| … | |
… | |
| 6 | #include <cstdlib> |
6 | #include <cstdlib> |
| 7 | |
7 | |
| 8 | #include <tr1/unordered_set> |
8 | #include <tr1/unordered_set> |
| 9 | |
9 | |
| 10 | #include "shstr.h" |
10 | #include "shstr.h" |
|
|
11 | |
|
|
12 | // NOTE: even with lots of stuff loaded, we do not usually have >>20000 strings. |
|
|
13 | // maybe refcounting is just overhead? |
| 11 | |
14 | |
| 12 | struct hash |
15 | struct hash |
| 13 | { |
16 | { |
| 14 | std::size_t operator ()(const char *s) const |
17 | std::size_t operator ()(const char *s) const |
| 15 | { |
18 | { |
| … | |
… | |
| 55 | return s; |
58 | return s; |
| 56 | |
59 | |
| 57 | HT::iterator i = ht.find (s); |
60 | HT::iterator i = ht.find (s); |
| 58 | |
61 | |
| 59 | return i != ht.end () |
62 | return i != ht.end () |
| 60 | ? (char *)*i |
63 | ? *i |
| 61 | : 0; |
64 | : 0; |
| 62 | } |
65 | } |
| 63 | |
66 | |
| 64 | const char * |
67 | const char * |
| 65 | shstr::intern (const char *s) |
68 | shstr::intern (const char *s) |
| 66 | { |
69 | { |
| 67 | if (!s) |
70 | if (!s) |
| 68 | return s; |
71 | return s; |
| 69 | |
72 | |
| 70 | if (const char *found = find (s)) |
73 | if (const char *found = find (s)) |
|
|
74 | { |
|
|
75 | ++refcnt (found); |
| 71 | return found; |
76 | return found; |
|
|
77 | } |
| 72 | |
78 | |
| 73 | int len = strlen (s); |
79 | int len = strlen (s); |
| 74 | |
80 | |
| 75 | int *v = (int *)malloc (sizeof (int) * 2 + len + 1); |
81 | const char *v = (const char *)(2 + (int *)malloc (sizeof (int) * 2 + len + 1)); |
| 76 | |
82 | |
| 77 | v [0] = len; |
83 | length (v) = len; |
| 78 | v [1] = 0; |
84 | refcnt (v) = 1; |
| 79 | |
85 | |
| 80 | v += 2; |
86 | memcpy ((char *)v, s, len + 1); |
| 81 | |
87 | |
| 82 | memcpy (v, s, len + 1); |
88 | ht.insert (v); |
| 83 | |
89 | |
| 84 | ht.insert ((char *)v); |
90 | return v; |
| 85 | |
|
|
| 86 | return (char *)v; |
|
|
| 87 | } |
91 | } |
| 88 | |
92 | |
| 89 | // TODO: periodically test refcounts == 0 for a few strings (e.g. one hash bucket, |
93 | // periodically test refcounts == 0 for a few strings |
| 90 | // exploiting the fatc that iterators stay valid for unordered_set). |
94 | // this is the ONLY thing that erases stuff from ht. keep it that way. |
| 91 | void |
95 | void |
| 92 | shstr::gc () |
96 | shstr::gc () |
| 93 | { |
97 | { |
|
|
98 | static const char *curpos; |
|
|
99 | |
|
|
100 | HT::iterator i = curpos ? ht.find (curpos) : ht.begin (); |
|
|
101 | |
|
|
102 | if (i == ht.end ()) |
|
|
103 | i = ht.begin (); |
|
|
104 | |
|
|
105 | // go through all strings roughly once every 4 minutes |
|
|
106 | for (int n = ht.size () / 256 + 16; --n; ) |
|
|
107 | { |
|
|
108 | if (i == ht.end ()) |
|
|
109 | { |
|
|
110 | curpos = 0; |
|
|
111 | return; |
|
|
112 | } |
|
|
113 | |
|
|
114 | if (!refcnt (*i)) |
|
|
115 | { |
|
|
116 | HT::iterator o = i++; |
|
|
117 | const char *s = *o; |
|
|
118 | ht.erase (o); |
|
|
119 | |
|
|
120 | //printf ("GC %4d %3d %d >%s<%d\n", (int)ht.size (), n, shstr::refcnt (s), s, shstr::length (s)); |
|
|
121 | free (-2 + (int *)s); |
|
|
122 | } |
|
|
123 | else |
|
|
124 | ++i; |
|
|
125 | } |
|
|
126 | |
|
|
127 | curpos = *i; |
| 94 | } |
128 | } |
| 95 | |
129 | |
|
|
130 | //TODO: this should of course not be here |
| 96 | /* buf_overflow() - we don't want to exceed the buffer size of |
131 | /* buf_overflow() - we don't want to exceed the buffer size of |
| 97 | * buf1 by adding on buf2! Returns true if overflow will occur. |
132 | * buf1 by adding on buf2! Returns true if overflow will occur. |
| 98 | */ |
133 | */ |
| 99 | |
134 | |
| 100 | int |
135 | int |
