|
|
1 | |
| 1 | /* |
2 | /* |
| 2 | * shstr.C |
3 | * shstr.C |
| 3 | */ |
4 | */ |
| 4 | |
5 | |
| 5 | #include <cstring> |
6 | #include <cstring> |
| 6 | #include <cstdlib> |
7 | #include <cstdlib> |
| 7 | |
8 | |
|
|
9 | #include <glib.h> |
|
|
10 | |
| 8 | #include <tr1/unordered_set> |
11 | #include <tr1/unordered_set> |
| 9 | |
12 | |
| 10 | #include "shstr.h" |
13 | #include "shstr.h" |
|
|
14 | #include "util.h" |
| 11 | |
15 | |
| 12 | struct hash |
16 | typedef |
|
|
17 | std::tr1::unordered_set < const char *, |
|
|
18 | str_hash, |
|
|
19 | str_equal > |
|
|
20 | HT; |
|
|
21 | |
|
|
22 | static HT |
|
|
23 | ht; |
|
|
24 | |
|
|
25 | static const char * |
|
|
26 | makevec (const char *s) |
| 13 | { |
27 | { |
| 14 | std::size_t operator ()(const char *s) const |
28 | int |
| 15 | { |
29 | len = strlen (s); |
| 16 | unsigned long hash = 0; |
|
|
| 17 | unsigned int i = 0; |
|
|
| 18 | |
30 | |
| 19 | /* use the one-at-a-time hash function, which supposedly is |
31 | const char * |
| 20 | * better than the djb2-like one used by perl5.005, but |
32 | v = (const char *) (2 + (int *) g_slice_alloc (sizeof (int) * 2 + len + 1)); |
| 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 | |
33 | |
| 31 | hash += hash << 3; |
34 | shstr::length (v) = len; |
| 32 | hash ^= hash >> 11; |
35 | shstr::refcnt (v) = 1; |
| 33 | hash += hash << 15; |
|
|
| 34 | |
36 | |
| 35 | return hash; |
37 | memcpy ((char *) v, s, len + 1); |
| 36 | } |
|
|
| 37 | }; |
|
|
| 38 | |
38 | |
| 39 | struct equal |
39 | return v; |
| 40 | { |
40 | } |
| 41 | bool operator ()(const char *a, const char *b) const |
|
|
| 42 | { |
|
|
| 43 | return !strcmp (a, b); |
|
|
| 44 | } |
|
|
| 45 | }; |
|
|
| 46 | |
41 | |
| 47 | typedef std::tr1::unordered_set<const char *, hash, equal> HT; |
42 | const char * |
|
|
43 | shstr::null = makevec ("<nil>"); |
| 48 | |
44 | |
| 49 | static HT ht; |
45 | // what weird misoptimisation is this again? |
|
|
46 | const shstr undead_name ("undead"); |
| 50 | |
47 | |
| 51 | const char * |
48 | const char * |
| 52 | shstr::find (const char *s) |
49 | shstr::find (const char *s) |
| 53 | { |
50 | { |
|
|
51 | if (!s) |
|
|
52 | return s; |
|
|
53 | |
| 54 | HT::iterator i = ht.find (s); |
54 | HT::iterator i = ht.find (s); |
| 55 | |
55 | |
| 56 | return i != ht.end () |
56 | return i != ht.end ()? *i : 0; |
| 57 | ? (char *)*i |
|
|
| 58 | : 0; |
|
|
| 59 | } |
57 | } |
| 60 | |
58 | |
| 61 | const char * |
59 | const char * |
| 62 | shstr::intern (const char *s) |
60 | shstr::intern (const char *s) |
| 63 | { |
61 | { |
| 64 | HT::iterator i = ht.find (s); |
62 | if (!s) |
|
|
63 | return null; |
| 65 | |
64 | |
| 66 | if (i != ht.end ()) |
65 | if (const char *found = find (s)) |
| 67 | return (char *)*i; |
66 | { |
|
|
67 | ++refcnt (found); |
|
|
68 | return found; |
|
|
69 | } |
| 68 | |
70 | |
| 69 | int len = strlen (s); |
71 | s = makevec (s); |
| 70 | |
|
|
| 71 | int *v = (int *)malloc (sizeof (int) * 2 + len + 1); |
|
|
| 72 | |
|
|
| 73 | v [0] = len; |
|
|
| 74 | v [1] = 0; |
|
|
| 75 | |
|
|
| 76 | v += 2; |
|
|
| 77 | |
|
|
| 78 | memcpy (v, s, len + 1); |
|
|
| 79 | |
|
|
| 80 | ht.insert ((char *)v); |
72 | ht.insert (s); |
| 81 | |
73 | return s; |
| 82 | return (char *)v; |
|
|
| 83 | } |
74 | } |
| 84 | |
75 | |
| 85 | // TODO: periodically test refcounts == 0 for a few strings (e.g. one hash bucket, |
76 | // periodically test refcounts == 0 for a few strings |
| 86 | // exploiting the fatc that iterators stay valid for unordered_set). |
77 | // this is the ONLY thing that erases stuff from ht. keep it that way. |
| 87 | void |
78 | void |
| 88 | shstr::gc () |
79 | shstr::gc () |
| 89 | { |
80 | { |
|
|
81 | static const char *curpos; |
|
|
82 | |
|
|
83 | HT::iterator i = curpos ? ht.find (curpos) : ht.begin (); |
|
|
84 | |
|
|
85 | if (i == ht.end ()) |
|
|
86 | i = ht.begin (); |
|
|
87 | |
|
|
88 | // go through all strings roughly once every 4 minutes |
|
|
89 | int n = ht.size () / 256 + 16; |
|
|
90 | |
|
|
91 | for (;;) |
|
|
92 | { |
|
|
93 | if (i == ht.end ()) |
|
|
94 | { |
|
|
95 | curpos = 0; |
|
|
96 | return; |
|
|
97 | } |
|
|
98 | else if (!--n) |
|
|
99 | break; |
|
|
100 | else if (!refcnt (*i)) |
|
|
101 | { |
|
|
102 | HT::iterator o = i++; |
|
|
103 | const char *s = *o; |
|
|
104 | |
|
|
105 | ht.erase (o); |
|
|
106 | |
|
|
107 | //printf ("GC %4d %3d %d >%s<%d\n", (int)ht.size (), n, shstr::refcnt (s), s, shstr::length (s)); |
|
|
108 | g_slice_free1 (sizeof (int) * 2 + length (s) + 1, -2 + (int *) s); |
|
|
109 | } |
|
|
110 | else |
|
|
111 | ++i; |
|
|
112 | } |
|
|
113 | |
|
|
114 | curpos = *i; |
| 90 | } |
115 | } |
|
|
116 | |
|
|
117 | //TODO: this should of course not be here |
| 91 | |
118 | |
| 92 | /* buf_overflow() - we don't want to exceed the buffer size of |
119 | /* buf_overflow() - we don't want to exceed the buffer size of |
| 93 | * buf1 by adding on buf2! Returns true if overflow will occur. |
120 | * buf1 by adding on buf2! Returns true if overflow will occur. |
| 94 | */ |
121 | */ |
| 95 | |
122 | |
| 96 | int |
123 | int |
| 97 | buf_overflow (const char *buf1, const char *buf2, int bufsize) |
124 | buf_overflow (const char *buf1, const char *buf2, int bufsize) |
| 98 | { |
125 | { |
| 99 | int len1 = 0, len2 = 0; |
126 | int len1 = 0, len2 = 0; |
| 100 | |
127 | |
| 101 | if (buf1) |
128 | if (buf1) |
| 102 | len1 = strlen (buf1); |
129 | len1 = strlen (buf1); |
| 103 | if (buf2) |
130 | if (buf2) |
| 104 | len2 = strlen (buf2); |
131 | len2 = strlen (buf2); |
| 105 | if ((len1 + len2) >= bufsize) |
132 | if ((len1 + len2) >= bufsize) |
| 106 | return 1; |
|
|
| 107 | return 0; |
133 | return 1; |
|
|
134 | return 0; |
| 108 | } |
135 | } |
| 109 | |
|
|
