Update to newer version of HighFive lib and implement new features

[LJoyL]

Current HighFive version : v2.4.1

Desired Update version : latest v2.7.1
mainly for these specific features :

• Enable choosing the allocation time of datasets (Add wrapper for allocation time. BlueBrain/HighFive#627).
• Enable page buffered reading (Update changelog for 2.5.1 BlueBrain/HighFive#639).
• Enable collective MPI IO using the Data Transfer Property (Enable collective MPI IO. BlueBrain/HighFive#623).
• Bug fixes

see all release logs:
https://github.com/BlueBrain/HighFive/releases

[LJoyL]

Install new version of HighFive locally

1. Installing library

git clone --recursive https://github.com/BlueBrain/HighFive.git HighFive

cmake -DHIGHFIVE_EXAMPLES=Off \
      -DHIGHFIVE_USE_BOOST=On \
      -DHIGHFIVE_PARALLEL_HDF5=Off \
      -DHIGHFIVE_UNIT_TESTS=Off \
      -DCMAKE_INSTALL_PREFIX=HighFive/build \
      -B HighFive/build \
      HighFive

cmake --build HighFive/build
cmake --install HighFive/build

2. Turn OFF WITH_HIGHFIVE_AS_PACKAGE option in CMakeLists.txt

option(WITH_HIGHFIVE_AS_PACKAGE "HIGHFIVE externals as a dunedaq package" OFF)

3. Adding export to env.sh

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/path/to/lib/HighFive/build
export HIGHFIVE_INC=/path/to/lib/HighFive/include

[LJoyL]

Adding new functionalities

New splitting methods

in HDF5RawDataFile.cpp and HDF5RawDataFile.hpp, adding the following methods for HDF5RawDataFile class:

• write_header_only : Write the header of a trigger record into file
  Need to initialize these objects before calling:

// the source_id_path map that we will build up as we write the TR header
HDF5SourceIDHandler::source_id_path_map_t source_id_path_map;

// the map of fragment types to SourceIDS
HDF5SourceIDHandler::fragment_type_source_id_map_t fragment_type_source_id_map;

// the map of subdetectors to SourceIDS
HDF5SourceIDHandler::subdetector_source_id_map_t subdetector_source_id_map;

// write the record header into the HDF5 file/group
HighFive::Group record_level_group;

• write_fragment_only : write a full fragment at once into file
• write_end_tr : terminate the write of a trigger record

Adding even more splitting with new methods :

• write_fragment_begin : Write only the header and initialize the next fragment to write
• write_fragment_part : Write a Chunk or buffer of the previously initialized fragment
• write_fragment_end : Terminate the write of the previously initialized and wrote fragment

Adding allocation time parameter

Added for Trigger record writing and new methods listed before.

• allocate_early : bool default false (incremental), decide to allocate the dataset at the creation time or at the writing time. dataset creation properties.

data_set_create_props.add(HighFive::AllocationTime(H5D_ALLOC_TIME_EARLY));

[LJoyL]

Adding new tests

HDF5lib tests

• HDF5LIBS_write_fullframe_test.cpp : Test app using new methods writing a full frame at once.
• HDF5LIBS_write_partframe_test.cpp : Test app using new methods writing chunked frames.
• HDF5LIBS_TestWriter.py : New script to easily run benchmark tests on new and old writing tests.

HighFive tests

• HighFive_full_bench.cpp : Test application to try new features implemented in last version of HighFive. This test only use HighFive functionalities and is not using HDF5libs structures.
• run_HighFive_benchmark.sh : New script to benchmark and test new HighFive fonctionalities.