4792 return 0;
4793 }
4794 /* Raises a SyntaxError and returns 0.
4795 If something goes wrong, a different exception may be raised.
4796 */
4797
4798 static int
compiler_error(struct compiler * c,const char * errstr)4799 compiler_error(struct compiler *c, const char *errstr)
4800 {
4801 PyObject *loc;
4802 PyObject *u = NULL, *v = NULL;
4803
4804 loc = PyErr_ProgramTextObject(c->c_filename, c->u->u_lineno);
4805 if (!loc) {
4806 Py_INCREF(Py_None);
4807 loc = Py_None;
4808 }
4809 u = Py_BuildValue("(OiiO)", c->c_filename, c->u->u_lineno,
4810 c->u->u_col_offset, loc);
4811 if (!u)
4812 goto exit;
4813 v = Py_BuildValue("(zO)", errstr, u);
4814 if (!v)
4815 goto exit;
4816 PyErr_SetObject(PyExc_SyntaxError, v);
4817 exit:
4818 Py_DECREF(loc);
4819 Py_XDECREF(u);
4820 Py_XDECREF(v);
4821 return 0;
4822 }
4823
4824 static int
compiler_handle_subscr(struct compiler * c,const char * kind,expr_context_ty ctx)4825 compiler_handle_subscr(struct compiler *c, const char *kind,
4826 expr_context_ty ctx)
4827 {
4828 int op = 0;
4829
4830 /* XXX this code is duplicated */
4831 switch (ctx) {
4832 case AugLoad: /* fall through to Load */
4833 case Load: op = BINARY_SUBSCR; break;
4834 case AugStore:/* fall through to Store */
4835 case Store: op = STORE_SUBSCR; break;
4836 case Del: op = DELETE_SUBSCR; break;
4837 case Param:
4838 PyErr_Format(PyExc_SystemError,
4839 "invalid %s kind %d in subscript\n",
4840 kind, ctx);
4841 return 0;
4842 }
4843 if (ctx == AugLoad) {
4844 ADDOP(c, DUP_TOP_TWO);
4845 }
4846 else if (ctx == AugStore) {
4847 ADDOP(c, ROT_THREE);
4848 }
4849 ADDOP(c, op);
4850 return 1;
4851 }
4852
4853 static int
compiler_slice(struct compiler * c,slice_ty s,expr_context_ty ctx)4854 compiler_slice(struct compiler *c, slice_ty s, expr_context_ty ctx)
4855 {
4856 int n = 2;
4857 assert(s->kind == Slice_kind);
4858
4859 /* only handles the cases where BUILD_SLICE is emitted */
4860 if (s->v.Slice.lower) {
4861 VISIT(c, expr, s->v.Slice.lower);
4862 }
4863 else {
4864 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4865 }
4866
4867 if (s->v.Slice.upper) {
4868 VISIT(c, expr, s->v.Slice.upper);
4869 }
4870 else {
4871 ADDOP_O(c, LOAD_CONST, Py_None, consts);
4872 }
4873
4874 if (s->v.Slice.step) {
4875 n++;
4876 VISIT(c, expr, s->v.Slice.step);
4877 }
4878 ADDOP_I(c, BUILD_SLICE, n);
4879 return 1;
4880 }
4881
4882 static int
compiler_visit_nested_slice(struct compiler * c,slice_ty s,expr_context_ty ctx)4883 compiler_visit_nested_slice(struct compiler *c, slice_ty s,
4884 expr_context_ty ctx)
4885 {
4886 switch (s->kind) {
4887 case Slice_kind:
4888 return compiler_slice(c, s, ctx);
4889 case Index_kind:
4890 VISIT(c, expr, s->v.Index.value);
4891 break;
4892 case ExtSlice_kind:
4893 default:
4894 PyErr_SetString(PyExc_SystemError,
4895 "extended slice invalid in nested slice");
4896 return 0;
4897 }
4898 return 1;
4899 }
4900
4901 static int
compiler_visit_slice(struct compiler * c,slice_ty s,expr_context_ty ctx)4902 compiler_visit_slice(struct compiler *c, slice_ty s, expr_context_ty ctx)
4903 {
4904 const char * kindname = NULL;
4905 switch (s->kind) {
4906 case Index_kind:
4907 kindname = "index";
4908 if (ctx != AugStore) {
4909 VISIT(c, expr, s->v.Index.value);
4910 }
4911 break;
4912 case Slice_kind:
4913 kindname = "slice";
4914 if (ctx != AugStore) {
4915 if (!compiler_slice(c, s, ctx))
4916 return 0;
4917 }
4918 break;
4919 case ExtSlice_kind:
4920 kindname = "extended slice";
4921 if (ctx != AugStore) {
4922 Py_ssize_t i, n = asdl_seq_LEN(s->v.ExtSlice.dims);
4923 for (i = 0; i < n; i++) {
4924 slice_ty sub = (slice_ty)asdl_seq_GET(
4925 s->v.ExtSlice.dims, i);
4926 if (!compiler_visit_nested_slice(c, sub, ctx))
4927 return 0;
4928 }
4929 ADDOP_I(c, BUILD_TUPLE, n);
4930 }
4931 break;
4932 default:
4933 PyErr_Format(PyExc_SystemError,
4934 "invalid subscript kind %d", s->kind);
4935 return 0;
4936 }
4937 return compiler_handle_subscr(c, kindname, ctx);
4938 }
4939
4940 /* End of the compiler section, beginning of the assembler section */
4941
4942 /* do depth-first search of basic block graph, starting with block.
4943 post records the block indices in post-order.
4944
4945 XXX must handle implicit jumps from one block to next
4946 */
4947
4948 struct assembler {
4949 PyObject *a_bytecode; /* string containing bytecode */
4950 int a_offset; /* offset into bytecode */
4951 int a_nblocks; /* number of reachable blocks */
4952 basicblock **a_postorder; /* list of blocks in dfs postorder */
4953 PyObject *a_lnotab; /* string containing lnotab */
4954 int a_lnotab_off; /* offset into lnotab */
4955 int a_lineno; /* last lineno of emitted instruction */
4956 int a_lineno_off; /* bytecode offset of last lineno */
4957 };
4958
4959 static void
dfs(struct compiler * c,basicblock * b,struct assembler * a,int end)4960 dfs(struct compiler *c, basicblock *b, struct assembler *a, int end)
4961 {
4962 int i, j;
4963
4964 /* Get rid of recursion for normal control flow.
4965 Since the number of blocks is limited, unused space in a_postorder
4966 (from a_nblocks to end) can be used as a stack for still not ordered
4967 blocks. */
4968 for (j = end; b && !b->b_seen; b = b->b_next) {
4969 b->b_seen = 1;
4970 assert(a->a_nblocks < j);
4971 a->a_postorder[--j] = b;
4972 }
4973 while (j < end) {
4974 b = a->a_postorder[j++];
4975 for (i = 0; i < b->b_iused; i++) {
4976 struct instr *instr = &b->b_instr[i];
4977 if (instr->i_jrel || instr->i_jabs)
4978 dfs(c, instr->i_target, a, j);
4979 }
4980 assert(a->a_nblocks < j);
4981 a->a_postorder[a->a_nblocks++] = b;
4982 }
4983 }
4984
4985 Py_LOCAL_INLINE(void)
stackdepth_push(basicblock *** sp,basicblock * b,int depth)4986 stackdepth_push(basicblock ***sp, basicblock *b, int depth)
4987 {
4988 /* XXX b->b_startdepth > depth only for the target of SETUP_FINALLY,
4989 * SETUP_WITH and SETUP_ASYNC_WITH. */
4990 assert(b->b_startdepth < 0 || b->b_startdepth >= depth);
4991 if (b->b_startdepth < depth) {
4992 assert(b->b_startdepth < 0);
4993 b->b_startdepth = depth;
4994 *(*sp)++ = b;
4995 }
4996 }
4997
4998 /* Find the flow path that needs the largest stack. We assume that
4999 * cycles in the flow graph have no net effect on the stack depth.
5000 */
5001 static int
stackdepth(struct compiler * c)5002 stackdepth(struct compiler *c)
5003 {
5004 basicblock *b, *entryblock = NULL;
5005 basicblock **stack, **sp;
5006 int nblocks = 0, maxdepth = 0;
5007 for (b = c->u->u_blocks; b != NULL; b = b->b_list) {
5008 b->b_startdepth = INT_MIN;
5009 entryblock = b;
5010 nblocks++;
5011 }
5012 if (!entryblock)
5013 return 0;
5014 stack = (basicblock **)PyObject_Malloc(sizeof(basicblock *) * nblocks);
5015 if (!stack) {
5016 PyErr_NoMemory();
5017 return -1;
5018 }
5019
5020 sp = stack;
5021 stackdepth_push(&sp, entryblock, 0);
5022 while (sp != stack) {
5023 b = *--sp;
5024 int depth = b->b_startdepth;
5025 assert(depth >= 0);
5026 basicblock *next = b->b_next;
5027 for (int i = 0; i < b->b_iused; i++) {
5028 struct instr *instr = &b->b_instr[i];
5029 int effect = stack_effect(instr->i_opcode, instr->i_oparg, 0);
5030 if (effect == PY_INVALID_STACK_EFFECT) {
5031 fprintf(stderr, "opcode = %d\n", instr->i_opcode);
5032 Py_FatalError("PyCompile_OpcodeStackEffect()");
5033 }
5034 int new_depth = depth + effect;
5035 if (new_depth > maxdepth) {
5036 maxdepth = new_depth;
5037 }
5038 assert(depth >= 0); /* invalid code or bug in stackdepth() */
5039 if (instr->i_jrel || instr->i_jabs) {
5040 effect = stack_effect(instr->i_opcode, instr->i_oparg, 1);
5041 assert(effect != PY_INVALID_STACK_EFFECT);
5042 int target_depth = depth + effect;
5043 if (target_depth > maxdepth) {
5044 maxdepth = target_depth;
5045 }
5046 assert(target_depth >= 0); /* invalid code or bug in stackdepth() */
5047 if (instr->i_opcode == CONTINUE_LOOP) {
5048 /* Pops a variable number of values from the stack,
5049 * but the target should be already proceeding.
5050 */
5051 assert(instr->i_target->b_startdepth >= 0);
5052 assert(instr->i_target->b_startdepth <= depth);
5053 /* remaining code is dead */
5054 next = NULL;
5055 break;
5056 }
5057 stackdepth_push(&sp, instr->i_target, target_depth);
5058 }
5059 depth = new_depth;
5060 if (instr->i_opcode == JUMP_ABSOLUTE ||
5061 instr->i_opcode == JUMP_FORWARD ||
5062 instr->i_opcode == RETURN_VALUE ||
5063 instr->i_opcode == RAISE_VARARGS ||
5064 instr->i_opcode == BREAK_LOOP)
5065 {
5066 /* remaining code is dead */
5067 next = NULL;
5068 break;
5069 }
5070 }
5071 if (next != NULL) {
5072 stackdepth_push(&sp, next, depth);
5073 }
5074 }
5075 PyObject_Free(stack);
5076 return maxdepth;
5077 }
5078
5079 static int
assemble_init(struct assembler * a,int nblocks,int firstlineno)5080 assemble_init(struct assembler *a, int nblocks, int firstlineno)
5081 {
5082 memset(a, 0, sizeof(struct assembler));
5083 a->a_lineno = firstlineno;
5084 a->a_bytecode = PyBytes_FromStringAndSize(NULL, DEFAULT_CODE_SIZE);
5085 if (!a->a_bytecode)
5086 return 0;
5087 a->a_lnotab = PyBytes_FromStringAndSize(NULL, DEFAULT_LNOTAB_SIZE);
5088 if (!a->a_lnotab)
5089 return 0;
5090 if ((size_t)nblocks > SIZE_MAX / sizeof(basicblock *)) {
5091 PyErr_NoMemory();
5092 return 0;
5093 }
5094 a->a_postorder = (basicblock **)PyObject_Malloc(
5095 sizeof(basicblock *) * nblocks);
5096 if (!a->a_postorder) {
5097 PyErr_NoMemory();
5098 return 0;
5099 }
5100 return 1;
5101 }
5102
5103 static void
assemble_free(struct assembler * a)5104 assemble_free(struct assembler *a)
5105 {
5106 Py_XDECREF(a->a_bytecode);
5107 Py_XDECREF(a->a_lnotab);
5108 if (a->a_postorder)
5109 PyObject_Free(a->a_postorder);
5110 }
5111
5112 static int
blocksize(basicblock * b)5113 blocksize(basicblock *b)
5114 {
5115 int i;
5116 int size = 0;
5117
5118 for (i = 0; i < b->b_iused; i++)
5119 size += instrsize(b->b_instr[i].i_oparg);
5120 return size;
5121 }
5122
5123 /* Appends a pair to the end of the line number table, a_lnotab, representing
5124 the instruction's bytecode offset and line number. See
5125 Objects/lnotab_notes.txt for the description of the line number table. */
5126
5127 static int
assemble_lnotab(struct assembler * a,struct instr * i)5128 assemble_lnotab(struct assembler *a, struct instr *i)
5129 {
5130 int d_bytecode, d_lineno;
5131 Py_ssize_t len;
5132 unsigned char *lnotab;
5133
5134 d_bytecode = (a->a_offset - a->a_lineno_off) * sizeof(_Py_CODEUNIT);
5135 d_lineno = i->i_lineno - a->a_lineno;
5136
5137 assert(d_bytecode >= 0);
5138
5139 if(d_bytecode == 0 && d_lineno == 0)
5140 return 1;
5141
5142 if (d_bytecode > 255) {
5143 int j, nbytes, ncodes = d_bytecode / 255;
5144 nbytes = a->a_lnotab_off + 2 * ncodes;
5145 len = PyBytes_GET_SIZE(a->a_lnotab);
5146 if (nbytes >= len) {
5147 if ((len <= INT_MAX / 2) && (len * 2 < nbytes))
5148 len = nbytes;
5149 else if (len <= INT_MAX / 2)
5150 len *= 2;
5151 else {
5152 PyErr_NoMemory();
5153 return 0;
5154 }
5155 if (_PyBytes_Resize(&a->a_lnotab, len) < 0)
5156 return 0;
5157 }
5158 lnotab = (unsigned char *)
5159 PyBytes_AS_STRING(a->a_lnotab) + a->a_lnotab_off;
5160 for (j = 0; j < ncodes; j++) {
5161 *lnotab++ = 255;
5162 *lnotab++ = 0;
5163 }
5164 d_bytecode -= ncodes * 255;
5165 a->a_lnotab_off += ncodes * 2;
5166 }
5167 assert(0 <= d_bytecode && d_bytecode <= 255);
5168
5169 if (d_lineno < -128 || 127 < d_lineno) {
5170 int j, nbytes, ncodes, k;
5171 if (d_lineno < 0) {
5172 k = -128;
5173 /* use division on positive numbers */
5174 ncodes = (-d_lineno) / 128;
5175 }
5176 else {
5177 k = 127;
5178 ncodes = d_lineno / 127;
5179 }
5180 d_lineno -= ncodes * k;
5181 assert(ncodes >= 1);
5182 nbytes = a->a_lnotab_off + 2 * ncodes;
5183 len = PyBytes_GET_SIZE(a->a_lnotab);
5184 if (nbytes >= len) {
5185 if ((len <= INT_MAX / 2) && len * 2 < nbytes)
5186 len = nbytes;
5187 else if (len <= INT_MAX / 2)
5188 len *= 2;
5189 else {
5190 PyErr_NoMemory();
5191 return 0;
5192 }
5193 if (_PyBytes_Resize(&a->a_lnotab, len) < 0)
5194 return 0;
5195 }
5196 lnotab = (unsigned char *)
5197 PyBytes_AS_STRING(a->a_lnotab) + a->a_lnotab_off;
5198 *lnotab++ = d_bytecode;
5199 *lnotab++ = k;
5200 d_bytecode = 0;
5201 for (j = 1; j < ncodes; j++) {
5202 *lnotab++ = 0;
5203 *lnotab++ = k;
5204 }
5205 a->a_lnotab_off += ncodes * 2;
5206 }
5207 assert(-128 <= d_lineno && d_lineno <= 127);
5208
5209 len = PyBytes_GET_SIZE(a->a_lnotab);
5210 if (a->a_lnotab_off + 2 >= len) {
5211 if (_PyBytes_Resize(&a->a_lnotab, len * 2) < 0)
5212 return 0;
5213 }
5214 lnotab = (unsigned char *)
5215 PyBytes_AS_STRING(a->a_lnotab) + a->a_lnotab_off;
5216
5217 a->a_lnotab_off += 2;
5218 if (d_bytecode) {
5219 *lnotab++ = d_bytecode;
5220 *lnotab++ = d_lineno;
5221 }
5222 else { /* First line of a block; def stmt, etc. */
5223 *lnotab++ = 0;
5224 *lnotab++ = d_lineno;
5225 }
5226 a->a_lineno = i->i_lineno;
5227 a->a_lineno_off = a->a_offset;
5228 return 1;
5229 }
5230
5231 /* assemble_emit()
5232 Extend the bytecode with a new instruction.
5233 Update lnotab if necessary.
5234 */
5235
5236 static int
assemble_emit(struct assembler * a,struct instr * i)5237 assemble_emit(struct assembler *a, struct instr *i)
5238 {
5239 int size, arg = 0;
5240 Py_ssize_t len = PyBytes_GET_SIZE(a->a_bytecode);
5241 _Py_CODEUNIT *code;
5242
5243 arg = i->i_oparg;
5244 size = instrsize(arg);
5245 if (i->i_lineno && !assemble_lnotab(a, i))
5246 return 0;
5247 if (a->a_offset + size >= len / (int)sizeof(_Py_CODEUNIT)) {
5248 if (len > PY_SSIZE_T_MAX / 2)
5249 return 0;
5250 if (_PyBytes_Resize(&a->a_bytecode, len * 2) < 0)
5251 return 0;
5252 }
5253 code = (_Py_CODEUNIT *)PyBytes_AS_STRING(a->a_bytecode) + a->a_offset;
5254 a->a_offset += size;
5255 write_op_arg(code, i->i_opcode, arg, size);
5256 return 1;
5257 }
5258
5259 static void
assemble_jump_offsets(struct assembler * a,struct compiler * c)5260 assemble_jump_offsets(struct assembler *a, struct compiler *c)
5261 {
5262 basicblock *b;
5263 int bsize, totsize, extended_arg_recompile;
5264 int i;
5265
5266 /* Compute the size of each block and fixup jump args.
5267 Replace block pointer with position in bytecode. */
5268 do {
5269 totsize = 0;
5270 for (i = a->a_nblocks - 1; i >= 0; i--) {
5271 b = a->a_postorder[i];
5272 bsize = blocksize(b);
5273 b->b_offset = totsize;
5274 totsize += bsize;
5275 }
5276 extended_arg_recompile = 0;
5277 for (b = c->u->u_blocks; b != NULL; b = b->b_list) {
5278 bsize = b->b_offset;
5279 for (i = 0; i < b->b_iused; i++) {
5280 struct instr *instr = &b->b_instr[i];
5281 int isize = instrsize(instr->i_oparg);
5282 /* Relative jumps are computed relative to
5283 the instruction pointer after fetching
5284 the jump instruction.
5285 */
5286 bsize += isize;
5287 if (instr->i_jabs || instr->i_jrel) {
5288 instr->i_oparg = instr->i_target->b_offset;
5289 if (instr->i_jrel) {
5290 instr->i_oparg -= bsize;
5291 }
5292 instr->i_oparg *= sizeof(_Py_CODEUNIT);
5293 if (instrsize(instr->i_oparg) != isize) {
5294 extended_arg_recompile = 1;
5295 }
5296 }
5297 }
5298 }
5299
5300 /* XXX: This is an awful hack that could hurt performance, but
5301 on the bright side it should work until we come up
5302 with a better solution.
5303
5304 The issue is that in the first loop blocksize() is called
5305 which calls instrsize() which requires i_oparg be set
5306 appropriately. There is a bootstrap problem because
5307 i_oparg is calculated in the second loop above.
5308
5309 So we loop until we stop seeing new EXTENDED_ARGs.
5310 The only EXTENDED_ARGs that could be popping up are
5311 ones in jump instructions. So this should converge
5312 fairly quickly.
5313 */
5314 } while (extended_arg_recompile);
5315 }
5316
5317 static PyObject *
dict_keys_inorder(PyObject * dict,Py_ssize_t offset)5318 dict_keys_inorder(PyObject *dict, Py_ssize_t offset)
5319 {
5320 PyObject *tuple, *k, *v;
5321 Py_ssize_t i, pos = 0, size = PyDict_GET_SIZE(dict);
5322
5323 tuple = PyTuple_New(size);
5324 if (tuple == NULL)
5325 return NULL;
5326 while (PyDict_Next(dict, &pos, &k, &v)) {
5327 i = PyLong_AS_LONG(v);
5328 /* The keys of the dictionary are tuples. (see compiler_add_o
5329 * and _PyCode_ConstantKey). The object we want is always second,
5330 * though. */
5331 k = PyTuple_GET_ITEM(k, 1);
5332 Py_INCREF(k);
5333 assert((i - offset) < size);
5334 assert((i - offset) >= 0);
5335 PyTuple_SET_ITEM(tuple, i - offset, k);
5336 }
5337 return tuple;
5338 }
5339
5340 static int
compute_code_flags(struct compiler * c)5341 compute_code_flags(struct compiler *c)
5342 {
5343 PySTEntryObject *ste = c->u->u_ste;
5344 int flags = 0;
5345 if (ste->ste_type == FunctionBlock) {
5346 flags |= CO_NEWLOCALS | CO_OPTIMIZED;
5347 if (ste->ste_nested)
5348 flags |= CO_NESTED;
5349 if (ste->ste_generator && !ste->ste_coroutine)
5350 flags |= CO_GENERATOR;
5351 if (!ste->ste_generator && ste->ste_coroutine)
5352 flags |= CO_COROUTINE;
5353 if (ste->ste_generator && ste->ste_coroutine)
5354 flags |= CO_ASYNC_GENERATOR;
5355 if (ste->ste_varargs)
5356 flags |= CO_VARARGS;
5357 if (ste->ste_varkeywords)
5358 flags |= CO_VARKEYWORDS;
5359 }
5360
5361 /* (Only) inherit compilerflags in PyCF_MASK */
5362 flags |= (c->c_flags->cf_flags & PyCF_MASK);
5363
5364 return flags;
5365 }
5366
5367 static PyCodeObject *
makecode(struct compiler * c,struct assembler * a)5368 makecode(struct compiler *c, struct assembler *a)
5369 {
5370 PyObject *tmp;
5371 PyCodeObject *co = NULL;
5372 PyObject *consts = NULL;
5373 PyObject *names = NULL;
5374 PyObject *varnames = NULL;
5375 PyObject *name = NULL;
5376 PyObject *freevars = NULL;
5377 PyObject *cellvars = NULL;
5378 PyObject *bytecode = NULL;
5379 Py_ssize_t nlocals;
5380 int nlocals_int;
5381 int flags;
5382 int argcount, kwonlyargcount, maxdepth;
5383
5384 tmp = dict_keys_inorder(c->u->u_consts, 0);
5385 if (!tmp)
5386 goto error;
5387 consts = PySequence_List(tmp); /* optimize_code requires a list */
5388 Py_DECREF(tmp);
5389
5390 names = dict_keys_inorder(c->u->u_names, 0);
5391 varnames = dict_keys_inorder(c->u->u_varnames, 0);
5392 if (!consts || !names || !varnames)
5393 goto error;
5394
5395 cellvars = dict_keys_inorder(c->u->u_cellvars, 0);
5396 if (!cellvars)
5397 goto error;
5398 freevars = dict_keys_inorder(c->u->u_freevars, PyTuple_Size(cellvars));
5399 if (!freevars)
5400 goto error;
5401
5402 nlocals = PyDict_GET_SIZE(c->u->u_varnames);
5403 assert(nlocals < INT_MAX);
5404 nlocals_int = Py_SAFE_DOWNCAST(nlocals, Py_ssize_t, int);
5405
5406 flags = compute_code_flags(c);
5407 if (flags < 0)
5408 goto error;
5409
5410 bytecode = PyCode_Optimize(a->a_bytecode, consts, names, a->a_lnotab);
5411 if (!bytecode)
5412 goto error;
5413
5414 tmp = PyList_AsTuple(consts); /* PyCode_New requires a tuple */
5415 if (!tmp)
5416 goto error;
5417 Py_DECREF(consts);
5418 consts = tmp;
5419
5420 argcount = Py_SAFE_DOWNCAST(c->u->u_argcount, Py_ssize_t, int);
5421 kwonlyargcount = Py_SAFE_DOWNCAST(c->u->u_kwonlyargcount, Py_ssize_t, int);
5422 maxdepth = stackdepth(c);
5423 if (maxdepth < 0) {
5424 goto error;
5425 }
5426 co = PyCode_New(argcount, kwonlyargcount,
5427 nlocals_int, maxdepth, flags,
5428 bytecode, consts, names, varnames,
5429 freevars, cellvars,
5430 c->c_filename, c->u->u_name,
5431 c->u->u_firstlineno,
5432 a->a_lnotab);
5433 error:
5434 Py_XDECREF(consts);
5435 Py_XDECREF(names);
5436 Py_XDECREF(varnames);
5437 Py_XDECREF(name);
5438 Py_XDECREF(freevars);
5439 Py_XDECREF(cellvars);
5440 Py_XDECREF(bytecode);
5441 return co;
5442 }
5443
5444
5445 /* For debugging purposes only */
5446 #if 0
5447 static void
5448 dump_instr(const struct instr *i)
5449 {
5450 const char *jrel = i->i_jrel ? "jrel " : "";
5451 const char *jabs = i->i_jabs ? "jabs " : "";
5452 char arg[128];
5453
5454 *arg = '\0';
5455 if (HAS_ARG(i->i_opcode)) {
5456 sprintf(arg, "arg: %d ", i->i_oparg);
5457 }
5458 fprintf(stderr, "line: %d, opcode: %d %s%s%s\n",
5459 i->i_lineno, i->i_opcode, arg, jabs, jrel);
5460 }
5461
5462 static void
5463 dump_basicblock(const basicblock *b)
5464 {
5465 const char *seen = b->b_seen ? "seen " : "";
5466 const char *b_return = b->b_return ? "return " : "";
5467 fprintf(stderr, "used: %d, depth: %d, offset: %d %s%s\n",
5468 b->b_iused, b->b_startdepth, b->b_offset, seen, b_return);
5469 if (b->b_instr) {
5470 int i;
5471 for (i = 0; i < b->b_iused; i++) {
5472 fprintf(stderr, " [%02d] ", i);
5473 dump_instr(b->b_instr + i);
5474 }
5475 }
5476 }
5477 #endif
5478
5479 static PyCodeObject *
assemble(struct compiler * c,int addNone)5480 assemble(struct compiler *c, int addNone)
5481 {
5482 basicblock *b, *entryblock;
5483 struct assembler a;
5484 int i, j, nblocks;
5485 PyCodeObject *co = NULL;
5486
5487 /* Make sure every block that falls off the end returns None.
5488 XXX NEXT_BLOCK() isn't quite right, because if the last
5489 block ends with a jump or return b_next shouldn't set.
5490 */
5491 if (!c->u->u_curblock->b_return) {
5492 NEXT_BLOCK(c);
5493 if (addNone)
5494 ADDOP_O(c, LOAD_CONST, Py_None, consts);
5495 ADDOP(c, RETURN_VALUE);
5496 }
5497
5498 nblocks = 0;
5499 entryblock = NULL;
5500 for (b = c->u->u_blocks; b != NULL; b = b->b_list) {
5501 nblocks++;
5502 entryblock = b;
5503 }
5504
5505 /* Set firstlineno if it wasn't explicitly set. */
5506 if (!c->u->u_firstlineno) {
5507 if (entryblock && entryblock->b_instr && entryblock->b_instr->i_lineno)
5508 c->u->u_firstlineno = entryblock->b_instr->i_lineno;
5509 else
5510 c->u->u_firstlineno = 1;
5511 }
5512 if (!assemble_init(&a, nblocks, c->u->u_firstlineno))
5513 goto error;
5514 dfs(c, entryblock, &a, nblocks);
5515
5516 /* Can't modify the bytecode after computing jump offsets. */
5517 assemble_jump_offsets(&a, c);
5518
5519 /* Emit code in reverse postorder from dfs. */
5520 for (i = a.a_nblocks - 1; i >= 0; i--) {
5521 b = a.a_postorder[i];
5522 for (j = 0; j < b->b_iused; j++)
5523 if (!assemble_emit(&a, &b->b_instr[j]))
5524 goto error;
5525 }
5526
5527 if (_PyBytes_Resize(&a.a_lnotab, a.a_lnotab_off) < 0)
5528 goto error;
5529 if (_PyBytes_Resize(&a.a_bytecode, a.a_offset * sizeof(_Py_CODEUNIT)) < 0)
5530 goto error;
5531
5532 co = makecode(c, &a);
5533 error:
5534 assemble_free(&a);
5535 return co;
5536 }
5537
5538 #undef PyAST_Compile
5539 PyAPI_FUNC(PyCodeObject *)
PyAST_Compile(mod_ty mod,const char * filename,PyCompilerFlags * flags,PyArena * arena)5540 PyAST_Compile(mod_ty mod, const char *filename, PyCompilerFlags *flags,
5541 PyArena *arena)
5542 {
5543 return PyAST_CompileEx(mod, filename, flags, -1, arena);
5544 }
5545