2 /* Parser implementation */
4 /* For a description, see the comments at end of this file */
6 /* XXX To do: error recovery */
9 #include "pgenheaders.h"
18 extern int Py_DebugFlag
;
19 #define D(x) if (!Py_DebugFlag); else x
27 static void s_reset(stack
*);
32 s
->s_top
= &s
->s_base
[MAXSTACK
];
35 #define s_empty(s) ((s)->s_top == &(s)->s_base[MAXSTACK])
38 s_push(register stack
*s
, dfa
*d
, node
*parent
)
40 register stackentry
*top
;
41 if (s
->s_top
== s
->s_base
) {
42 fprintf(stderr
, "s_push: parser stack overflow\n");
47 top
->s_parent
= parent
;
55 s_pop(register stack
*s
)
58 Py_FatalError("s_pop: parser stack underflow -- FATAL");
64 #define s_pop(s) (s)->s_top++
72 PyParser_New(grammar
*g
, int start
)
77 PyGrammar_AddAccelerators(g
);
78 ps
= (parser_state
*)PyMem_MALLOC(sizeof(parser_state
));
82 #ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
85 ps
->p_tree
= PyNode_New(start
);
86 if (ps
->p_tree
== NULL
) {
90 s_reset(&ps
->p_stack
);
91 (void) s_push(&ps
->p_stack
, PyGrammar_FindDFA(g
, start
), ps
->p_tree
);
96 PyParser_Delete(parser_state
*ps
)
98 /* NB If you want to save the parse tree,
99 you must set p_tree to NULL before calling delparser! */
100 PyNode_Free(ps
->p_tree
);
105 /* PARSER STACK OPERATIONS */
108 shift(register stack
*s
, int type
, char *str
, int newstate
, int lineno
, int col_offset
)
112 err
= PyNode_AddChild(s
->s_top
->s_parent
, type
, str
, lineno
, col_offset
);
115 s
->s_top
->s_state
= newstate
;
120 push(register stack
*s
, int type
, dfa
*d
, int newstate
, int lineno
, int col_offset
)
124 n
= s
->s_top
->s_parent
;
126 err
= PyNode_AddChild(n
, type
, (char *)NULL
, lineno
, col_offset
);
129 s
->s_top
->s_state
= newstate
;
130 return s_push(s
, d
, CHILD(n
, NCH(n
)-1));
137 classify(parser_state
*ps
, int type
, char *str
)
139 grammar
*g
= ps
->p_grammar
;
140 register int n
= g
->g_ll
.ll_nlabels
;
143 register char *s
= str
;
144 register label
*l
= g
->g_ll
.ll_label
;
146 for (i
= n
; i
> 0; i
--, l
++) {
147 if (l
->lb_type
!= NAME
|| l
->lb_str
== NULL
||
148 l
->lb_str
[0] != s
[0] ||
149 strcmp(l
->lb_str
, s
) != 0)
151 #ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
152 if (!(ps
->p_flags
& CO_FUTURE_WITH_STATEMENT
)) {
153 if (s
[0] == 'w' && strcmp(s
, "with") == 0)
154 break; /* not a keyword yet */
155 else if (s
[0] == 'a' && strcmp(s
, "as") == 0)
156 break; /* not a keyword yet */
159 D(printf("It's a keyword\n"));
165 register label
*l
= g
->g_ll
.ll_label
;
167 for (i
= n
; i
> 0; i
--, l
++) {
168 if (l
->lb_type
== type
&& l
->lb_str
== NULL
) {
169 D(printf("It's a token we know\n"));
175 D(printf("Illegal token\n"));
179 #ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
181 future_hack(parser_state
*ps
)
183 node
*n
= ps
->p_stack
.s_top
->s_parent
;
187 /* from __future__ import ..., must have at least 4 children */
192 if (STR(ch
) == NULL
|| strcmp(STR(ch
), "from") != 0)
195 if (NCH(ch
) == 1 && STR(CHILD(ch
, 0)) &&
196 strcmp(STR(CHILD(ch
, 0)), "__future__") != 0)
198 for (i
= 3; i
< NCH(n
); i
+= 2) {
199 /* XXX: assume we don't have parentheses in import:
200 from __future__ import (x, y, z)
205 if (NCH(ch
) >= 1 && TYPE(CHILD(ch
, 0)) == NAME
&&
206 strcmp(STR(CHILD(ch
, 0)), "with_statement") == 0) {
207 ps
->p_flags
|= CO_FUTURE_WITH_STATEMENT
;
212 #endif /* future keyword */
215 PyParser_AddToken(register parser_state
*ps
, register int type
, char *str
,
216 int lineno
, int col_offset
, int *expected_ret
)
221 D(printf("Token %s/'%s' ... ", _PyParser_TokenNames
[type
], str
));
223 /* Find out which label this token is */
224 ilabel
= classify(ps
, type
, str
);
228 /* Loop until the token is shifted or an error occurred */
230 /* Fetch the current dfa and state */
231 register dfa
*d
= ps
->p_stack
.s_top
->s_dfa
;
232 register state
*s
= &d
->d_state
[ps
->p_stack
.s_top
->s_state
];
234 D(printf(" DFA '%s', state %d:",
235 d
->d_name
, ps
->p_stack
.s_top
->s_state
));
237 /* Check accelerator */
238 if (s
->s_lower
<= ilabel
&& ilabel
< s
->s_upper
) {
239 register int x
= s
->s_accel
[ilabel
- s
->s_lower
];
242 /* Push non-terminal */
243 int nt
= (x
>> 8) + NT_OFFSET
;
244 int arrow
= x
& ((1<<7)-1);
245 dfa
*d1
= PyGrammar_FindDFA(
247 if ((err
= push(&ps
->p_stack
, nt
, d1
,
248 arrow
, lineno
, col_offset
)) > 0) {
249 D(printf(" MemError: push\n"));
252 D(printf(" Push ...\n"));
256 /* Shift the token */
257 if ((err
= shift(&ps
->p_stack
, type
, str
,
258 x
, lineno
, col_offset
)) > 0) {
259 D(printf(" MemError: shift.\n"));
262 D(printf(" Shift.\n"));
263 /* Pop while we are in an accept-only state */
264 while (s
= &d
->d_state
265 [ps
->p_stack
.s_top
->s_state
],
266 s
->s_accept
&& s
->s_narcs
== 1) {
267 D(printf(" DFA '%s', state %d: "
270 ps
->p_stack
.s_top
->s_state
));
271 #ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
272 if (d
->d_name
[0] == 'i' &&
278 if (s_empty(&ps
->p_stack
)) {
279 D(printf(" ACCEPT.\n"));
282 d
= ps
->p_stack
.s_top
->s_dfa
;
289 #ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
290 if (d
->d_name
[0] == 'i' &&
291 strcmp(d
->d_name
, "import_stmt") == 0)
294 /* Pop this dfa and try again */
296 D(printf(" Pop ...\n"));
297 if (s_empty(&ps
->p_stack
)) {
298 D(printf(" Error: bottom of stack.\n"));
304 /* Stuck, report syntax error */
305 D(printf(" Error.\n"));
307 if (s
->s_lower
== s
->s_upper
- 1) {
308 /* Only one possible expected token */
309 *expected_ret
= ps
->p_grammar
->
310 g_ll
.ll_label
[s
->s_lower
].lb_type
;
325 dumptree(grammar
*g
, node
*n
)
335 printf("%s", PyGrammar_LabelRepr(&l
));
336 if (ISNONTERMINAL(TYPE(n
))) {
338 for (i
= 0; i
< NCH(n
); i
++) {
341 dumptree(g
, CHILD(n
, i
));
349 showtree(grammar
*g
, node
*n
)
355 if (ISNONTERMINAL(TYPE(n
))) {
356 for (i
= 0; i
< NCH(n
); i
++)
357 showtree(g
, CHILD(n
, i
));
359 else if (ISTERMINAL(TYPE(n
))) {
360 printf("%s", _PyParser_TokenNames
[TYPE(n
)]);
361 if (TYPE(n
) == NUMBER
|| TYPE(n
) == NAME
)
362 printf("(%s)", STR(n
));
370 printtree(parser_state
*ps
)
373 printf("Parse tree:\n");
374 dumptree(ps
->p_grammar
, ps
->p_tree
);
377 showtree(ps
->p_grammar
, ps
->p_tree
);
380 printf("Listing:\n");
381 PyNode_ListTree(ps
->p_tree
);
385 #endif /* Py_DEBUG */
392 The parser's interface is different than usual: the function addtoken()
393 must be called for each token in the input. This makes it possible to
394 turn it into an incremental parsing system later. The parsing system
395 constructs a parse tree as it goes.
397 A parsing rule is represented as a Deterministic Finite-state Automaton
398 (DFA). A node in a DFA represents a state of the parser; an arc represents
399 a transition. Transitions are either labeled with terminal symbols or
400 with non-terminals. When the parser decides to follow an arc labeled
401 with a non-terminal, it is invoked recursively with the DFA representing
402 the parsing rule for that as its initial state; when that DFA accepts,
403 the parser that invoked it continues. The parse tree constructed by the
404 recursively called parser is inserted as a child in the current parse tree.
406 The DFA's can be constructed automatically from a more conventional
407 language description. An extended LL(1) grammar (ELL(1)) is suitable.
408 Certain restrictions make the parser's life easier: rules that can produce
409 the empty string should be outlawed (there are other ways to put loops
410 or optional parts in the language). To avoid the need to construct
411 FIRST sets, we can require that all but the last alternative of a rule
412 (really: arc going out of a DFA's state) must begin with a terminal
415 As an example, consider this grammar:
417 expr: term (OP term)*
418 term: CONSTANT | '(' expr ')'
420 The DFA corresponding to the rule for expr is:
422 ------->.---term-->.------->
427 The parse tree generated for the input a+b is:
429 (expr: (term: (NAME: a)), (OP: +), (term: (NAME: b)))