23rd October, 2019

Attending: RyanW, JoshM, WardF, JohnK, Stephan Saalfeld (20:09), TonyT (20:20)

  • Deliverables & deadlines

    • Current setup

      • Someone at Mt. Sinai and the rest to Quant*
      • SciPy: share accommodations?
      • Overhead for NumFOCUS and Mt. Sinai
    • Cf.


      • John: Quant’s Ralf, community outreach
      • Stephan: like ITK’s MattM
    • Stephan:

      • V2 backend worked on during the summer
      • V3 looks to be a subset of N5, can coexist for a while
    • Ryan:

      • Leave spec related things

      • Implementation best-effort (stretch the goals; towards the end)

      • More towards spec designers (hard to get external person)

      • Josh: but perhaps for implementing the TCK

      • John: sprints together to focus? (another version)

        • NumFOCUS can take reimbursements
    • 10:30 finish up? John: more!

    • To cut/change:

      • Cut Julia [2.2]

      • C++ [2.4] becomes C/C++

        • Josh: may need to split NetCDF library to work like N5


        • Stephan: v3 isn’t layered as needed by C or Java. Can’t

          build on top. Would love to see more supercore, core, then the other stuff.

          • Or perhaps we’re just saying, “Extensions are different stuff”

          • E.g. compression schemes - uses numcodecs but API is clear (Mem in; Mem out). Great.

            • But: Filters in V2. Unclear how they are

              parameterized, named, etc.

          • Chunk storage in V2. Struct formatting language. Can build new types. Makes sense.

            • But: In V3 we don’t have this. Can’t do the

              var-length chunks.

          • Attributes interface: spec says “JSON”, but extensions are possible, and then there’s no common interface.

        • TBD: in first couple of months.

        • John: extension API? Will come. SS: Not natural for

          Python thinking. JK: but MutableMapping is one counterexample.

        • Q to Ward: interface for C? WF: can speculate but Dennis

          isn’t here.

      • Josh: cut some of governance (do that via the community)?

        • Cut [3.1-3.3]
      • Josh: happy to keep 4.1 & 4.2

  • One vs many discussions (Josh) - 20:48

    • Various people are asking: Zeiss, Napari, Fiji, etc.

    • Stephan: HDF5 has less problems than a tarball (cf. N5/BDV)

    • John: “just implement mutable mapping”

    • Josh: is the proposal that add a HDF5 storage and them them to

      use it? (i.e. what communities like BDV, OME, NetCDF are already looking to do)

    • John: there are some backends. Don’t know what works for their

      use case. Need to do benchmarking. Perhaps something for the position to do.

    • Stephan: to let someone be able to do it.

    • (Tabling Python angle for next time)
  • Next meeting

    • John: Re-do doodle?

    • Stephan: don’t need to also be there

    • Ryan: happy to do every two weeks right now

    • Record?

  • AOB (likely tabled)

    • Roadmapping

    • Any ongoing issues: permissions, etc.