|
| 1 | +Zarr v3 design update |
| 2 | +--------------------- |
| 3 | + |
| 4 | +Alistair Miles ([@alimanfoo](https://github.com/alimanfoo)) |
| 5 | + |
| 6 | +19 June 2019 |
| 7 | + |
| 8 | +==== |
| 9 | + |
| 10 | +### Current status |
| 11 | + |
| 12 | +* Development is ongoing via the core-protocol-v3.0-dev branch in the |
| 13 | + [zarr-specs repo](https://github.com/zarr-developers/zarr-specs), |
| 14 | + follow |
| 15 | + [PR#16](https://github.com/zarr-developers/zarr-specs/pull/16) for |
| 16 | + current status. |
| 17 | + |
| 18 | +* Rendered docs from this branch can be [viewed on |
| 19 | + RTFD](https://zarr-specs.readthedocs.io/en/core-protocol-v3.0-dev/). |
| 20 | + |
| 21 | +* This is still a straw man, everything is up for discussion. |
| 22 | + |
| 23 | +==== |
| 24 | + |
| 25 | +### Modular spec architecture |
| 26 | + |
| 27 | +* [Core protocol spec](https://zarr-specs.readthedocs.io/en/core-protocol-v3.0-dev/protocol/core/v3.0.html) |
| 28 | +* [Protocol extension specs](https://zarr-specs.readthedocs.io/en/core-protocol-v3.0-dev/protocol/extensions.html) |
| 29 | +* Codec specs |
| 30 | +* Storage specs |
| 31 | + |
| 32 | +=== |
| 33 | + |
| 34 | +### [Core protocol spec](https://zarr-specs.readthedocs.io/en/core-protocol-v3.0-dev/protocol/core/v3.0.html) |
| 35 | + |
| 36 | +* Minimal, easy as possible to do full implementation in any language. |
| 37 | + |
| 38 | +* Aiming for intersection of N5 and Zarr v2 features. |
| 39 | + |
| 40 | +* Defines a variety of **extension points** so can also serve as a |
| 41 | + foundation for growth and experimentation. |
| 42 | + |
| 43 | +=== |
| 44 | + |
| 45 | +### Protocol extension specs |
| 46 | + |
| 47 | +* Each protocol extension gets its own spec. |
| 48 | + |
| 49 | +* Currently the core-protocol-v3.0-dev branch also includes some |
| 50 | + protocol extension specs, to illustrate the concept. |
| 51 | + |
| 52 | + * E.g., [Datetime data types |
| 53 | + spec](https://zarr-specs.readthedocs.io/en/core-protocol-v3.0-dev/protocol/extensions/datetime-dtypes/v1.0.html) |
| 54 | + |
| 55 | +* Ultimately these extension specs should get split out each into a |
| 56 | + separate development branch, so we decouple them from the core |
| 57 | + protocol spec. |
| 58 | + |
| 59 | +=== |
| 60 | + |
| 61 | +### Codec specs |
| 62 | + |
| 63 | +* Each codec intended for use as a compressor or filter gets it's own |
| 64 | + spec. |
| 65 | + |
| 66 | +* @@TODO create an example to illustrate the concept. |
| 67 | + |
| 68 | +=== |
| 69 | + |
| 70 | +### Storage specs |
| 71 | + |
| 72 | +* Each concrete storage system (e.g., file system, cloud object |
| 73 | + storage, Zip file, LMDB, ...) gets it's own spec. |
| 74 | + |
| 75 | +* @@TODO create an example to illustrate the concept. |
| 76 | + |
| 77 | +==== |
| 78 | + |
| 79 | +### Spec development process (1) |
| 80 | + |
| 81 | +* Currently @alimanfoo is acting as editor for the v3.0 core protocol |
| 82 | + spec. |
| 83 | + |
| 84 | + * Feedback/comments/ideas welcome from anyone at any time |
| 85 | + ([PR#16](https://github.com/zarr-developers/zarr-specs/pull/16) is |
| 86 | + probably the best place for comments). |
| 87 | + |
| 88 | + * Still in a conceptualisation phase, no need for formal decision |
| 89 | + process as yet. |
| 90 | + |
| 91 | +=== |
| 92 | + |
| 93 | +### Spec development process (2) |
| 94 | + |
| 95 | +* Ultimately I think we'll need to define a community process for spec |
| 96 | + development, so that: |
| 97 | + |
| 98 | + * It's clear how others can contribute. |
| 99 | + |
| 100 | + * It's clear how decisions get made. |
| 101 | + |
| 102 | + * It's clear what stage of maturity each spec is at. |
| 103 | + |
| 104 | +* Postpone that for now, may need external advice/help on best |
| 105 | + approach. |
| 106 | + |
| 107 | +=== |
| 108 | + |
| 109 | +### Spec development process (3) |
| 110 | + |
| 111 | +* Hopefully the [zarr-specs |
| 112 | + repo](https://github.com/zarr-developers/zarr-specs) can serve as a |
| 113 | + focus for community spec development. |
| 114 | + |
| 115 | +* However, don't want to force all spec development down the same |
| 116 | + route. |
| 117 | + |
| 118 | + * This is one reason why currently the core spec makes use of |
| 119 | + **URIs** in metadata to refer to protocol extensions and codecs - |
| 120 | + allow for anyone to publish their own spec. |
| 121 | + |
| 122 | +==== |
| 123 | + |
| 124 | +### |
0 commit comments