1 | /* Parse tree node implementation */
|
---|
2 |
|
---|
3 | #include "Python.h"
|
---|
4 | #include "node.h"
|
---|
5 | #include "errcode.h"
|
---|
6 |
|
---|
7 | node *
|
---|
8 | PyNode_New(int type)
|
---|
9 | {
|
---|
10 | node *n = (node *) PyObject_MALLOC(1 * sizeof(node));
|
---|
11 | if (n == NULL)
|
---|
12 | return NULL;
|
---|
13 | n->n_type = type;
|
---|
14 | n->n_str = NULL;
|
---|
15 | n->n_lineno = 0;
|
---|
16 | n->n_nchildren = 0;
|
---|
17 | n->n_child = NULL;
|
---|
18 | return n;
|
---|
19 | }
|
---|
20 |
|
---|
21 | /* See comments at XXXROUNDUP below. Returns -1 on overflow. */
|
---|
22 | static int
|
---|
23 | fancy_roundup(int n)
|
---|
24 | {
|
---|
25 | /* Round up to the closest power of 2 >= n. */
|
---|
26 | int result = 256;
|
---|
27 | assert(n > 128);
|
---|
28 | while (result < n) {
|
---|
29 | result <<= 1;
|
---|
30 | if (result <= 0)
|
---|
31 | return -1;
|
---|
32 | }
|
---|
33 | return result;
|
---|
34 | }
|
---|
35 |
|
---|
36 | /* A gimmick to make massive numbers of reallocs quicker. The result is
|
---|
37 | * a number >= the input. In PyNode_AddChild, it's used like so, when
|
---|
38 | * we're about to add child number current_size + 1:
|
---|
39 | *
|
---|
40 | * if XXXROUNDUP(current_size) < XXXROUNDUP(current_size + 1):
|
---|
41 | * allocate space for XXXROUNDUP(current_size + 1) total children
|
---|
42 | * else:
|
---|
43 | * we already have enough space
|
---|
44 | *
|
---|
45 | * Since a node starts out empty, we must have
|
---|
46 | *
|
---|
47 | * XXXROUNDUP(0) < XXXROUNDUP(1)
|
---|
48 | *
|
---|
49 | * so that we allocate space for the first child. One-child nodes are very
|
---|
50 | * common (presumably that would change if we used a more abstract form
|
---|
51 | * of syntax tree), so to avoid wasting memory it's desirable that
|
---|
52 | * XXXROUNDUP(1) == 1. That in turn forces XXXROUNDUP(0) == 0.
|
---|
53 | *
|
---|
54 | * Else for 2 <= n <= 128, we round up to the closest multiple of 4. Why 4?
|
---|
55 | * Rounding up to a multiple of an exact power of 2 is very efficient, and
|
---|
56 | * most nodes with more than one child have <= 4 kids.
|
---|
57 | *
|
---|
58 | * Else we call fancy_roundup() to grow proportionately to n. We've got an
|
---|
59 | * extreme case then (like test_longexp.py), and on many platforms doing
|
---|
60 | * anything less than proportional growth leads to exorbitant runtime
|
---|
61 | * (e.g., MacPython), or extreme fragmentation of user address space (e.g.,
|
---|
62 | * Win98).
|
---|
63 | *
|
---|
64 | * In a run of compileall across the 2.3a0 Lib directory, Andrew MacIntyre
|
---|
65 | * reported that, with this scheme, 89% of PyObject_REALLOC calls in
|
---|
66 | * PyNode_AddChild passed 1 for the size, and 9% passed 4. So this usually
|
---|
67 | * wastes very little memory, but is very effective at sidestepping
|
---|
68 | * platform-realloc disasters on vulnerable platforms.
|
---|
69 | *
|
---|
70 | * Note that this would be straightforward if a node stored its current
|
---|
71 | * capacity. The code is tricky to avoid that.
|
---|
72 | */
|
---|
73 | #define XXXROUNDUP(n) ((n) <= 1 ? (n) : \
|
---|
74 | (n) <= 128 ? (((n) + 3) & ~3) : \
|
---|
75 | fancy_roundup(n))
|
---|
76 |
|
---|
77 |
|
---|
78 | int
|
---|
79 | PyNode_AddChild(register node *n1, int type, char *str, int lineno, int col_offset)
|
---|
80 | {
|
---|
81 | const int nch = n1->n_nchildren;
|
---|
82 | int current_capacity;
|
---|
83 | int required_capacity;
|
---|
84 | node *n;
|
---|
85 |
|
---|
86 | if (nch == INT_MAX || nch < 0)
|
---|
87 | return E_OVERFLOW;
|
---|
88 |
|
---|
89 | current_capacity = XXXROUNDUP(nch);
|
---|
90 | required_capacity = XXXROUNDUP(nch + 1);
|
---|
91 | if (current_capacity < 0 || required_capacity < 0)
|
---|
92 | return E_OVERFLOW;
|
---|
93 | if (current_capacity < required_capacity) {
|
---|
94 | if (required_capacity > PY_SIZE_MAX / sizeof(node)) {
|
---|
95 | return E_NOMEM;
|
---|
96 | }
|
---|
97 | n = n1->n_child;
|
---|
98 | n = (node *) PyObject_REALLOC(n,
|
---|
99 | required_capacity * sizeof(node));
|
---|
100 | if (n == NULL)
|
---|
101 | return E_NOMEM;
|
---|
102 | n1->n_child = n;
|
---|
103 | }
|
---|
104 |
|
---|
105 | n = &n1->n_child[n1->n_nchildren++];
|
---|
106 | n->n_type = type;
|
---|
107 | n->n_str = str;
|
---|
108 | n->n_lineno = lineno;
|
---|
109 | n->n_col_offset = col_offset;
|
---|
110 | n->n_nchildren = 0;
|
---|
111 | n->n_child = NULL;
|
---|
112 | return 0;
|
---|
113 | }
|
---|
114 |
|
---|
115 | /* Forward */
|
---|
116 | static void freechildren(node *);
|
---|
117 | static Py_ssize_t sizeofchildren(node *n);
|
---|
118 |
|
---|
119 |
|
---|
120 | void
|
---|
121 | PyNode_Free(node *n)
|
---|
122 | {
|
---|
123 | if (n != NULL) {
|
---|
124 | freechildren(n);
|
---|
125 | PyObject_FREE(n);
|
---|
126 | }
|
---|
127 | }
|
---|
128 |
|
---|
129 | Py_ssize_t
|
---|
130 | _PyNode_SizeOf(node *n)
|
---|
131 | {
|
---|
132 | Py_ssize_t res = 0;
|
---|
133 |
|
---|
134 | if (n != NULL)
|
---|
135 | res = sizeof(node) + sizeofchildren(n);
|
---|
136 | return res;
|
---|
137 | }
|
---|
138 |
|
---|
139 | static void
|
---|
140 | freechildren(node *n)
|
---|
141 | {
|
---|
142 | int i;
|
---|
143 | for (i = NCH(n); --i >= 0; )
|
---|
144 | freechildren(CHILD(n, i));
|
---|
145 | if (n->n_child != NULL)
|
---|
146 | PyObject_FREE(n->n_child);
|
---|
147 | if (STR(n) != NULL)
|
---|
148 | PyObject_FREE(STR(n));
|
---|
149 | }
|
---|
150 |
|
---|
151 | static Py_ssize_t
|
---|
152 | sizeofchildren(node *n)
|
---|
153 | {
|
---|
154 | Py_ssize_t res = 0;
|
---|
155 | int i;
|
---|
156 | for (i = NCH(n); --i >= 0; )
|
---|
157 | res += sizeofchildren(CHILD(n, i));
|
---|
158 | if (n->n_child != NULL)
|
---|
159 | /* allocated size of n->n_child array */
|
---|
160 | res += XXXROUNDUP(NCH(n)) * sizeof(node);
|
---|
161 | if (STR(n) != NULL)
|
---|
162 | res += strlen(STR(n)) + 1;
|
---|
163 | return res;
|
---|
164 | }
|
---|