Kirill,
I see what you're getting at. The valgrind error comes about when we call get_column_info in streams.c. No string is allocated for table_name there, so the pointer remains uninitialized.
I am unfamiliar with the streams use case, but looking at tests/db.c:test_streamtable(), I think that the correct tablename to be returning is '_Streams'. Would you agree?
Thanks, ~Nathan
On Sun, 1 Nov 2009, Kirill K. Smirnov wrote:
Hi Nathan,
haystack_table_name = NULL; r = table->ops->get_column_info( table, i, &col_name, NULL, NULL, &haystack_table_name ); if( r != ERROR_SUCCESS )I'm afraid, but this patch silences valgrind only, but does not solves the problem. It seems to me, that if table->ops->get_column_info() succeeds, it initializes haystack_table_name, and if it fails, we do not use haystack_table_name at all (we immediately return from this function). Thus, the problem lies somewhere in get_column_info().
So it is useless (and harmful, because it confuses valgrind) to initialize it like this.
-- Kirill