# March 2020
###### tags: `2020 dev call`

Call co-ordinates: Mondays @ 15:00 Eastern US time (UTC -05:00) https://zoom.us/j/384435716

Previous Notes: [2020 Meeting Agenda](/zljR-pZrQ0O5J_j4NZ-9yw)
Archive: [matplotlib/ProjectManagement](https://github.com/matplotlib/ProjectManagement)

# Needs Discussion At Some Point
- parallel coordinates plot (PCP)
- https://matplotlib.org/devel/MEP/index.html
- GSOC: https://github.com/matplotlib/matplotlib/projects/12
- named roles
- roll-off procedure
- blog PR/merge guidelines
- mission statement/vision
- ownership of notebook backends?
- provisional API
- semantics of when the cycler gets consulted

# March 30
## Agenda
- GSOD: https://opensource.googleblog.com/2020/03/announcing-season-of-docs-2020.html?m=1
- Application: April - May 4
- The path related PRs
- https://github.com/matplotlib/matplotlib/pull/16773
- https://github.com/matplotlib/matplotlib/pull/16812
- https://github.com/matplotlib/matplotlib/issues/16891
- Naming `VectorTransform`: https://github.com/matplotlib/matplotlib/pull/13696
- Units interface: https://github.com/matplotlib/matplotlib/pull/16922
- color map mutability
- https://github.com/matplotlib/matplotlib/pull/16943
- https://github.com/matplotlib/matplotlib/issues/16296

## Notes
@BrunoBeltran @GregLucas joined us. (@anntzer, @story645 @tacaswell, @timhoffman, @qulogic)
### Introductions
- Bruno Beltran working on paths, bio physics grad students
- Jody Klymak, physical oceanographer
- Greg Lucas, aerospace engineer, now doing software
- Antony Lee, bio physicist
- Elliott, geo physicist, RSE
- Hannah, CS grad student doing mpl as dissertation
- Tim Hoffmann, also physicist, works

### Colormap PR(s)
- https://github.com/matplotlib/matplotlib/pull/16943
- https://github.com/matplotlib/matplotlib/issues/16296
- https://github.com/matplotlib/matplotlib/pull/15127

have two problems:
- have color maps that can be mutated
- `plt.get_cmap()` returns a singleton

These two together means you can accidentally mutate the global color map and get your self into trouble

two ways to go:
- make cmaps immutable
- return a copy from `get_cmap`

boundary conditions:
- should be able to get the original color map
- if you want to modify, re-register
- make immutable -> can still re-register

pro for making immutable
- only thing that is mutable is ~the over/under/bad

concern: passing color map by names -> no control without re-registering

levels of immutable-ness:

two things:
- always making 'jet' or 'viridis' always mean the same thing
- should get a copy?
- can defer immutability decision

important to provide a path to re-register global color maps.
`plt.register_cmap(cmap=plt.get_cmap('viridis').set_bad('k'))`

Concerns about the prevalence of the "mutate the global state and mean to" usage due to discussion in https://github.com/matplotlib/matplotlib/issues/12513

Additional wrinkle: `plt.cm.viridis` exists
- maybe handle with module attribute
- replace`cm` in pyplot samespace with a proxy

#### deprecation/implementation plan
In 3.3:
- still return global
- warn on set_*
- add copy kwarg (?)
- make cmap_d warn on read/write and make new private one for our use?

In 3.4:
- return a copy
- don't warn on set_*
- warn on re-accessing a color map that would have been changed

### aside about 3.3 release timeline
- do we have anything blocking 3.3?
- @tim has a big clean up waiting for 3.3 https://github.com/matplotlib/matplotlib/issues/16181
- we have a decent set of new fetures
- concerns that our deprecations windows are too short
- should we block this on the colormaps and paths?
- Hannah volunteers to help with comms on next release

### path related PRs

overall goal: trying to remove inkscape edits

need to get things about bounding boxes etc

this overlaps nicely with the uniform sized makers

plan is to provide the users a way to inject their own paths without having it rescaled

PRs in the stack

(This PR) (*) <- #16832 (*) <- #16859 (*) <- #16888 <- #16889 (*) <- #16891 (MarkerStyle improvements!)
1. https://github.com/matplotlib/matplotlib/pull/16812

This re-arranges things that will pay off in later PRs. @anntzer deprecated a bunch of things which simplified things

The problem is that bezier.py imports path.py, but want to be able to use helpers in bezier.py from path.py.

There are several functions in in bezier.py which are basically methods on Path (but just implemented someplace funny)

suggestion: take the easy (minimal change) here

2. https://github.com/matplotlib/matplotlib/pull/16832

# March 23
- #Covid19 social policy: https://twitter.com/matplotlib/status/1239301597127217152
- Framework Classifier: [16592](https://github.com/matplotlib/matplotlib/issues/16592)
- We have one now: `Framework :: Matplotlib`

## Notes

### ssh config settings for multi-hop

Host ssh_proxy
     HostName something_on_the_edge
     User tcaswell
     ForwardAgent yes


Host *.myplace.edu
    User tcaswell
    ForwardAgent yes
    ProxyCommand ssh ssh_proxy -W %h:%p 22

 ssh -D 8124 -C -vvvv -N [e-mail]

### path refactoring
- https://github.com/matplotlib/matplotlib/issues/15703
- need a richer set of normalization
- need the ability to control the path normalization done in side of `MarkerStyle`
- https://github.com/matplotlib/matplotlib/pull/16773
- This adds the ability to not normalize the path of a custom marker
- hopefully in a way that we can extend in the future
- use a transform instead of a normalization?
- may have to depend on path, might want set of keys - normalization = 'unit'...
- `m = MarkerStyle(sp, normalization='unit-area'); plt.plot(..., ..., marker=m)`
- what @jody is suggesting
- `m = MarkerStyle(sp, transform=None) # does the current thing`
- `m = MarkerStyle(sp, transform='none') # does lookup based on string`
- `m = MarkerStyle(sp, transform=IdentityTransform()) # uses this transfrom`
- `m = MarkerStyle(sp, transfrom=transform_factory: Callable [Path, Transform]) # uses the callable as a factory to get the transform`
- transform is a slightly better name, we already have the name in the code
- would have to decide if these would be applied to the "special" ones
- the pattern of string for lookup or callable is common in our code and numpy/scipy
- https://github.com/matplotlib/matplotlib/pull/16812
- code re-arrangement to support future work on the callables / norms / transforms that will be fed to the code in the previous PR
- currently bezier.py imports path.py (because some of the helpers create `Path` objects)
- all `Path` objects can have the bezier codes in them and bezier.py contains functions that do useful things if your `Path` objects happen to contain bezier code points
-

# March 16
## Agenda
- named roles
- [#15065](https://github.com/matplotlib/matplotlib/pull/15065)

## Notes

### 3.2.1
- Almost out. Elliott and Thomas working on it.

### Governance

- Hannah and Thomas have been working on
- More named roles?
- responsibilities and names?
- BDFL -> Lead Developer(s)
- https://github.com/matplotlib/governance/pull/7
- Avoid implicit jobs versus explicit
- empowers them to make final decisions
- add to resume
- Role timeline
- 1-2 years.
- Allow easy way to stop being in charge
- Number of roles?
- fewer?
- same person does multiple?
- low cost to having more roles?
- small enough to be manageable
- missing roles
- secretary
- treasurer (co-incident with finance sub-committee)

#### Release manager: yes

- per time or per release?
- sign up for X.Y.z series?
- this is how cpython does
- also in charge of deciding backports
- writing release notes and publication
- make sure website gets rebuilt/published as needed
- **per minor release family gig**

#### API breakage shepherd:

- future course of API critical
- clean up and bugfix versus new features
- justify breakage
- new API if we redesign
- lead developer role...
- forward looking and backwards looking very different
- Technical committee?
- need a layer before lead developer as a first line decision / facilitation.
- some things can get lost in indecision circles
- (champions help
- forward looking breaks into two categories
- self-contained new feautres
- long-term API change
- instead of forward / reverse split
- triage / switch yard
- more careful consideration
- different split
- big vs small
- small
- if you add a feature -> need to add a new kwarg / function to make sure they are named correctly and future-proof
- may want to remove kwargs (same level)
- big
- complete overhaul of API
- conclusion **keep to "Small"**
- conclusion **keep routing and facilitation in one**
- conclusion **keep forward and reverse in one role**

#### Technical (reference) docs lead
- docstring validity checking
- properly-formatted
- correct
- **keep and call "reference something"**

#### User (narrative) Documentation lead

- https://mail.python.org/pipermail/numpy-discussion/2020-March/080487.html
- https://labs.quansight.org/blog/2020/03/documentation-as-a-way-to-build-community/
- covers everything in /examples and /tutorials
- covers tutorials in separate repos in org
- **keep and call "narrative something"**

#### Community Manager (communications lead)
- publicity, communications
- Social media: twitter, instagram
- blog (matplotblog)
- recruiting other people to give matplotlib talks/conference organization&involvment
- matplotlib update slide decks
- [ann] mailing list

#### Developer relations
- GSOC/mentoring programs
- 3rd party package interactions & publicity
- discourse - shepherding people to it, making sure questions get answered
- [user]/[dev] mailing list

#### secretary
- manage weekly meeting notes

#### NF laison
- monthly project meetings
- newsletter updates
- signing up for gsoc
- coordinate finance communication
- **leave in lead developer's responsibility**

#### domain specific project leads
- lead their projects mostly as they see fit
- have to use the Matplotlib CoC

#### Lead developer (project leader)
- BDFL must go
- some concern that "lead developer" is too code-focused
- Project leader is more grammatically correct
- suggests leadership, but more collaborative operation

**change to Project Leader**

### [#15065] Drawstyle between
- Discussion about how to manage complexity
- @tcaswell convinced that we should add new public API for this (a Line2D sub-class and `hist_draw()` method an proposed by Jody)
- still up for discussion if we should keep the drawstyle as an implemenattion

# March 9
## Agenda
- meeting agenda
- repo inventory
- team inventory
- named roles
- code coverage
- [#16702](https://github.com/matplotlib/matplotlib/issues/16702): Open Source in CS ED

## Notes
### Agenda
- anyone can add admin, interpersonal, code, etc topics to the agenda
- agenda can be added to in real time
- topics can be added in real time even if not on agenda

### Repo inventory
- Should we archive any more?
- Who is responsible for each of these?
- lives under mpl vs. umbrella orgs?
- flat vs. nested
- Do we want to move tutorials/books/etc to an mpl-teaching org?
Anything with P will be archived in next week or so

  matplotlib/matplotlib              4840 10886
  matplotlib/AnatomyOfMatplotlib     493  777
  matplotlib/basemap                 291  555
  matplotlib/mpl-finance             188  267
P matplotlib/sample_data             100  15
  matplotlib/jupyter-matplotlib      77   650
P matplotlib/sampledoc               75   25
  matplotlib/matplotlib.github.com   41   16
  matplotlib/mplfinance              36   96
  matplotlib/cmocean                 33   131
  matplotlib/pytest-mpl              23   101
  matplotlib/viscm                   22   96
  matplotlib/cycler                  21   23
  matplotlib/GettingStarted          19   17
  matplotlib/interactive_tutorial    18   24
  matplotlib/trendvis                14   112
  matplotlib/grid-strategy           13   24
P matplotlib/natgrid                 11   12
  matplotlib/mplcairo                11   40
  matplotlib/matplotblog             9    18
  matplotlib/mpl-probscale           8    19
  matplotlib/governance              7    0
  matplotlib/mpl-altair              6    6
  matplotlib/mpl-docker              6    1
P matplotlib/freetypy                5    3
A matplotlib/mpl_mac_testing         4    0
  matplotlib/devdocs                 4    1
  matplotlib/legacycontour           4    5
A matplotlib/CZI_2019-07_mpl         4    3
P matplotlib/fcpy                    3    0
  matplotlib/mpl-bench               3    0
P matplotlib/ProjectManagement       3    0
P matplotlib/mplsizer                2    0
P matplotlib/matplotlib-jenkins      2    0
  matplotlib/presentations           2    2

Created a new github organization 'matplotlib-teaching', made Hannah an owner.

@tacaswell will reach out to original trendvis author to see if they are are still interested in maintaining or if we should look for a new maintainers.

Discovered you can not restirct who can make new repos -> need to invistgate outside contirbutors for triage rights.

### Teams inventory
Proposal:
- SC (admin on governance)
- Core (only team with write access to matplotlib/matplotlib)
- Projects
- domain
- project
- teams
- communication and teaching
- blog (admin on blog)
- teaching (write/admin on tutorials)
- docs (write access on website)

**@tacaswell Will re-org teams this week,also sort which repository belongs to which team**

### CS Classes
- Keep on w/ Toronto
- our biggest limitation is review capacity
- on one hand, having a policy makes clear we won't do anything special for course
- on other hand, don't want to give people ideas
- concerns that these students will be less-sticky than an average new contributor
- also concerns that students won't follow through after the project is over
- Arguments:
- review capacity + loosened merging rules
- Mpl is not well scoped or well constrained, which can be hard for school projects
- Expectations for new contributors
- intrinsinc motivations,
- motivated to help GSOC/small bursts/occasional new contributors
- you know that there's a funnel down to significant contributors
- too much work: one time contributions that need to be hand held throughout the process

#### Outcomes
- The issue is the documentation
- Email Anya: Heads up + gating please
- Tom post at bottom w/ sum up of heads up + pre-filtering for good PRs that are likely to make it in w/o hand holding and that students/profs will push through even after class ends if necessary. Welcoming but not as unpaid as TAs.

# March 2
## Agenda
- RSE announce blog
- meeting note babysitter
- GSOC
- https://github.com/matplotlib/matplotlib/projects/12
- Small dev grant: who to hire/how to spend the money
- Consolidating gitter channels
- delete/archive gsoc, mpl-altair, community since they're dormant
- named roles
- using assign
- color cycle turn on turn off
## Notes
### RSE Hiring
- We have hired Elliott (qulogic)!
- https://paper.dropbox.com/doc/announcement-blog-post-E5Qm4GMpVEwGxt1DmQ2sQ (need to finish)
### meeting note baby sitting
- need to manually sync to github at end of meeting
- --- (menu) -> versions->sync -> push
- 

- need to be added to project management repository
- need to create a new link / file each month
- maybe move to bigger time windows?
- update the year file that links out to each month
- set sharing/editing permissions & publish:
- 

### GSOC
- @anntzer's rolling image comparison project
- see Iris for some perceptual image comparison setup https://github.com/SciTools/iris
- I think https://github.com/SciTools/iris/pull/2206 was the PR that introduced it
- [their docs on graphics tests](https://scitools.org.uk/iris/docs/latest/developers_guide/graphics_tests.html#developer-graphics-tests)
- official GSOC page https://github.com/matplotlib/matplotlib/projects/12
- maybe fonts, but low priority
- maybe detexify (http://detexify.kirelabs.org/classify.html) for navigating the library
-
### Using assign

- ask folks if we can assign issues/PRs to them?
- assign things so they have a champion to move them along
- ask folks to (agressively) self-assign
- expectation is that if you comment on a PR and don't self assign you have weak opinions.
- also indicated by request-changes vs just a comment review
- "just comments" -> means "I won't block on this"
- these things make sense for small PRs, may not work well for big PRs
- may have a significant comment, but not be willing to take over championing the PR
- https://doist.com/blog/decision-making-flat-organization/
- like how pydata ranks proposals
- -2 (will fight)
- -1 (against, but meh)
- 0 (neutral)
- +1 (pro, but meh)
- +2 (will fight)
- this missing the pre-consensus step of "please give more information"
- start a -2 and be willing to be moved?
- issue is more getting engagement at all (not reaching consensus once everyone is engaged)
- how do we get people engaged?
- ping them directly?
- dev list?
- deal with over-engagement
- assigned champion can moderate
- but this is the minority of cases
- use draft PRs in lieu of self-assigning on issues in some cases
- or both
- not going to solve the general problem of over-notified people
- concern is less about people on calls, more about making sure it happens for new contributors.
- like the idea of a sponsor for the PR
- losing good possible contributors due to their PRs languishing
- should have some sort of default timeout on PRs
- do we want a bot?
- start with 12-18mo timeout
- assigner does not have to do the review / decsion, just responsible for shepherding
- the assigment can be handed off
- due to force-pushing master last month, all PRs were last touched 23 days ago
- set up bot to start timing things out at 6 months, gives us 5 months to keep things alive
- hope the bot gives a warning message?
- @hannah will research bots

### provisional API