We've run into an alignment issue in SQLite 3.6.14.2 which only
seems to cause a problem under Solaris Sparc in our testing.
Failure narrowed down to:
src/rowset.c:186
pEntry->v = rowid
pEntry is 0xXXXXX4
__alignof(*pEntry) is 8 (because of an i64)
However sizeof(RowSet) is 28 (and 28%8 = 4), and pEntry starts 1 RowSet
after freshly allocated, 8-bytes aligned) memory (see sqlite3RowSetInit)
So it crashes. This is definitely a bug in sqlite.
Suggested patch that seems to work for us:
===== sqlite/src/rowset.c 1.1 vs edited =====
--- 1.1/sqlite/src/rowset.c 2009-05-19 14:07:53 -07:00
+++ edited/sqlite/src/rowset.c
2009-05-26 15:43:56 -07:00
@@ -127,6 +127,7 @@
*/
RowSet *sqlite3RowSetInit(sqlite3 *db, void *pSpace, unsigned int N){
RowSet *p;
+ int n;
assert( N >= sizeof(*p) );
p = pSpace;
p->pChunk = 0;
@@ -134,8 +135,14 @@
p->pEntry = 0;
p->pLast = 0;
p->pTree = 0;
- p->pFresh = (struct RowSetEntry*)&p[1];
- p->nFresh = (u16)((N - sizeof(*p))/sizeof(struct RowSetEntry));
+ /* Alignment must be a power of 2, and at least equal to
+ __alignof(struct RowSetEntry) */
+ #define MIN_ALIGNMENT 8
+ n = sizeof(*p);
+ /* Round up to next alignment */
+ n = (n - 1) / MIN_ALIGNMENT * MIN_ALIGNMENT + MIN_ALIGNMENT;
+ p->pFresh = (struct RowSetEntry*)((char *)p + n);
+ p->nFresh = (u16)((N - n)/sizeof(struct RowSetEntry));
p->isSorted = 1;
p->iBatch = 0;
return p;
_______________________________________________
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
