| … | |
… | |
| 3 | */ |
3 | */ |
| 4 | |
4 | |
| 5 | #include <cstring> |
5 | #include <cstring> |
| 6 | #include <cstdlib> |
6 | #include <cstdlib> |
| 7 | |
7 | |
|
|
8 | #include <glib.h> |
|
|
9 | |
| 8 | #include <tr1/unordered_set> |
10 | #include <tr1/unordered_set> |
| 9 | |
11 | |
| 10 | #include "shstr.h" |
12 | #include "shstr.h" |
|
|
13 | #include "util.h" |
| 11 | |
14 | |
| 12 | struct hash |
|
|
| 13 | { |
|
|
| 14 | std::size_t operator ()(const char *s) const |
|
|
| 15 | { |
|
|
| 16 | unsigned long hash = 0; |
|
|
| 17 | unsigned int i = 0; |
|
|
| 18 | |
|
|
| 19 | /* use the one-at-a-time hash function, which supposedly is |
|
|
| 20 | * better than the djb2-like one used by perl5.005, but |
|
|
| 21 | * certainly is better then the bug used here before. |
|
|
| 22 | * see http://burtleburtle.net/bob/hash/doobs.html |
|
|
| 23 | */ |
|
|
| 24 | while (*s) |
|
|
| 25 | { |
|
|
| 26 | hash += *s++; |
|
|
| 27 | hash += hash << 10; |
|
|
| 28 | hash ^= hash >> 6; |
|
|
| 29 | } |
|
|
| 30 | |
|
|
| 31 | hash += hash << 3; |
|
|
| 32 | hash ^= hash >> 11; |
|
|
| 33 | hash += hash << 15; |
|
|
| 34 | |
|
|
| 35 | return hash; |
|
|
| 36 | } |
|
|
| 37 | }; |
|
|
| 38 | |
|
|
| 39 | struct equal |
|
|
| 40 | { |
|
|
| 41 | bool operator ()(const char *a, const char *b) const |
|
|
| 42 | { |
|
|
| 43 | return !strcmp (a, b); |
|
|
| 44 | } |
|
|
| 45 | }; |
|
|
| 46 | |
|
|
| 47 | typedef std::tr1::unordered_set<const char *, hash, equal> HT; |
15 | typedef std::tr1::unordered_set<const char *, str_hash, str_equal> HT; |
| 48 | |
16 | |
| 49 | static HT ht; |
17 | static HT ht; |
| 50 | |
18 | |
| 51 | static const char *makevec (const char *s) |
19 | static const char *makevec (const char *s) |
| 52 | { |
20 | { |
| 53 | int len = strlen (s); |
21 | int len = strlen (s); |
| 54 | |
22 | |
| 55 | const char *v = (const char *)(2 + (int *)malloc (sizeof (int) * 2 + len + 1)); |
23 | const char *v = (const char *)(2 + (int *)g_slice_alloc (sizeof (int) * 2 + len + 1)); |
| 56 | |
24 | |
| 57 | shstr::length (v) = len; |
25 | shstr::length (v) = len; |
| 58 | shstr::refcnt (v) = 1; |
26 | shstr::refcnt (v) = 1; |
| 59 | |
27 | |
| 60 | memcpy ((char *)v, s, len + 1); |
28 | memcpy ((char *)v, s, len + 1); |
| … | |
… | |
| 108 | |
76 | |
| 109 | if (i == ht.end ()) |
77 | if (i == ht.end ()) |
| 110 | i = ht.begin (); |
78 | i = ht.begin (); |
| 111 | |
79 | |
| 112 | // go through all strings roughly once every 4 minutes |
80 | // go through all strings roughly once every 4 minutes |
| 113 | int n = ht.size () / 256 + 16; |
81 | int n = ht.size () / 256 + 16000; |
| 114 | |
82 | |
| 115 | for (;;) |
83 | for (;;) |
| 116 | { |
84 | { |
| 117 | if (i == ht.end ()) |
85 | if (i == ht.end ()) |
| 118 | { |
86 | { |
| … | |
… | |
| 125 | { |
93 | { |
| 126 | HT::iterator o = i++; |
94 | HT::iterator o = i++; |
| 127 | const char *s = *o; |
95 | const char *s = *o; |
| 128 | ht.erase (o); |
96 | ht.erase (o); |
| 129 | |
97 | |
|
|
98 | int len = length (s); |
|
|
99 | |
| 130 | //printf ("GC %4d %3d %d >%s<%d\n", (int)ht.size (), n, shstr::refcnt (s), s, shstr::length (s)); |
100 | //printf ("GC %4d %3d %d >%s<%d\n", (int)ht.size (), n, shstr::refcnt (s), s, shstr::length (s)); |
| 131 | free (-2 + (int *)s); |
101 | g_slice_free1 (sizeof (int) * 2 + length (s) + 1, -2 + (int *)s); |
| 132 | } |
102 | } |
| 133 | else |
103 | else |
| 134 | ++i; |
104 | ++i; |
| 135 | } |
105 | } |
| 136 | |
106 | |
