Code Camp 2017

NeXus Code Camp - 24th to the 26th of October, 2017

Location: The Cosener’s house, Abingdon UK

Venue Website with contact details

Agenda

Tuesday

Morning

• Shape descriptions (Dean Keeble and Tim Spain will be visiting)

Afternoon

• C++ wrappers for HDF5
• Versioning

After Coffee

• Features
• Example files for Ray.

Evening

• 6:00 - 6:30 pre dinner drink
• 7:00 - Dinner at cosners house
  • Main-Course: Rack of lamb with rosemary crumb, dauphinoise potato, chargrilled courgette, vine roasted cherry tomato, fine bean parcel and rosemary jus
  • Vegetarian Option: Asparagus and broad bean risotto, parmesan tuile and lemon balm
  • Pudding: Classic glazed lemon tart, crème fraiche and berries

Wednesday

Morning

• PDB integration (Greame Winter and Charles Mita will be visiting)
• Dealing with dropped frames

Afternoon

• 3pm VC with Pete on Versioning
• Detectors with modules in modules. (Aaron Brewster may be VCing in)

Evening

• Reservartion for 7:00 at wildwood

Thursday

Morning

• Event data (Alan Greer also attending)
• Event log disscussion and focus on example files.

Afternoon

• Wrap up and Telco

The aim is to start at 10:00 on the 24th, 9:00 the other days and finish early afternoon on the 26th.

Minutes for the meeting

Tuesday Morning

Introductions

• Move the PDB and versions to the afternoons so that Pete Jemian can VC in

First item is to investigate the shape methodoloy as investigated by the ESS team

• There is concern over how the shape can accurately describe detectors, and how the data is mapped to the shape.
• Visualisation is important, but so it the use of full descriptions for analyis
• The OFF format will be used for Mantid and McStas
• A new description for Quadrics will be developed by DLS to deal with more complex descriptions for use with xrmc. There should be a converter for this to the OFF NeXus description.

Tuesday Afternoon

2 Seperate threads

Versioning

• Looking at the versioning concept as in the document
• Big discussion about the complexity of local vrs. global versioning, i.e version on baseclasses vrs. versioning on the file.
• Came to the conclusion of a single version number which increments gradually and incrementally to deal with new versions.
• Things to do
  • UPDATE WORKFLOW - Questions here, but we will try to address at the code camp.
  • VERSIONING SEMANTIC - Big changes bump version by 1
  • VALIDATION - is easier now as there is only one global version to validate against
  • LEGACY DOC - Still in the repository.

C++ wrappers for HDF5

• On ess-dmsc github h5cpp
• Came out of some HDF work, didn’t like the C++ API.
• Various parties were doing their own - Tobias asked them to work together.
• Eugen and Martin, ocassionally Jonas Nilison
• Auto builds docs, doxygen
• Migration of Eugen PNI code at the beginning. Lack of tests, now adding more tests.
• Aim to cover all of HDF5 features, e.g. SWMR, compound data types etc.
• Examples of implementation: write_simple_vector
• Suggest to use clang format.
• Using google’s style guide - need consistency
• Fortnightly meetings
• Implemented types: basic types int, float etc.
• Going to be used in the file writer
• Trying to make it as easy to use as h5py
• Looked at internal Paths
• Plan to create performance tests
• Will put on to conan
• Martin will replace the naïve version he is currently using.
• Detector people are interested in using it to store diagnostic data.
• Happy to take suggestions for things to be implement from other parties.
• Discussed using in-memory files for performance tests
• Setting up Windows automated build and hoping to OSX
• MPI support in the future?
• Using runtime exceptions - should it use something more tailored? Do something with the hdf error stack

Wednesday Morning

NXmx PDB discussion

• It was decided that NXmx could easily contain an NXpdb which contains the basic information required for PDB ingest
• Charles Mita from DLS will make a first pass at this for VMXi data

Dropped frames

• could the value be set to NXdetector.undefined, or NXdetector.saturation_value
• Could the count_time value be set to zero (or annother value) for the frame which is not there.
  • a valid count time is anything zero or above, anything other than this is an invalid frame.
• The solution was to create a new dataset called frame_acquired
  • a boolean dataset
  • True if the frame is correcly collected, if the frame has been dropped by the datawriter and padded with fill value, then this shoud be set to False
  • The fill value should be set to a value larger than the NXdetector.saturation_value

Wednesday Afternoon

Features

We all continued to work on features.

Versions revisitited with Pete

• It was agreed that Base class versions were not that useful, and so these can be made optional to inherit from the main version
• Agreed with the concept of global versioning
• in the January release the version numbers will move to Year.Month instead of 3.4

Modules in Modules, with Aaron Brewster.

• Seems like a sencible option
• TASK on AB to update the documentaion on modules to clarify the usage of the depends-on and pixel size
• TASK on AB to make an NXlog index class for multiple event streams.

Thursday Morning

• Discussion on NXevent_data
• Continue on features

Thursday Afternoon