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
>
