NCollection containers update:
- NCollection_Array1 - updated functionality
- NCollection_Array2 - NCollection_Array1 as a wrapper for 2array
- NCollection_Vector -> NCollection_DynamicArray was renamed and reworked.
TCollection:
- Use static empty string to avoid allocations on empty string
NCollection allocators update:
- NCollection_Allocator - allocator that used Standard::Allocate
- NCollection_OccAllocator - allocator-wrapper that used OCC BaseAllocator objects
- NCollection_IncAllocator - rework to increase performance
Standard:
- Rework functionality to use different allocation libs
- Implement basic of new way to wrap allocations tools
- Define 4 ways to allocation (defines in configure stage)
Additional changes:
- Hash function uses std::hash functionality
- size_t as a hash value
- New HashUtils with Murmur and FVN hash algo for x32 and x64
- Deprecated _0.cxx and .gxx DE classes reorganized
- Create own utility for std memory
- Update Standard_Transient to be more platform-independent
Math TK changes:
- math_Vector -> match_BaseVector<>
- Buffer decreased to cash 32 elements instead of 512
Implementation of new format for quick reading and writing parts of the documents (sub-set of labels and sub-set of attributes). It consists in writing shapes and all their contents right in the TNaming_NamedShape attribute placement and skipping the shape section. New format 12 for Binary file types is assigned to this version.
Added PCDM_ReaderFilter class that could be used in Open methods of TDocStd_Application. If it is defined, it allows to read:
- into already opened document in append mode AppendMode_Protect (do not overwrite existing attributes) or AppendMode_Overwrite
- only specified sub-trees of the document using AddPath (const TCollection_AsciiString& theEntryToRead)
- only specified attributes using AddRead (const TCollection_AsciiString& theRead) where theRead could be "TDataStd_Name", for example
- to skip specified attributes read using AddSkipped (const TCollection_AsciiString& theSkipped) where theSkipped could be "TDF_Reference", for example
The current limitations:
- only in Bin format
- if shapes have in the document shared topology, loaded in "append" mode in different "load" operations, they will have no shared topology anymore
Modifications:
BinLDrivers and BinDrivers packages - modifications related to the quick part tree format flag usage, skipping shape section writing and adding labels sizes into the document to be able to pass labels during the reading quickly.
BinObjMgt_Persistent and BinObjMgt_Position - to add possibility to write directly into the stream some data just after the attribute. Before this record a data-size is recorded.
BinMXCAFDoc package modifications to write BinMXCAFDoc_LocationDriver location in the same way as shapes write location data right after the attribute (empty) data in this new format.
BinTools package: creation of ShapeReader and ShapeWriter classes with same root class ShapeSetBase with ShapeSet class. These classes allows to write/read shapes directly to the stream. If some object is already in the stream, write a reference - relative position of the duplicated object. The old format of documents is still supported by Bin_ToolsShapeSet class.
PCDM_ReaderFilter - Allows the user to create a reading filter. It contains algorithm to browse labels tree quickly, without usage of referencing by entry-strings.
TDocStd, CDF and some other packages are changed for supporting reading filters API and options.
Tests, documentation and upgrade information are also added for both issues: 31839 and 31918 related to this commit.
