kohei-us / mdds Goto Github PK

If I try to build mdds with boost-1.57 I get the following errors during make 
check:

"In file included from include/mdds/multi_type_vector.hpp:34:0,
                 from ./src/multi_type_vector_test_default.cpp:31:
include/mdds/multi_type_vector_types.hpp: In static member function 'static 
void mdds::mtv::element_block<_Self, _TypeId, _Data>::print_block(const 
mdds::mtv::base_element_block&)':
include/mdds/multi_type_vector_types.hpp:258:9: error: 'for_each' is not a 
member of 'std'
         std::for_each(blk2.begin(), blk2.end(), print_block_array());
...
(more of the same kind)"

The attached patch fixes that. The reason is probably the newer boost version 
dropping a #include.

variables set but unused

Support a cached search operation where the last search result is stored, and 
is used instead of running full search each and every time.

Tree search is not necessarily cheap, and there are many cases where the client 
code has to cache the last search result and re-use it for better performance.  
Why not support this in the library itself?

It has been broken by changeset 7b807e5a4295 "Re-defined 
flat_segment_tree::node using a template-only, with no inheritance."

Fix attached.

Currently, flat_segment_tree::search() only searches from the head position.  
Add a new method to allow searching from the end position.

While I'm at it, I should also rename search() and search_tree() to find() and 
find_tree(), respectively, to make the naming more in line with STL's.

When all nodes are shifted right and the leftmost node has a value other than 
initial, a new node is inserted, but it is not linked from the node right to 
it. This causes crash in the loop in disconnect_all_leaves, because 
disconnect_all_nodes(cur_node) deletes the node next_node is currently pointing 
to.

Fix attached.

This is just to test how this bug tracker works.

What steps will reproduce the problem?
1. make test.recset.perf

What version of the product are you using?
mdds-0.3.0

Fix attached.

flat_segment_tree should have swap() method which should swap the guts of two 
flat_segment_tree instances.

If the tree is shifted from the leftmost leaf and the value of it differs from 
the initial value, a new node that carries the value of the leftmost leaf is 
inserted. But if the shift size is bigger than the span of the tree, its key 
will be greater that that of the rightmost leaf.

Demonstration example and fix attached.

As is visible from the attached reproducer, the default operator=() leads to 
crash. So it must either be implemented (probably in the same way as the copy 
constructor) or disabled.

I.e., fst.shift_right(fst.min_key(), anysize, true) does not skip the leftmost 
node and fills the default value from left. If this is a special case, it 
should be documented. Otherwise, the attached patch fixes it.

max_key() and min_key() should return the max and min key values, respectively, 
and default_value() should return the default initial value.

Please ship Makefile.am and configure.{ac,in} files in your release tarballs in 
order for us to be able modify build system if needed.

It is much easier for packager to alter the above files and run autogen.sh 
again then to crawl throught the raw configure sh script.

Thanks