X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=bfile%2Fcfile.c;h=2c5d06bb0c826a27cfaf8dc52e85dbea48977dba;hb=a52e4b70e079601037de557b9a95c07f915f567a;hp=bacbe8392a0e9d16c5172f8e7fc619a0a1547119;hpb=bee234a243b7bfc164f7096f35d1bcb98e7b24a9;p=idzebra-moved-to-github.git

diff --git a/bfile/cfile.c b/bfile/cfile.c
index bacbe83..2c5d06b 100644
--- a/bfile/cfile.c
+++ b/bfile/cfile.c
@@ -4,7 +4,40 @@
  * Sebastian Hammer, Adam Dickmeiss
  *
  * $Log: cfile.c,v $
- * Revision 1.9  1996-02-07 10:08:43  adam
+ * Revision 1.19  1997-02-12 20:37:17  adam
+ * Changed the messages logged. No real code changed.
+ *
+ * Revision 1.18  1996/10/29 13:56:15  adam
+ * Include of zebrautl.h instead of alexutil.h.
+ *
+ * Revision 1.17  1996/04/19 16:49:00  adam
+ * Minor changes.
+ *
+ * Revision 1.16  1996/04/19  16:23:47  adam
+ * Serious bug fix in shadow implementation; function new_bucket might
+ * set wrong bucket number on new bucket.
+ *
+ * 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
@@ -39,7 +72,7 @@
 #include <stdlib.h>
 #include <string.h>
 
-#include <alexutil.h>
+#include <zebrautl.h>
 #include <mfile.h>
 #include "cfile.h"
 
@@ -49,6 +82,8 @@ 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);
@@ -66,6 +101,8 @@ 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);
@@ -87,6 +124,7 @@ CFile cf_open (MFile mf, MFile_area area, const char *fname,
     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)))
     {
@@ -106,9 +144,9 @@ 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)
@@ -126,18 +164,29 @@ CFile cf_open (MFile mf, MFile_area area, const char *fname,
         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; i<cf->head.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; i<cf->head.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;
 }
 
@@ -226,13 +275,13 @@ static struct CFile_hash_bucket *get_bucket (CFile cf, int block_no, int hno)
     return p;
 }
 
-static struct CFile_hash_bucket *new_bucket (CFile cf, int *block_no, int hno)
+static struct CFile_hash_bucket *new_bucket (CFile cf, int *block_nop, int hno)
 {
     struct CFile_hash_bucket *p;
-    int i;
+    int i, block_no;
 
-    *block_no = cf->head.next_bucket++;
-    p = alloc_bucket (cf, *block_no, hno);
+    block_no = *block_nop = cf->head.next_bucket++;
+    p = alloc_bucket (cf, block_no, hno);
 
     for (i = 0; i<HASH_BUCKET; i++)
     {
@@ -240,7 +289,7 @@ static struct CFile_hash_bucket *new_bucket (CFile cf, int *block_no, int hno)
         p->ph.no[i] = 0;
     }
     p->ph.next_bucket = 0;
-    p->ph.this_bucket = *block_no;
+    p->ph.this_bucket = block_no;
     p->dirty = 1;
     return p;
 }
@@ -248,7 +297,7 @@ static struct CFile_hash_bucket *new_bucket (CFile cf, int *block_no, int hno)
 static int cf_lookup_flat (CFile cf, int no)
 {
     int hno = (no*sizeof(int))/HASH_BSIZE;
-    int off = (no*sizeof(int)) - hno*sizeof(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);
@@ -265,7 +314,10 @@ static int cf_lookup_hash (CFile cf, int no)
     {
         for (i = 0; i<HASH_BUCKET && hb->ph.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)
     {
@@ -276,6 +328,24 @@ static int cf_lookup_hash (CFile cf, int no)
         }
         if (hb)
             continue;
+#if 0
+        /* extra check ... */
+        for (hb = cf->bucket_lru_back; hb; hb = hb->lru_next)
+        {
+            if (hb->ph.this_bucket == block_no)
+            {
+                logf (LOG_FATAL, "Found hash bucket on other chain (1)");
+                abort ();
+            }
+            for (i = 0; i<HASH_BUCKET && hb->ph.vno[i]; i++)
+                if (hb->ph.no[i] == no)
+                {
+                    logf (LOG_FATAL, "Found hash bucket on other chain (2)");
+                    abort ();
+                }
+        }
+#endif
+        (cf->no_miss)++;
         hb = get_bucket (cf, block_no, hno);
         for (i = 0; i<HASH_BUCKET && hb->ph.vno[i]; i++)
             if (hb->ph.no[i] == no)
@@ -284,6 +354,50 @@ static int cf_lookup_hash (CFile cf, int no)
     return 0;
 }
 
+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, "cf: Moving to flat shadow: %s", cf->rmf->name);
+    logf (LOG_LOG, "cf: 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)
@@ -291,13 +405,11 @@ static int cf_lookup (CFile cf, int no)
     return cf_lookup_hash (cf, no);
 }
 
-int cf_new_flat (CFile cf, int no)
+static int cf_new_flat (CFile cf, int no)
 {
-    int hno = (no*sizeof(int))/HASH_BSIZE;
-    int off = (no*sizeof(int)) - hno*sizeof(HASH_BSIZE);
     int vno = (cf->head.next_block)++;
 
-    mf_write (cf->hash_mf, hno+cf->head.next_bucket, off, sizeof(int), &vno);
+    cf_write_flat (cf, no, vno);
     return vno;
 }
 
@@ -313,6 +425,7 @@ static int cf_new_hash (CFile cf, int no)
             for (i = 0; i<HASH_BUCKET; i++)
                 if (!hb->ph.vno[i])
                 {
+                    (cf->no_hits)++;
                     hb->ph.no[i] = no;
                     hb->ph.vno[i] = vno;
                     hb->dirty = 1;
@@ -330,6 +443,19 @@ static int cf_new_hash (CFile cf, int no)
             }
         if (hb)
             continue;
+
+#if 0
+        /* extra check ... */
+        for (hb = cf->bucket_lru_back; hb; hb = hb->lru_next)
+        {
+            if (hb->ph.this_bucket == *bucketpp)
+            {
+                logf (LOG_FATAL, "Found hash bucket on other chain");
+                abort ();
+            }
+        }
+#endif
+        (cf->no_miss)++;
         hb = get_bucket (cf, *bucketpp, hno);
         assert (hb);
         for (i = 0; i<HASH_BUCKET; i++)
@@ -355,6 +481,12 @@ int cf_new (CFile cf, int no)
 {
     if (cf->head.state > 1)
         return cf_new_flat (cf, no);
+    if (cf->no_miss*2 > cf->no_hits)
+    {
+        cf_moveto_flat (cf);
+        assert (cf->head.state > 1);
+        return cf_new_flat (cf, no);
+    }
     return cf_new_hash (cf, no);
 }
 
@@ -401,6 +533,9 @@ int cf_write (CFile cf, int no, int offset, int num, const void *buf)
 
 int cf_close (CFile cf)
 {
+    logf (LOG_DEBUG, "cf: 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)
     {