cpython/Parser/parser.c

463 lines
12 KiB
C
Raw Normal View History

1991-02-19 20:39:46 +08:00
1990-10-14 20:07:46 +08:00
/* Parser implementation */
/* For a description, see the comments at end of this file */
/* XXX To do: error recovery */
#include "Python.h"
1990-10-14 20:07:46 +08:00
#include "token.h"
#include "grammar.h"
#include "node.h"
#include "parser.h"
#include "errcode.h"
#include "graminit.h"
1990-10-14 20:07:46 +08:00
1996-12-31 00:17:54 +08:00
#ifdef Py_DEBUG
1997-04-30 05:03:06 +08:00
extern int Py_DebugFlag;
#define D(x) if (!Py_DebugFlag); else x
1990-10-14 20:07:46 +08:00
#else
#define D(x)
#endif
/* STACK DATA TYPE */
static void s_reset(stack *);
1990-10-14 20:07:46 +08:00
static void
s_reset(stack *s)
1990-10-14 20:07:46 +08:00
{
s->s_top = &s->s_base[MAXSTACK];
1990-10-14 20:07:46 +08:00
}
#define s_empty(s) ((s)->s_top == &(s)->s_base[MAXSTACK])
static int
s_push(stack *s, const dfa *d, node *parent)
1990-10-14 20:07:46 +08:00
{
stackentry *top;
if (s->s_top == s->s_base) {
fprintf(stderr, "s_push: parser stack overflow\n");
return E_NOMEM;
}
top = --s->s_top;
top->s_dfa = d;
top->s_parent = parent;
top->s_state = 0;
return 0;
1990-10-14 20:07:46 +08:00
}
1996-12-31 00:17:54 +08:00
#ifdef Py_DEBUG
1990-10-14 20:07:46 +08:00
static void
s_pop(stack *s)
1990-10-14 20:07:46 +08:00
{
if (s_empty(s))
Py_FatalError("s_pop: parser stack underflow -- FATAL");
s->s_top++;
1990-10-14 20:07:46 +08:00
}
1996-12-31 00:17:54 +08:00
#else /* !Py_DEBUG */
1990-10-14 20:07:46 +08:00
#define s_pop(s) (s)->s_top++
#endif
/* PARSER CREATION */
parser_state *
PyParser_New(grammar *g, int start)
1990-10-14 20:07:46 +08:00
{
parser_state *ps;
if (!g->g_accel)
PyGrammar_AddAccelerators(g);
ps = (parser_state *)PyMem_MALLOC(sizeof(parser_state));
if (ps == NULL)
return NULL;
ps->p_grammar = g;
#ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
ps->p_flags = 0;
#endif
ps->p_tree = PyNode_New(start);
if (ps->p_tree == NULL) {
PyMem_FREE(ps);
return NULL;
}
s_reset(&ps->p_stack);
(void) s_push(&ps->p_stack, PyGrammar_FindDFA(g, start), ps->p_tree);
return ps;
1990-10-14 20:07:46 +08:00
}
void
PyParser_Delete(parser_state *ps)
1990-10-14 20:07:46 +08:00
{
/* NB If you want to save the parse tree,
you must set p_tree to NULL before calling delparser! */
PyNode_Free(ps->p_tree);
PyMem_FREE(ps);
1990-10-14 20:07:46 +08:00
}
/* PARSER STACK OPERATIONS */
static int
bpo-33416: Add end positions to Python AST (GH-11605) The majority of this PR is tediously passing `end_lineno` and `end_col_offset` everywhere. Here are non-trivial points: * It is not possible to reconstruct end positions in AST "on the fly", some information is lost after an AST node is constructed, so we need two more attributes for every AST node `end_lineno` and `end_col_offset`. * I add end position information to both CST and AST. Although it may be technically possible to avoid adding end positions to CST, the code becomes more cumbersome and less efficient. * Since the end position is not known for non-leaf CST nodes while the next token is added, this requires a bit of extra care (see `_PyNode_FinalizeEndPos`). Unless I made some mistake, the algorithm should be linear. * For statements, I "trim" the end position of suites to not include the terminal newlines and dedent (this seems to be what people would expect), for example in ```python class C: pass pass ``` the end line and end column for the class definition is (2, 8). * For `end_col_offset` I use the common Python convention for indexing, for example for `pass` the `end_col_offset` is 4 (not 3), so that `[0:4]` gives one the source code that corresponds to the node. * I added a helper function `ast.get_source_segment()`, to get source text segment corresponding to a given AST node. It is also useful for testing. An (inevitable) downside of this PR is that AST now takes almost 25% more memory. I think however it is probably justified by the benefits.
2019-01-22 19:18:22 +08:00
shift(stack *s, int type, char *str, int newstate, int lineno, int col_offset,
int end_lineno, int end_col_offset)
1990-10-14 20:07:46 +08:00
{
int err;
assert(!s_empty(s));
bpo-33416: Add end positions to Python AST (GH-11605) The majority of this PR is tediously passing `end_lineno` and `end_col_offset` everywhere. Here are non-trivial points: * It is not possible to reconstruct end positions in AST "on the fly", some information is lost after an AST node is constructed, so we need two more attributes for every AST node `end_lineno` and `end_col_offset`. * I add end position information to both CST and AST. Although it may be technically possible to avoid adding end positions to CST, the code becomes more cumbersome and less efficient. * Since the end position is not known for non-leaf CST nodes while the next token is added, this requires a bit of extra care (see `_PyNode_FinalizeEndPos`). Unless I made some mistake, the algorithm should be linear. * For statements, I "trim" the end position of suites to not include the terminal newlines and dedent (this seems to be what people would expect), for example in ```python class C: pass pass ``` the end line and end column for the class definition is (2, 8). * For `end_col_offset` I use the common Python convention for indexing, for example for `pass` the `end_col_offset` is 4 (not 3), so that `[0:4]` gives one the source code that corresponds to the node. * I added a helper function `ast.get_source_segment()`, to get source text segment corresponding to a given AST node. It is also useful for testing. An (inevitable) downside of this PR is that AST now takes almost 25% more memory. I think however it is probably justified by the benefits.
2019-01-22 19:18:22 +08:00
err = PyNode_AddChild(s->s_top->s_parent, type, str, lineno, col_offset,
end_lineno, end_col_offset);
if (err)
return err;
s->s_top->s_state = newstate;
return 0;
1990-10-14 20:07:46 +08:00
}
static int
push(stack *s, int type, const dfa *d, int newstate, int lineno, int col_offset,
bpo-33416: Add end positions to Python AST (GH-11605) The majority of this PR is tediously passing `end_lineno` and `end_col_offset` everywhere. Here are non-trivial points: * It is not possible to reconstruct end positions in AST "on the fly", some information is lost after an AST node is constructed, so we need two more attributes for every AST node `end_lineno` and `end_col_offset`. * I add end position information to both CST and AST. Although it may be technically possible to avoid adding end positions to CST, the code becomes more cumbersome and less efficient. * Since the end position is not known for non-leaf CST nodes while the next token is added, this requires a bit of extra care (see `_PyNode_FinalizeEndPos`). Unless I made some mistake, the algorithm should be linear. * For statements, I "trim" the end position of suites to not include the terminal newlines and dedent (this seems to be what people would expect), for example in ```python class C: pass pass ``` the end line and end column for the class definition is (2, 8). * For `end_col_offset` I use the common Python convention for indexing, for example for `pass` the `end_col_offset` is 4 (not 3), so that `[0:4]` gives one the source code that corresponds to the node. * I added a helper function `ast.get_source_segment()`, to get source text segment corresponding to a given AST node. It is also useful for testing. An (inevitable) downside of this PR is that AST now takes almost 25% more memory. I think however it is probably justified by the benefits.
2019-01-22 19:18:22 +08:00
int end_lineno, int end_col_offset)
1990-10-14 20:07:46 +08:00
{
int err;
node *n;
n = s->s_top->s_parent;
assert(!s_empty(s));
bpo-33416: Add end positions to Python AST (GH-11605) The majority of this PR is tediously passing `end_lineno` and `end_col_offset` everywhere. Here are non-trivial points: * It is not possible to reconstruct end positions in AST "on the fly", some information is lost after an AST node is constructed, so we need two more attributes for every AST node `end_lineno` and `end_col_offset`. * I add end position information to both CST and AST. Although it may be technically possible to avoid adding end positions to CST, the code becomes more cumbersome and less efficient. * Since the end position is not known for non-leaf CST nodes while the next token is added, this requires a bit of extra care (see `_PyNode_FinalizeEndPos`). Unless I made some mistake, the algorithm should be linear. * For statements, I "trim" the end position of suites to not include the terminal newlines and dedent (this seems to be what people would expect), for example in ```python class C: pass pass ``` the end line and end column for the class definition is (2, 8). * For `end_col_offset` I use the common Python convention for indexing, for example for `pass` the `end_col_offset` is 4 (not 3), so that `[0:4]` gives one the source code that corresponds to the node. * I added a helper function `ast.get_source_segment()`, to get source text segment corresponding to a given AST node. It is also useful for testing. An (inevitable) downside of this PR is that AST now takes almost 25% more memory. I think however it is probably justified by the benefits.
2019-01-22 19:18:22 +08:00
err = PyNode_AddChild(n, type, (char *)NULL, lineno, col_offset,
end_lineno, end_col_offset);
if (err)
return err;
s->s_top->s_state = newstate;
return s_push(s, d, CHILD(n, NCH(n)-1));
1990-10-14 20:07:46 +08:00
}
/* PARSER PROPER */
static int
classify(parser_state *ps, int type, const char *str)
1990-10-14 20:07:46 +08:00
{
grammar *g = ps->p_grammar;
int n = g->g_ll.ll_nlabels;
if (type == NAME) {
const label *l = g->g_ll.ll_label;
int i;
for (i = n; i > 0; i--, l++) {
if (l->lb_type != NAME || l->lb_str == NULL ||
l->lb_str[0] != str[0] ||
strcmp(l->lb_str, str) != 0)
continue;
#ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
#if 0
/* Leaving this in as an example */
if (!(ps->p_flags & CO_FUTURE_WITH_STATEMENT)) {
if (str[0] == 'w' && strcmp(str, "with") == 0)
break; /* not a keyword yet */
else if (str[0] == 'a' && strcmp(str, "as") == 0)
break; /* not a keyword yet */
}
#endif
#endif
D(printf("It's a keyword\n"));
return n - i;
}
}
{
const label *l = g->g_ll.ll_label;
int i;
for (i = n; i > 0; i--, l++) {
if (l->lb_type == type && l->lb_str == NULL) {
D(printf("It's a token we know\n"));
return n - i;
}
}
}
D(printf("Illegal token\n"));
return -1;
1990-10-14 20:07:46 +08:00
}
#ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
#if 0
/* Leaving this in as an example */
static void
future_hack(parser_state *ps)
{
node *n = ps->p_stack.s_top->s_parent;
node *ch, *cch;
int i;
/* from __future__ import ..., must have at least 4 children */
n = CHILD(n, 0);
if (NCH(n) < 4)
return;
ch = CHILD(n, 0);
if (STR(ch) == NULL || strcmp(STR(ch), "from") != 0)
return;
ch = CHILD(n, 1);
if (NCH(ch) == 1 && STR(CHILD(ch, 0)) &&
strcmp(STR(CHILD(ch, 0)), "__future__") != 0)
return;
ch = CHILD(n, 3);
/* ch can be a star, a parenthesis or import_as_names */
if (TYPE(ch) == STAR)
return;
if (TYPE(ch) == LPAR)
ch = CHILD(n, 4);
for (i = 0; i < NCH(ch); i += 2) {
cch = CHILD(ch, i);
if (NCH(cch) >= 1 && TYPE(CHILD(cch, 0)) == NAME) {
char *str_ch = STR(CHILD(cch, 0));
if (strcmp(str_ch, FUTURE_WITH_STATEMENT) == 0) {
ps->p_flags |= CO_FUTURE_WITH_STATEMENT;
} else if (strcmp(str_ch, FUTURE_PRINT_FUNCTION) == 0) {
ps->p_flags |= CO_FUTURE_PRINT_FUNCTION;
} else if (strcmp(str_ch, FUTURE_UNICODE_LITERALS) == 0) {
ps->p_flags |= CO_FUTURE_UNICODE_LITERALS;
}
}
}
}
#endif
#endif /* future keyword */
1990-10-14 20:07:46 +08:00
int
PyParser_AddToken(parser_state *ps, int type, char *str,
bpo-33416: Add end positions to Python AST (GH-11605) The majority of this PR is tediously passing `end_lineno` and `end_col_offset` everywhere. Here are non-trivial points: * It is not possible to reconstruct end positions in AST "on the fly", some information is lost after an AST node is constructed, so we need two more attributes for every AST node `end_lineno` and `end_col_offset`. * I add end position information to both CST and AST. Although it may be technically possible to avoid adding end positions to CST, the code becomes more cumbersome and less efficient. * Since the end position is not known for non-leaf CST nodes while the next token is added, this requires a bit of extra care (see `_PyNode_FinalizeEndPos`). Unless I made some mistake, the algorithm should be linear. * For statements, I "trim" the end position of suites to not include the terminal newlines and dedent (this seems to be what people would expect), for example in ```python class C: pass pass ``` the end line and end column for the class definition is (2, 8). * For `end_col_offset` I use the common Python convention for indexing, for example for `pass` the `end_col_offset` is 4 (not 3), so that `[0:4]` gives one the source code that corresponds to the node. * I added a helper function `ast.get_source_segment()`, to get source text segment corresponding to a given AST node. It is also useful for testing. An (inevitable) downside of this PR is that AST now takes almost 25% more memory. I think however it is probably justified by the benefits.
2019-01-22 19:18:22 +08:00
int lineno, int col_offset,
int end_lineno, int end_col_offset,
int *expected_ret)
1990-10-14 20:07:46 +08:00
{
int ilabel;
int err;
D(printf("Token %s/'%s' ... ", _PyParser_TokenNames[type], str));
/* Find out which label this token is */
ilabel = classify(ps, type, str);
if (ilabel < 0)
return E_SYNTAX;
/* Loop until the token is shifted or an error occurred */
for (;;) {
/* Fetch the current dfa and state */
const dfa *d = ps->p_stack.s_top->s_dfa;
state *s = &d->d_state[ps->p_stack.s_top->s_state];
D(printf(" DFA '%s', state %d:",
d->d_name, ps->p_stack.s_top->s_state));
/* Check accelerator */
if (s->s_lower <= ilabel && ilabel < s->s_upper) {
int x = s->s_accel[ilabel - s->s_lower];
if (x != -1) {
if (x & (1<<7)) {
/* Push non-terminal */
int nt = (x >> 8) + NT_OFFSET;
int arrow = x & ((1<<7)-1);
if (nt == func_body_suite && !(ps->p_flags & PyCF_TYPE_COMMENTS)) {
/* When parsing type comments is not requested,
we can provide better errors about bad indentation
by using 'suite' for the body of a funcdef */
D(printf(" [switch func_body_suite to suite]"));
nt = suite;
}
const dfa *d1 = PyGrammar_FindDFA(
ps->p_grammar, nt);
if ((err = push(&ps->p_stack, nt, d1,
bpo-33416: Add end positions to Python AST (GH-11605) The majority of this PR is tediously passing `end_lineno` and `end_col_offset` everywhere. Here are non-trivial points: * It is not possible to reconstruct end positions in AST "on the fly", some information is lost after an AST node is constructed, so we need two more attributes for every AST node `end_lineno` and `end_col_offset`. * I add end position information to both CST and AST. Although it may be technically possible to avoid adding end positions to CST, the code becomes more cumbersome and less efficient. * Since the end position is not known for non-leaf CST nodes while the next token is added, this requires a bit of extra care (see `_PyNode_FinalizeEndPos`). Unless I made some mistake, the algorithm should be linear. * For statements, I "trim" the end position of suites to not include the terminal newlines and dedent (this seems to be what people would expect), for example in ```python class C: pass pass ``` the end line and end column for the class definition is (2, 8). * For `end_col_offset` I use the common Python convention for indexing, for example for `pass` the `end_col_offset` is 4 (not 3), so that `[0:4]` gives one the source code that corresponds to the node. * I added a helper function `ast.get_source_segment()`, to get source text segment corresponding to a given AST node. It is also useful for testing. An (inevitable) downside of this PR is that AST now takes almost 25% more memory. I think however it is probably justified by the benefits.
2019-01-22 19:18:22 +08:00
arrow, lineno, col_offset,
end_lineno, end_col_offset)) > 0) {
D(printf(" MemError: push\n"));
return err;
}
D(printf(" Push '%s'\n", d1->d_name));
continue;
}
/* Shift the token */
if ((err = shift(&ps->p_stack, type, str,
bpo-33416: Add end positions to Python AST (GH-11605) The majority of this PR is tediously passing `end_lineno` and `end_col_offset` everywhere. Here are non-trivial points: * It is not possible to reconstruct end positions in AST "on the fly", some information is lost after an AST node is constructed, so we need two more attributes for every AST node `end_lineno` and `end_col_offset`. * I add end position information to both CST and AST. Although it may be technically possible to avoid adding end positions to CST, the code becomes more cumbersome and less efficient. * Since the end position is not known for non-leaf CST nodes while the next token is added, this requires a bit of extra care (see `_PyNode_FinalizeEndPos`). Unless I made some mistake, the algorithm should be linear. * For statements, I "trim" the end position of suites to not include the terminal newlines and dedent (this seems to be what people would expect), for example in ```python class C: pass pass ``` the end line and end column for the class definition is (2, 8). * For `end_col_offset` I use the common Python convention for indexing, for example for `pass` the `end_col_offset` is 4 (not 3), so that `[0:4]` gives one the source code that corresponds to the node. * I added a helper function `ast.get_source_segment()`, to get source text segment corresponding to a given AST node. It is also useful for testing. An (inevitable) downside of this PR is that AST now takes almost 25% more memory. I think however it is probably justified by the benefits.
2019-01-22 19:18:22 +08:00
x, lineno, col_offset,
end_lineno, end_col_offset)) > 0) {
D(printf(" MemError: shift.\n"));
return err;
}
D(printf(" Shift.\n"));
/* Pop while we are in an accept-only state */
while (s = &d->d_state
[ps->p_stack.s_top->s_state],
s->s_accept && s->s_narcs == 1) {
D(printf(" DFA '%s', state %d: "
"Direct pop.\n",
d->d_name,
ps->p_stack.s_top->s_state));
#ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
#if 0
if (d->d_name[0] == 'i' &&
strcmp(d->d_name,
"import_stmt") == 0)
future_hack(ps);
#endif
#endif
s_pop(&ps->p_stack);
if (s_empty(&ps->p_stack)) {
D(printf(" ACCEPT.\n"));
return E_DONE;
}
d = ps->p_stack.s_top->s_dfa;
}
return E_OK;
}
}
if (s->s_accept) {
#ifdef PY_PARSER_REQUIRES_FUTURE_KEYWORD
#if 0
if (d->d_name[0] == 'i' &&
strcmp(d->d_name, "import_stmt") == 0)
future_hack(ps);
#endif
#endif
/* Pop this dfa and try again */
s_pop(&ps->p_stack);
D(printf(" Pop ...\n"));
if (s_empty(&ps->p_stack)) {
D(printf(" Error: bottom of stack.\n"));
return E_SYNTAX;
}
continue;
}
/* Stuck, report syntax error */
D(printf(" Error.\n"));
if (expected_ret) {
if (s->s_lower == s->s_upper - 1) {
/* Only one possible expected token */
*expected_ret = ps->p_grammar->
g_ll.ll_label[s->s_lower].lb_type;
}
else
*expected_ret = -1;
}
return E_SYNTAX;
}
1990-10-14 20:07:46 +08:00
}
1996-12-31 00:17:54 +08:00
#ifdef Py_DEBUG
1990-10-14 20:07:46 +08:00
/* DEBUG OUTPUT */
void
dumptree(grammar *g, node *n)
1990-10-14 20:07:46 +08:00
{
int i;
if (n == NULL)
printf("NIL");
else {
label l;
l.lb_type = TYPE(n);
l.lb_str = STR(n);
printf("%s", PyGrammar_LabelRepr(&l));
if (ISNONTERMINAL(TYPE(n))) {
printf("(");
for (i = 0; i < NCH(n); i++) {
if (i > 0)
printf(",");
dumptree(g, CHILD(n, i));
}
printf(")");
}
}
1990-10-14 20:07:46 +08:00
}
void
showtree(grammar *g, node *n)
1990-10-14 20:07:46 +08:00
{
int i;
if (n == NULL)
return;
if (ISNONTERMINAL(TYPE(n))) {
for (i = 0; i < NCH(n); i++)
showtree(g, CHILD(n, i));
}
else if (ISTERMINAL(TYPE(n))) {
printf("%s", _PyParser_TokenNames[TYPE(n)]);
if (TYPE(n) == NUMBER || TYPE(n) == NAME)
printf("(%s)", STR(n));
printf(" ");
}
else
printf("? ");
1990-10-14 20:07:46 +08:00
}
void
printtree(parser_state *ps)
1990-10-14 20:07:46 +08:00
{
if (Py_DebugFlag) {
printf("Parse tree:\n");
dumptree(ps->p_grammar, ps->p_tree);
printf("\n");
printf("Tokens:\n");
showtree(ps->p_grammar, ps->p_tree);
printf("\n");
}
printf("Listing:\n");
PyNode_ListTree(ps->p_tree);
printf("\n");
1990-10-14 20:07:46 +08:00
}
1996-12-31 00:17:54 +08:00
#endif /* Py_DEBUG */
1990-10-14 20:07:46 +08:00
/*
Description
-----------
The parser's interface is different than usual: the function addtoken()
must be called for each token in the input. This makes it possible to
turn it into an incremental parsing system later. The parsing system
constructs a parse tree as it goes.
A parsing rule is represented as a Deterministic Finite-state Automaton
(DFA). A node in a DFA represents a state of the parser; an arc represents
a transition. Transitions are either labeled with terminal symbols or
with non-terminals. When the parser decides to follow an arc labeled
with a non-terminal, it is invoked recursively with the DFA representing
the parsing rule for that as its initial state; when that DFA accepts,
the parser that invoked it continues. The parse tree constructed by the
recursively called parser is inserted as a child in the current parse tree.
The DFA's can be constructed automatically from a more conventional
language description. An extended LL(1) grammar (ELL(1)) is suitable.
Certain restrictions make the parser's life easier: rules that can produce
the empty string should be outlawed (there are other ways to put loops
or optional parts in the language). To avoid the need to construct
FIRST sets, we can require that all but the last alternative of a rule
(really: arc going out of a DFA's state) must begin with a terminal
symbol.
As an example, consider this grammar:
expr: term (OP term)*
term: CONSTANT | '(' expr ')'
1990-10-14 20:07:46 +08:00
The DFA corresponding to the rule for expr is:
------->.---term-->.------->
^ |
| |
\----OP----/
1990-10-14 20:07:46 +08:00
The parse tree generated for the input a+b is:
(expr: (term: (NAME: a)), (OP: +), (term: (NAME: b)))
*/