|
|
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 "global.h" |
11 | |
14 | |
12 | // NOTE: even with lots of stuff loaded, we do not usually have >>20000 strings. |
|
|
13 | // maybe refcounting is just overhead? |
|
|
14 | |
|
|
15 | struct hash |
|
|
16 | { |
|
|
17 | std::size_t operator ()(const char *s) const |
|
|
18 | { |
|
|
19 | unsigned long hash = 0; |
|
|
20 | unsigned int i = 0; |
|
|
21 | |
|
|
22 | /* use the one-at-a-time hash function, which supposedly is |
|
|
23 | * better than the djb2-like one used by perl5.005, but |
|
|
24 | * certainly is better then the bug used here before. |
|
|
25 | * see http://burtleburtle.net/bob/hash/doobs.html |
|
|
26 | */ |
|
|
27 | while (*s) |
|
|
28 | { |
|
|
29 | hash += *s++; |
|
|
30 | hash += hash << 10; |
|
|
31 | hash ^= hash >> 6; |
|
|
32 | } |
|
|
33 | |
|
|
34 | hash += hash << 3; |
|
|
35 | hash ^= hash >> 11; |
|
|
36 | hash += hash << 15; |
|
|
37 | |
|
|
38 | return hash; |
|
|
39 | } |
|
|
40 | }; |
|
|
41 | |
|
|
42 | struct equal |
|
|
43 | { |
|
|
44 | bool operator ()(const char *a, const char *b) const |
|
|
45 | { |
|
|
46 | return !strcmp (a, b); |
|
|
47 | } |
|
|
48 | }; |
|
|
49 | |
|
|
50 | typedef std::tr1::unordered_set<const char *, hash, equal> HT; |
15 | typedef std::tr1::unordered_set<const char *, str_hash, str_equal> HT; |
51 | |
16 | |
52 | static HT ht; |
17 | static HT ht; |
|
|
18 | |
|
|
19 | static const char * |
|
|
20 | makevec (const char *s) |
|
|
21 | { |
|
|
22 | int len = strlen (s); |
|
|
23 | |
|
|
24 | const char *v = (const char *) (2 + (int *) g_slice_alloc (sizeof (int) * 2 + len + 1)); |
|
|
25 | |
|
|
26 | shstr::length (v) = len; |
|
|
27 | shstr::refcnt (v) = 1; |
|
|
28 | |
|
|
29 | memcpy ((char *) v, s, len + 1); |
|
|
30 | |
|
|
31 | return v; |
|
|
32 | } |
|
|
33 | |
|
|
34 | static const char * |
|
|
35 | makenull () |
|
|
36 | { |
|
|
37 | const char *s = makevec ("(null)"); |
|
|
38 | shstr::length (s) = 0; |
|
|
39 | return s; |
|
|
40 | } |
|
|
41 | |
|
|
42 | const char *shstr::null = makenull (); |
53 | |
43 | |
54 | const char * |
44 | const char * |
55 | shstr::find (const char *s) |
45 | shstr::find (const char *s) |
56 | { |
46 | { |
57 | if (!s) |
47 | if (!s) |
58 | return s; |
48 | return s; |
59 | |
49 | |
60 | HT::iterator i = ht.find (s); |
50 | HT::iterator i = ht.find (s); |
61 | |
51 | |
62 | return i != ht.end () |
52 | return i != ht.end ()? *i : 0; |
63 | ? *i |
|
|
64 | : 0; |
|
|
65 | } |
53 | } |
66 | |
54 | |
67 | const char * |
55 | const char * |
68 | shstr::intern (const char *s) |
56 | shstr::intern (const char *s) |
69 | { |
57 | { |
70 | if (!s) |
58 | if (!s) |
71 | return s; |
59 | return null; |
72 | |
60 | |
73 | if (const char *found = find (s)) |
61 | if (const char *found = find (s)) |
74 | { |
62 | { |
75 | ++refcnt (found); |
63 | ++refcnt (found); |
76 | return found; |
64 | return found; |
77 | } |
65 | } |
78 | |
66 | |
79 | int len = strlen (s); |
67 | s = makevec (s); |
80 | |
|
|
81 | const char *v = (const char *)(2 + (int *)malloc (sizeof (int) * 2 + len + 1)); |
|
|
82 | |
|
|
83 | length (v) = len; |
|
|
84 | refcnt (v) = 1; |
|
|
85 | |
|
|
86 | memcpy ((char *)v, s, len + 1); |
|
|
87 | |
|
|
88 | ht.insert (v); |
68 | ht.insert (s); |
89 | |
|
|
90 | return v; |
69 | return s; |
91 | } |
70 | } |
92 | |
71 | |
93 | // periodically test refcounts == 0 for a few strings |
72 | // periodically test refcounts == 0 for a few strings |
94 | // this is the ONLY thing that erases stuff from ht. keep it that way. |
73 | // this is the ONLY thing that erases stuff from ht. keep it that way. |
95 | void |
74 | void |
96 | shstr::gc () |
75 | shstr::gc () |
97 | { |
76 | { |
|
|
77 | return; //D |
|
|
78 | //D currently disabled: some datastructures might still store them |
|
|
79 | //D but their pointers will become invalidated |
98 | static const char *curpos; |
80 | static const char *curpos; |
99 | |
81 | |
100 | HT::iterator i = curpos ? ht.find (curpos) : ht.begin (); |
82 | HT::iterator i = curpos ? ht.find (curpos) : ht.begin (); |
101 | |
83 | |
102 | if (i == ht.end ()) |
84 | if (i == ht.end ()) |
103 | i = ht.begin (); |
85 | i = ht.begin (); |
104 | |
86 | |
105 | // go through all strings roughly once every 4 minutes |
87 | // go through all strings roughly once every 4 minutes |
106 | for (int n = ht.size () / 256 + 16; --n; ) |
88 | int n = ht.size () / 256 + 16; |
|
|
89 | |
|
|
90 | for (;;) |
107 | { |
91 | { |
108 | if (i == ht.end ()) |
92 | if (i == ht.end ()) |
109 | { |
93 | { |
110 | curpos = 0; |
94 | curpos = 0; |
111 | return; |
95 | return; |
112 | } |
96 | } |
113 | |
97 | else if (!--n) |
|
|
98 | break; |
114 | if (!refcnt (*i)) |
99 | else if (!refcnt (*i)) |
115 | { |
100 | { |
116 | HT::iterator o = i++; |
101 | HT::iterator o = i++; |
117 | const char *s = *o; |
102 | const char *s = *o; |
|
|
103 | |
118 | ht.erase (o); |
104 | ht.erase (o); |
119 | |
105 | |
120 | //printf ("GC %4d %3d %d >%s<%d\n", (int)ht.size (), n, shstr::refcnt (s), s, shstr::length (s)); |
106 | //printf ("GC %4d %3d %d >%s<%d\n", (int)ht.size (), n, shstr::refcnt (s), s, shstr::length (s)); |
121 | free (-2 + (int *)s); |
107 | g_slice_free1 (sizeof (int) * 2 + length (s) + 1, -2 + (int *) s); |
122 | } |
108 | } |
123 | else |
109 | else |
124 | ++i; |
110 | ++i; |
125 | } |
111 | } |
126 | |
112 | |
127 | curpos = *i; |
113 | curpos = *i; |
128 | } |
114 | } |
129 | |
115 | |
|
|
116 | shstr skill_names[NUM_SKILLS]; |
|
|
117 | |
|
|
118 | // what weird misoptimisation is this again? |
|
|
119 | const shstr undead_name ("undead"); |
|
|
120 | |
130 | //TODO: this should of course not be here |
121 | //TODO: this should of course not be here |
|
|
122 | |
131 | /* buf_overflow() - we don't want to exceed the buffer size of |
123 | /* buf_overflow() - we don't want to exceed the buffer size of |
132 | * buf1 by adding on buf2! Returns true if overflow will occur. |
124 | * buf1 by adding on buf2! Returns true if overflow will occur. |
133 | */ |
125 | */ |
134 | |
126 | |
135 | int |
127 | int |
136 | buf_overflow (const char *buf1, const char *buf2, int bufsize) |
128 | buf_overflow (const char *buf1, const char *buf2, int bufsize) |
137 | { |
129 | { |
138 | int len1 = 0, len2 = 0; |
130 | int len1 = 0, len2 = 0; |
139 | |
131 | |
140 | if (buf1) |
132 | if (buf1) |
141 | len1 = strlen (buf1); |
133 | len1 = strlen (buf1); |
142 | if (buf2) |
134 | if (buf2) |
143 | len2 = strlen (buf2); |
135 | len2 = strlen (buf2); |
144 | if ((len1 + len2) >= bufsize) |
136 | if ((len1 + len2) >= bufsize) |
145 | return 1; |
|
|
146 | return 0; |
137 | return 1; |
|
|
138 | return 0; |
147 | } |
139 | } |
148 | |
|
|