X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=bfile%2Fcfile.c;h=6c228f353964182df8845fe13bb1ab2445cb8d6b;hb=7edae22b5d556f3db505615888ef02b950dd5e83;hp=57be83322c31f325401c28ba20063c4471c4079b;hpb=13b6e86170cf02550bb4abc6e1598d09be7d5bae;p=idzebra-moved-to-github.git diff --git a/bfile/cfile.c b/bfile/cfile.c index 57be833..6c228f3 100644 --- a/bfile/cfile.c +++ b/bfile/cfile.c @@ -4,7 +4,34 @@ * Sebastian Hammer, Adam Dickmeiss * * $Log: cfile.c,v $ - * Revision 1.7 1995-12-15 10:35:07 adam + * Revision 1.15 1996-04-18 16:02:56 adam + * Changed logging a bit. + * Removed warning message when commiting flat shadow files. + * + * Revision 1.14 1996/04/12 07:01:55 adam + * Yet another bug fix (next_block was initialized to 0; now set to 1). + * + * Revision 1.13 1996/04/09 14:48:49 adam + * Bug fix: offset calculation when using flat files was completely broken. + * + * Revision 1.12 1996/04/09 06:47:28 adam + * Function scan_areadef doesn't use sscanf (%n fails on this Linux). + * + * Revision 1.11 1996/03/26 15:59:05 adam + * The directory of the shadow table file can be specified by the new + * bf_lockDir call. + * + * Revision 1.10 1996/02/07 14:03:46 adam + * Work on flat indexed shadow files. + * + * Revision 1.9 1996/02/07 10:08:43 adam + * Work on flat shadow (not finished yet). + * + * Revision 1.8 1995/12/15 12:36:52 adam + * Moved hash file information to union. + * Renamed commit files. + * + * Revision 1.7 1995/12/15 10:35:07 adam * Changed names of commit files. * * Revision 1.6 1995/12/11 09:03:53 adam @@ -42,6 +69,8 @@ static int write_head (CFile cf) int bno = 1; const char *tab = (char*) cf->array; + if (!tab) + return 0; while (left >= HASH_BSIZE) { mf_write (cf->hash_mf, bno++, 0, 0, tab); @@ -59,6 +88,8 @@ static int read_head (CFile cf) int bno = 1; char *tab = (char*) cf->array; + if (!tab) + return 0; while (left >= HASH_BSIZE) { mf_read (cf->hash_mf, bno++, 0, 0, tab); @@ -74,12 +105,13 @@ static int read_head (CFile cf) CFile cf_open (MFile mf, MFile_area area, const char *fname, int block_size, int wflag, int *firstp) { - char path[256]; + char path[1024]; int i; CFile cf = xmalloc (sizeof(*cf)); int hash_bytes; cf->rmf = mf; + logf (LOG_LOG, "cf_open %s %s", cf->rmf->name, wflag ? "rdwr" : "rd"); sprintf (path, "%s-b", fname); if (!(cf->block_mf = mf_open (area, path, block_size, wflag))) { @@ -99,9 +131,9 @@ CFile cf_open (MFile mf, MFile_area area, const char *fname, *firstp = 1; cf->head.state = 1; cf->head.block_size = block_size; - cf->head.hash_size = 401; + cf->head.hash_size = 199; hash_bytes = cf->head.hash_size * sizeof(int); - cf->head.next_bucket = + cf->head.flat_bucket = cf->head.next_bucket = cf->head.first_bucket = (hash_bytes+sizeof(cf->head))/HASH_BSIZE + 2; cf->head.next_block = 1; if (wflag) @@ -116,21 +148,32 @@ CFile cf_open (MFile mf, MFile_area area, const char *fname, { *firstp = 0; assert (cf->head.block_size == block_size); - assert (cf->head.hash_size > 2 && cf->head.hash_size < 200000); + assert (cf->head.hash_size > 2); hash_bytes = cf->head.hash_size * sizeof(int); assert (cf->head.next_bucket > 0); - cf->array = xmalloc (hash_bytes); + assert (cf->head.next_block > 0); + if (cf->head.state == 1) + cf->array = xmalloc (hash_bytes); + else + cf->array = NULL; read_head (cf); } - cf->parray = xmalloc (cf->head.hash_size * sizeof(*cf->parray)); - for (i = 0; ihead.hash_size; i++) - cf->parray[i] = NULL; + if (cf->head.state == 1) + { + cf->parray = xmalloc (cf->head.hash_size * sizeof(*cf->parray)); + for (i = 0; ihead.hash_size; i++) + cf->parray[i] = NULL; + } + else + cf->parray = NULL; cf->bucket_lru_front = cf->bucket_lru_back = NULL; cf->bucket_in_memory = 0; - cf->max_bucket_in_memory = 400; + cf->max_bucket_in_memory = 100; cf->dirty = 0; cf->iobuf = xmalloc (cf->head.block_size); memset (cf->iobuf, 0, cf->head.block_size); + cf->no_hits = 0; + cf->no_miss = 0; return cf; } @@ -238,7 +281,17 @@ static struct CFile_hash_bucket *new_bucket (CFile cf, int *block_no, int hno) return p; } -int cf_lookup (CFile cf, int no) +static int cf_lookup_flat (CFile cf, int no) +{ + int hno = (no*sizeof(int))/HASH_BSIZE; + int off = (no*sizeof(int)) - hno*HASH_BSIZE; + int vno = 0; + + mf_read (cf->hash_mf, hno+cf->head.next_bucket, off, sizeof(int), &vno); + return vno; +} + +static int cf_lookup_hash (CFile cf, int no) { int hno = cf_hash (cf, no); struct CFile_hash_bucket *hb; @@ -248,7 +301,10 @@ int cf_lookup (CFile cf, int no) { for (i = 0; iph.vno[i]; i++) if (hb->ph.no[i] == no) + { + (cf->no_hits)++; return hb->ph.vno[i]; + } } for (block_no = cf->array[hno]; block_no; block_no = hb->ph.next_bucket) { @@ -259,6 +315,7 @@ int cf_lookup (CFile cf, int no) } if (hb) continue; + (cf->no_miss)++; hb = get_bucket (cf, block_no, hno); for (i = 0; iph.vno[i]; i++) if (hb->ph.no[i] == no) @@ -267,19 +324,78 @@ int cf_lookup (CFile cf, int no) return 0; } -int cf_new (CFile cf, int no) +static void cf_write_flat (CFile cf, int no, int vno) +{ + int hno = (no*sizeof(int))/HASH_BSIZE; + int off = (no*sizeof(int)) - hno*HASH_BSIZE; + + hno += cf->head.next_bucket; + if (hno >= cf->head.flat_bucket) + cf->head.flat_bucket = hno+1; + cf->dirty = 1; + mf_write (cf->hash_mf, hno, off, sizeof(int), &vno); +} + +static void cf_moveto_flat (CFile cf) +{ + struct CFile_hash_bucket *p; + int i, j; + + logf (LOG_LOG, "Moving to flat shadow: %s", cf->rmf->name); + logf (LOG_LOG, "hits=%d miss=%d bucket_in_memory=%d total=%d", + cf->no_hits, cf->no_miss, cf->bucket_in_memory, + cf->head.next_bucket - cf->head.first_bucket); + assert (cf->head.state == 1); + flush_bucket (cf, -1); + assert (cf->bucket_in_memory == 0); + p = xmalloc (sizeof(*p)); + for (i = cf->head.first_bucket; i < cf->head.next_bucket; i++) + { + if (!mf_read (cf->hash_mf, i, 0, 0, &p->ph)) + { + logf (LOG_FATAL|LOG_ERRNO, "read bucket moveto flat"); + exit (1); + } + for (j = 0; j < HASH_BUCKET && p->ph.vno[j]; j++) + cf_write_flat (cf, p->ph.no[j], p->ph.vno[j]); + } + xfree (p); + xfree (cf->array); + cf->array = NULL; + xfree (cf->parray); + cf->parray = NULL; + cf->head.state = 2; + cf->dirty = 1; +} + +static int cf_lookup (CFile cf, int no) +{ + if (cf->head.state > 1) + return cf_lookup_flat (cf, no); + return cf_lookup_hash (cf, no); +} + +static int cf_new_flat (CFile cf, int no) +{ + int vno = (cf->head.next_block)++; + + cf_write_flat (cf, no, vno); + return vno; +} + +static int cf_new_hash (CFile cf, int no) { int hno = cf_hash (cf, no); struct CFile_hash_bucket *hbprev = NULL, *hb = cf->parray[hno]; - int *bucketpp = &cf->array[hno]; - int i; - int vno = (cf->head.next_block)++; - + int *bucketpp = &cf->array[hno]; + int i, vno = (cf->head.next_block)++; + for (hb = cf->parray[hno]; hb; hb = hb->h_next) if (!hb->ph.vno[HASH_BUCKET-1]) for (i = 0; iph.vno[i]) { + (cf->no_hits)++; hb->ph.no[i] = no; hb->ph.vno[i] = vno; hb->dirty = 1; @@ -297,6 +413,7 @@ int cf_new (CFile cf, int no) } if (hb) continue; + (cf->no_miss)++; hb = get_bucket (cf, *bucketpp, hno); assert (hb); for (i = 0; ihead.state > 1) + return cf_new_flat (cf, no); + if (cf->no_miss*5 > cf->no_hits) + { + cf_moveto_flat (cf); + assert (cf->head.state > 1); + return cf_new_flat (cf, no); + } + return cf_new_hash (cf, no); +} + + int cf_read (CFile cf, int no, int offset, int num, void *buf) { int block; @@ -360,9 +491,13 @@ int cf_write (CFile cf, int no, int offset, int num, const void *buf) int cf_close (CFile cf) { + logf (LOG_LOG, "hits=%d miss=%d bucket_in_memory=%d total=%d", + cf->no_hits, cf->no_miss, cf->bucket_in_memory, + cf->head.next_bucket - cf->head.first_bucket); flush_bucket (cf, -1); if (cf->dirty) { + logf (LOG_LOG, "dirty. write header"); mf_write (cf->hash_mf, 0, 0, sizeof(cf->head), &cf->head); write_head (cf); } @@ -372,6 +507,7 @@ int cf_close (CFile cf) xfree (cf->parray); xfree (cf->iobuf); xfree (cf); + logf (LOG_LOG, "cf_close %s", cf->rmf->name); return 0; }