X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~mdw/git/mLib/blobdiff_plain/03d53b73fc6663cef4493cfda7193c958557c240..17119f584e90f494da54aa9e2d1d10227213d96b:/sym.c diff --git a/sym.c b/sym.c index 03d2e82..a5cec57 100644 --- a/sym.c +++ b/sym.c @@ -1,6 +1,6 @@ /* -*-c-*- * - * $Id: sym.c,v 1.8 1999/08/02 14:45:48 mdw Exp $ + * $Id: sym.c,v 1.15 2004/04/08 01:36:13 mdw Exp $ * * Symbol table management * @@ -27,36 +27,6 @@ * MA 02111-1307, USA. */ -/*----- Revision history --------------------------------------------------* - * - * $Log: sym.c,v $ - * Revision 1.8 1999/08/02 14:45:48 mdw - * Break low-level hashtable code out from sym. - * - * Revision 1.7 1999/06/01 09:49:08 mdw - * Allow things to be looked up by just their caller-supplied hashes. This - * actually needs to be thought through better. - * - * Revision 1.6 1999/05/26 21:08:31 mdw - * Rename symbols in line with newer conventions. - * - * Revision 1.5 1999/05/13 22:48:37 mdw - * Twiddle the extension threshold. Change `-ise' to `-ize' throughout. - * - * Revision 1.4 1999/05/06 19:51:35 mdw - * Reformatted the LGPL notice a little bit. - * - * Revision 1.3 1999/05/05 18:50:31 mdw - * Change licensing conditions to LGPL. - * - * Revision 1.2 1998/11/26 19:27:33 mdw - * Move SYM_NAME into the header file. Fix bugs. - * - * Revision 1.1.1.1 1998/06/17 23:44:42 mdw - * Initial version of mLib - * - */ - /*----- Header files ------------------------------------------------------*/ /* --- ANSI headers --- */ @@ -68,34 +38,13 @@ /* --- Local headers --- */ #include "alloc.h" +#include "arena.h" #include "bits.h" -#include "crc32.h" #include "exc.h" #include "hash.h" #include "sub.h" #include "sym.h" -#include "track.h" - -/*----- Tuning parameters -------------------------------------------------*/ - -/* --- Initial hash table size --- * - * - * This is the initial @mask@ value. It must be of the form %$2^n - 1$%, - * so that it can be used to mask of the bottom bits of a hash value. - */ - -#define SYM_INITSZ 64 /* Size of a new hash table */ - -/* --- Maximum load factor --- * - * - * This parameter controls how much the table has to be loaded before the - * table is extended. The number of elements %$n$%, the number of bins %$b$% - * and the limit %$l$% satisfy the relation %$n < bl$%; if a new item is - * added to the table and this relation is found to be false, the table is - * doubled in size. - */ - -#define SYM_LIMIT(n) ((n) * 2) /* Load factor for growing table */ +#include "unihash.h" /*----- Main code ---------------------------------------------------------*/ @@ -111,11 +60,9 @@ void sym_create(sym_table *t) { - TRACK_CTX("symbol table creation"); - TRACK_PUSH; hash_create(&t->t, SYM_INITSZ); + t->s = &sub_global; t->load = SYM_LIMIT(SYM_INITSZ); - TRACK_POP; } /* --- @sym_destroy@ --- * @@ -132,28 +79,21 @@ void sym_destroy(sym_table *t) { sym_iter i; - TRACK_CTX("symbol table destruction"); - TRACK_PUSH; - SYM_MKITER(&i, t); for (;;) { sym_base *p; SYM_NEXT(&i, p); if (!p) break; - if (p->len > SYM_BUFSZ) - sub_free(p->name.p, p->len); - free(p); + x_free(t->t.a, p); } hash_destroy(&t->t); - - TRACK_POP; } /* --- @sym_find@ --- * * * Arguments: @sym_table *t@ = pointer to symbol table in question - * @const char *n@ = pointer to symbol table to look up + * @const char *n@ = pointer to symbol name to look up * @long l@ = length of the name string or negative to measure * @size_t sz@ = size of desired symbol object, or zero * @unsigned *f@ = pointer to a flag, or null. @@ -190,8 +130,8 @@ void *sym_find(sym_table *t, const char *n, long l, size_t sz, unsigned *f) /* --- Find the correct bin --- */ - len = l < 0 ? strlen(n) + 1 : l; - CRC32(hash, 0, n, len); + len = l < 0 ? strlen(n) : l; + hash = UNIHASH(&unihash_global, n, len); bin = HASH_BIN(&t->t, hash); /* --- Search the bin list --- */ @@ -222,34 +162,18 @@ void *sym_find(sym_table *t, const char *n, long l, size_t sz, unsigned *f) if (f) *f = 0; if (!sz) return (0); - /* --- Create a new symbol block and initialize it --- */ - - { - TRACK_CTX("new symbol creation"); - TRACK_PUSH; - - q = xmalloc(sz); - q->b.next = *bin; - q->b.hash = hash; - q->len = len; - if (n) { - if (len <= SYM_BUFSZ) - memcpy(q->name.b, n, len); - else { - TRY { - q->name.p = sub_alloc(len); - memcpy(q->name.p, n, len); - } CATCH { - free(q); - TRACK_POP; - RETHROW; - } END_TRY; - } - } - - TRACK_POP; - } + /* --- Create a new symbol block and initialize it --- * + * + * The name is attached to the end of the symbol block. + */ + q = x_alloc(t->t.a, sz + len + 1); + q->b.next = *bin; + q->b.hash = hash; + q->name = (char *)q + sz; + memcpy(q->name, n, len); + q->name[len] = 0; + q->len = len; *bin = &q->b; /* --- Consider growing the array --- */ @@ -257,7 +181,7 @@ void *sym_find(sym_table *t, const char *n, long l, size_t sz, unsigned *f) if (t->load) t->load--; if (!t->load && hash_extend(&t->t)) - t->load = SYM_LIMIT(t->t.mask / 2 + 1); + t->load = SYM_LIMIT(t->t.mask + 1); /* --- Finished that, so return the new symbol block --- */ @@ -266,7 +190,7 @@ void *sym_find(sym_table *t, const char *n, long l, size_t sz, unsigned *f) /* --- @sym_remove@ --- * * - * Arguments: @sym_table *i@ = pointer to a symbol table object + * Arguments: @sym_table *t@ = pointer to a symbol table object * @void *p@ = pointer to symbol table entry * * Returns: --- @@ -280,9 +204,7 @@ void sym_remove(sym_table *t, void *p) { sym_base *q = p; hash_remove(&t->t, &q->b); - if (q->len > SYM_BUFSZ) - sub_free(q->name.p, q->len); - free(q); + xfree(q); t->load++; } @@ -316,216 +238,4 @@ void *sym_next(sym_iter *i) return (p); } -/*----- Symbol table test code --------------------------------------------*/ - -#ifdef TEST_RIG - -#include -#include - -typedef struct sym_word { - sym_base base; - size_t i; -} sym_word; - - -/* --- What it does --- * - * - * Reads the file /usr/dict/words (change to some other file full of - * interesting and appropriate bits of text to taste) into a big buffer and - * picks apart into lines. Then picks lines at random and enters them into - * the symbol table. - */ - -int main(void) -{ - char *buff, *p, *lim; - size_t sz, done; - FILE *fp; - int i; - char **line; - sym_word **flag; - sym_table tbl; - int entries; - - /* --- Initialize for reading the file --- */ - - sz = BUFSIZ; - buff = xmalloc(sz + 1); - done = 0; - sub_init(); - - if ((fp = fopen("/usr/dict/words", "r")) == 0) - fprintf(stderr, "buggered ;-( (%s)\n", strerror(errno)); - - /* --- Read buffers of text --- * - * - * Read a buffer. If more to come, double the buffer size and try again. - * This is the method I recommended to comp.lang.c, so I may as well try - * it. - */ - - for (;;) { - i = fread(buff + done, 1, sz - done, fp); - done += i; - if (done != sz) - break; - sz <<= 1; - buff = xrealloc(buff, sz + 1); - } - - /* --- Count the lines --- */ - - lim = buff + done; - - sz = 1; - for (p = buff; p < lim; p++) - if (*p == '\n') sz++; - - /* --- Build a table of line starts --- */ - - line = xmalloc(sz * sizeof(char *)); - i = 0; - line[i++] = buff; - for (p = buff; p < lim; p++) - if (*p == '\n') *p = 0, line[i++] = p + 1; - *lim = 0; - - /* --- Build a table of lines --- * - * - * This reverses the mapping which the symbol table performs, so that its - * accuracy can be tested. - */ - - flag = xmalloc(sz * sizeof(sym_word *)); - for (i = 0; i < sz; i++) - flag[i] = 0; - entries = 0; - - sym_create(&tbl); - - for (;;) { - i = (unsigned)rand() % sz; - - switch (rand() % 5) - { - case 0: { - sym_word *w; - - printf("? %s\n", line[i]); - - w = sym_find(&tbl, line[i], -1, 0, 0); - if (w != flag[i]) - printf("*** error: find `%s' gave %p not %p\n", - line[i], (void *)w, (void *)flag[i]); - else if (w && w->i != i) - printf("*** error: find sym for `%s' gives index %i not %i\n", - line[i], w->i, i); - } break; - - case 1: { - unsigned f; - sym_word *w; - - printf("+ %s\n", line[i]); - - w = sym_find(&tbl, line[i], -1, sizeof(sym_word), &f); - if (f) - { - if (w != flag[i]) - printf("*** error: create `%s' gave %p not %p\n", - line[i], (void *)w, (void *)flag[i]); - else if (w && w->i != i) - printf("*** error: create sym for `%s' gives index %i not %i\n", - line[i], w->i, i); - } - else - { - if (flag[i]) - printf("*** error: create `%s' gave new block, should be %p\n", - line[i], (void *)flag[i]); - else { - flag[i] = w; - w->i = i; - entries++; - } - } - } break; - - case 2: { - sym_iter it; - sym_word *w, **ntbl; - int v; - - if (!entries) - break; - v = (rand() % entries) == 0; - if (!v) - break; - - printf(".\n"); - - ntbl = xmalloc(sz * sizeof(sym_word *)); - memcpy(ntbl, flag, sz * sizeof(sym_word *)); - sym_mkiter(&it, &tbl); - - while ((w = sym_next(&it)) != 0) { - if (ntbl[w->i] == 0) - printf("*** error: iterate returned duff item %s\n", SYM_NAME(w)); - else { - printf(": %s\n", SYM_NAME(w)); - ntbl[w->i] = 0; - } - } - - for (i = 0; i < sz; i++) - if (ntbl[i]) printf("*** error: iterate didn't return item %s\n", - SYM_NAME(ntbl[i])); - free(ntbl); - } break; - - case 3: { - sym_base *b; - int v = rand() & 255 ? 0 : 1; - break; - - printf("dump\n"); - - for (i = 0; i <= tbl.b.mask; i++) { - if (!tbl.b.v[i]) continue; - if (v) printf(" %i: ", i); - b = (sym_base *)tbl.b.v[i]; - while (b) { - if ((b->b.hash & tbl.b.mask) != i) - printf("*** error: bad hash value found"); - if (v) printf("`%s'(%08lx:%lu) ", - line[((sym_word *)b)->i], - b->b.hash, - b->b.hash & tbl.b.mask); - b = (sym_base *)b->b.next; - } - if (v) putchar('\n'); - } - } break; - - case 4: { - if (flag[i]) { - printf("- %s\n", SYM_NAME(&flag[i]->base)); - if ((rand() & 1023) == 0) { - putchar('-'); fflush(stdout); - } - sym_remove(&tbl, flag[i]); - flag[i] = 0; - entries--; - } - } break; - } - - } - - return (0); -} - -#endif - /*----- That's all, folks -------------------------------------------------*/