11th March, 2020

Attending: Josh, John, Ralf, Martin, James, Joe in Seattle! (20:33)

• Cancellations

  • PyCon? Not yet.

• Quansight SoW

  • Josh: anything needed? Clarifications?

    • Ralf: looks ok. As well from Travis.

    • @Josh to send gdoc to Quansight & NumFOCUS and publish to github after ok via email

      • Cc Leah & ping on slack

  • James: starting points

    • John: Reading on zarr-spec
    • Josh: Perhaps summary of reading to https://github.com/zarr-developers/zarr-specs/pull/16

• Martin: Xpublish: https://medium.com/pangeo/xpublish-ff788f900bbf

  • Possible to do pre-processing to generate chunks on the fly

  • Discussion versus Intake server happened in pangeo meeting

    • Overlap. Potentially making them more similar to one another.
    • Similarly similar but more generic. Includes catalogs.

  • Joe: Fastapi has its own event loop. James added a workaround

    for jupyter.

    • https://github.com/jhamman/xpublish/issues/7
    • Dask Array: Stackoverflow Question

• Martin will work on generic fsspec backend: github comment

  • Working on zarr open with fsspec. Few additional methods that

    could be useful.

  • Downsides of get_mapper: need to import fsspec, only get raw

    dictionary i/f (no files backend)

  • Josh: similar to

    https://github.com/zarr-developers/zarr-specs/issues/49 ?

    • E.g. detecting NestedDirectoryStorage

• NumPy docs (Josh) → ISSUE

  • https://numpy.org/devdocs/reference/routines.io.html

  • Any reason to not do it? Ralf: nah.

  • Ralf: question of which one relies on which one.

  • Description at the top of the page.

• EOSS Report: Governance, etc. → ISSUE

  • Ralf: need something more than consensus model of numpy

  • Josh: Napari example 2 of 3 tie breaker.

    • Propose 3 of 5 with non-permanent membership
    • Especially review after the first year.

  • John: good to discuss. Happy without BDFL.

  • Committee based: John- conda-forge,

    • Ralf: steering council alone isn’t enough. Naming of core team isn’t great.
    • Suggestion: Steering council (shouldn’t do much), teams as they make sense (cf. matplotlib). Code team → Core team. Perhaps also API team separate since that pulls in other projects like from dask, etc. Smaller concentrated teams seem to attract more people.
    • John: conda-forge has separate for subprojects that they are trusted to just solve.

• Calendar / Subsequent meeting

  • Josh sent James & Ralf to RyanW. for adding to event

• AOB (from xpublish)

  • To go along with zarr.js and xpublish

  • Ralf: Scipy+numpy in Javascript

  • John: https://github.com/scijs

• Done: 20:41 UK