|
| 1 | +--- |
| 2 | +tags: NumPy |
| 3 | +--- |
| 4 | + |
| 5 | +# 2019-08-28 NumPy Development Meeting |
| 6 | + |
| 7 | +- Time: 11am Pacific Time |
| 8 | +- Join via Zoom at https://berkeley.zoom.us/j/762261535 |
| 9 | +- [Trello workboard](https://trello.com/b/Azg4fYZH/numpy-at-bids) |
| 10 | +- [Previous meetings](https://github.com/BIDS-numpy/docs/tree/master/status_meetings) |
| 11 | + |
| 12 | +**Present:** Matti, Stefan, Tyler, Hameer, Inessa, Warren, Ralf, Sebastian, ZJ |
| 13 | + |
| 14 | +## Follow-up from last meeting / discussions |
| 15 | +- Move multibuild to MacPython and change its name |
| 16 | + Matti is now a maintainer on the multibuild repo |
| 17 | + - move numpy-wheels to the numpy organization |
| 18 | + - move wheel builds [to azure from appveyor](https://github.com/MacPython/numpy-wheels/pull/60) |
| 19 | + Stopped work on this until after the current release cycle |
| 20 | + |
| 21 | +- 1.17.1 and 1.16.5 released!! Documentation at https://numpy.org/doc not updated yet. |
| 22 | + |
| 23 | +- Add a ci-tests-requirements.txt and use it in the CI system. [Working PR](https://github.com/numpy/numpy/pull/14378). Needs a bit of guidance |
| 24 | + |
| 25 | +- Possible NumPy developer weekend: Nov 22-23 maybe move it back to the weekdays before that? Some of the documentation team may join. |
| 26 | + |
| 27 | +- Inaugural NumPy Community Survey (see section below for details) |
| 28 | + - translation to other languages: Zj Poh volunteered. |
| 29 | + |
| 30 | +# Topics |
| 31 | + |
| 32 | +- Refresh the admins of the github [numpy organization admins](https://github.com/orgs/numpy/teams/owners-legacy) |
| 33 | +- Rackspace [wheels](https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com/) and [openblas](https://3f23b170c54c2533c070-1c8a9b3114517dc5fe17b7c3f8c63a43.ssl.cf2.rackcdn.com/) uploads. |
| 34 | + - Matthew Brett has the unencrypted key. Does anyone else? |
| 35 | + - Should be part of the private repo for NumPy management. |
| 36 | + - we should upload arm and powerpc OpenBLAS to there as well (currently on DroBox since Tyler builds them) |
| 37 | + |
| 38 | +- our [information at nabble](http://numpy-discussion.10968.n7.nabble.com/) is outdated. The [admins](http://numpy-discussion.10968.n7.nabble.com/template/NamlServlet.jtp?macro=app_people&node=2&filter=Administrators) can change it, does anyone know who Timmie and Israel are? |
| 39 | + - Maybe try the "send email to" link on their user pages (mattip tried that, "Timmie" seems to be Tim Michelsen?, could not get info on "Israel")? |
| 40 | + |
| 41 | +- Private vs. public submodules in namespace, we're continuing to just add random public-looking things, it's a bit of a mess. What do we do? |
| 42 | + - For `numpy.random` see e.g. [gh-14360](https://github.com/numpy/numpy/pull/14360) |
| 43 | + - Another example: `dir(np.lib.mixins)` |
| 44 | + - We should add a test that hardcodes the current namespace, and will fail if anything is added without a PR. |
| 45 | + - Polynomial namespace, multiple ways of calling the poly functions; Eric Wieser has a [suggestion](https://github.com/numpy/numpy/pull/13601#issuecomment-525591808) on how to fix |
| 46 | + |
| 47 | +- Open PRs, waiting for review/merge: |
| 48 | + - ship our own [numpy.pxd](https://github.com/numpy/numpy/pull/12284) |
| 49 | + - [use a release summary instead of a wall of text](https://github.com/numpy/numpy/pull/13886) |
| 50 | + |
| 51 | +- Nep [website](https://numpy.org/neps) not updating from [repo](https://github.com/numpy/neps), Nep 29 not appearing |
| 52 | + - NEP 30 is showing up, so something is broken with NEP 29? |
| 53 | + |
| 54 | +- Hameer proposed NEP 31: [Context-local and global overrides of the NumPy API](https://github.com/numpy/numpy/pull/14389/files) |
| 55 | + |
| 56 | + |
| 57 | +### Inaugural NumPy Community Survey |
| 58 | +GitHub repo: github.com/numpy/numpy-surveys |
| 59 | + |
| 60 | +Proposal |
| 61 | +github.com/numpy/numpy-surveys/blob/master/NumPy-survey-proposal.md |
| 62 | + |
| 63 | +Topics Overview |
| 64 | +https://github.com/numpy/numpy-surveys/blob/master/NumPy-survey-topics.md |
| 65 | + |
| 66 | +Questionnaire Draft |
| 67 | +https://github.com/numpy/numpy-surveys/blob/master/NumPy-survey-questionnaire.md |
| 68 | + |
| 69 | +This survey shall have up to 40 questions, mainly close-ended to allow easy processing, and take 5-10 minutes to complete. |
| 70 | + |
| 71 | +All NumPy community members are invited to contribute to the creation and selection of questions by submitting pull requests against the [`numpy-surveys` repo](github.com/numpy/numpy-surveys). |
| 72 | + |
| 73 | +## Additional Details |
| 74 | + |
| 75 | +- Matti |
| 76 | + - [cython `@property` decorator](https://github.com/cython/cython/pull/3095), seems to work but needs a refactor |
| 77 | + - reviving dtype allocation and initialization changes in preparation for Sebastian's larger refactoring |
| 78 | + - preparing a [talk](https://github.com/BIDS-numpy/docs/blob/scipy-europe-2019/presentations/inside_numpy_talk_scipy2019.pptx) for EuroSciPy Sept 5 based on Ralf's SciPy talk and a ctypes, cffi, Cython [tutorial](https://github.com/mattip/c_from_python) |
| 79 | + |
| 80 | +- Sebastian |
| 81 | + - [Extensible Datatypes for NumPy](https://hackmd.io/kxuh15QGSjueEKft5SaMug) is pretty far along (good to read through, details probably missing, but some of them might still be in flux). Start to change the basics in NumPy: https://github.com/seberg/numpy/tree/dtypemeta (alpha, trying a few different approaches). Inserting dtype classes works. |
| 82 | + - Next step: wrapping new slots with legacy compatible versions (implement a basic CastingImpl). |
| 83 | + - Redirect current API functions to use new slots (old API functions may be renamed and used as fallbacks for compatibility slot implementations). |
| 84 | + - Slight issue: ABI compatibility of ``PyArray_DescrCheck``, but we can backport a fix to old numpy versions (So compiling with an updated old version of numpy would work). |
| 85 | + - Got lost in Dynamically created Type Objects, so for now "malloc" them and use the typical "static" python API (this seems OK for legacy wrapping, but not ideal?). |
| 86 | + |
| 87 | +- Warren |
| 88 | + - Old pull request triage: closed some abandoned PRs, pinged devs on others., push for close or complete. |
| 89 | + - Bug fix for histogram. |
| 90 | + - Restarted conversation about deprecating the financial functions. |
| 91 | + - Updated PR for the multivariate hypergeometric distribution, waiting for review. |
| 92 | + - Started reviewing ufunc code, and worked on improving the internal code documentation in generate_umath.py. |
| 93 | + - Catch up on Sebastian's NEP / dtype work |
| 94 | + |
0 commit comments