SciTools core developer diary 2026

[trexfeathers]

News of the the SciTools core developers' work during 2026

• Finished work
• Current work
• Planned work

Tip

Click the Subscribe button to get automatic updates as soon as we post them 🔔

[trexfeathers]

May

• user support surgery : 8 questions
• peloton devs meetings :
  • May 13
    • decided to deprecate the iris curl function
  • May 27
    • We started our first development sprint for iris 3.16 from 26th May (8 weeks scheduled)
• Ran a dask "dash" : investigation for team learning and advice improvement
  • 2 PRs merged on Dask project !
    • Fix LaTeX formula for split_every in docstrings by stephenworsley · Pull Request #12379 · dask/dask
    • Fix typo in array-chunks.rst by HGWright · Pull Request #12375 · dask/dask
  • Implemented a speed-up to PP Fields loading · Pull Request #7089
• assisted with ESMValTool User Workshop at UK MetOffice, May 11-13
• @rcomer @bjlittle @ESadek-MO met to discuss dev progress with tephi
• Team upskilling on Agentic Development 🤖⚡
• Big 💚 to @SgtVarmint for their first-contribution to iris · Pull Request #7087
• A productive Dragon Taming session! 🐉🐲
  These refinement sessions are essential for prioritising and planning upcoming large projects
  • JOSS Paper Submission  bjlittle/geovista#389
  • Allow global attributes on a Mesh iris#6085
  • Regridder Partitioning iris-esmf-regrid#427
  • Typing support iris#5924

[trexfeathers]

June

• User support surgery: 10 questions
• Preparation and delivery of Momentum User Training.
• Finishing a three week Iris sprint for Iris 3.16
• Undertook a three day dash on agentic development:
  • Opened X pull requests:
    • DOCS: Add sitemap support iris#7100
    • DOCS: Added llms.txt support using sphinx-llm (replaced sphinx_llms_txt) iris#7105
    • Add AGENTS.md iris#7160
    • [Agentic] Address warnings in tests - part 1 (Phases 1 - 3) iris#7161
    • docs: adopt towncrier changelog iris#7166
  • and updated issue Improve utility of warnings in Iris tests iris#5466 to include an agent generated plan.
  • Created a new Agentic label in Iris and GeoVista.

Peloton devs meetings:

• 3rd June
  • Concrete decision to NOT action Should our handling of STASH objects be rationalised? iris#6286
  • Our repos are entering the Brave New World™️ of Agentic Pull Requests 🚀
    • WIP: Feature/depth dimension bjlittle/geovista#2294
    • 8 PRs from @gaoflow on Iris: link
• 10th June
  • Discussed the merits of fork based PRs versus branch based PRs.
  • Closed two PRs:
    • Closed Improvements to GitHub Actions package caching iris#6803 in favour of exploring Pixi as an alternative: Adopt Pixi for reliable 'partitioning' of locked core and optional dependencies iris#6833
    • Requested Support for environment.yml/lockfiles SciTools-incubator/conda-rpms#26 to be closed as use of "conda rpms" is no longer relevant.
• 24th June
  • Discussed linkcheck failures across all SciTools repositories
  • Closed Dask Best Practice issue Update Dask Best Practice docs iris#6304
  • We will have @pt331 working with us for a month
    • First issue! iris_course out of date links and phrasing scitools-classroom/courses#197

[trexfeathers]

@bjlittle and @trexfeathers just finished a pair-programming marathon hardening Iris' GitHub Actions using Zizmor. Among dozens of useful security tips, the highlight was probably "Dude, just use the gh CLI for this. No way do you need that marketplace action!" ❤️.

• Adopt zizmor iris#6809
• Adopt zizmor for secure GHAs iris#7138
• Some spill-over into other SciTools repos:
  • adopt zizmor workflows#191
  • zizmor: follow-up fixes workflows#193
  • Update GHA for workflows 2026.06.0 iris-grib#798
  • gha: bump for latest scitools/workflows nc-time-axis#506
  • gha: bump for latest scitools/workflows python-stratify#429
  • Update templates from SciTools/iris#7138 #266

[trexfeathers]

July 2026

[trexfeathers]

Iris documentation improvements

Thanks to @bjlittle for these!

• docs: adopt towncrier changelog iris#7166
  Hear ye 🔔! A modernisation of how we manage our What's New pages, using the Towncrier tool. The pages retain the existing format that we love, but under the hood:
  • No worrying about correct paragraph formatting
  • No Git conflicts!
• docs: add custom readingtime directive iris#7150
  The majority of our docs pages now give you a heads-up on how long it will take to read them (approximately)

[trexfeathers]

More agentic goodness

• Finalised the Iris AGENTS.md file: Setup of AGENTS.md iris#7175
• Merged several more agentically generated PRs. Thanks @gaoflow!
  • Allow printing of a length-0 coordinate (#6531) iris#7122
  • Copy target coordinates onto regridded result cube (#6844) iris#7124
  • Allow concatenation of string coordinates of differing width (#6676) iris#7125
  • Keep lazy_elementwise blocks as arrays for scalar-returning ops (#6965) iris#7126

[trexfeathers]

iris-esmf-regrid v0.15 release!

Some key stuff in this release - click the link for more 😊

[pp-mo]

Peloton devs meeting 2026-07-01

• Guest collaborators: @pt331, @2319bli, @GitRylee, @king56180468-droid

[pp-mo]

Peloton devs meeting 2026-07-08

• considered multiple linkcheck failures, notably those for netCDF docs
  decided we should probably adopt the lychee "--insecure" option : think that's safe for only checking link
  • see Consider accepting SSL insecurity warnings for linkchecks #271
• reviewed Refactor bad practices in Iris' tests iris#6625
  • ticking a few of these off now
  • recent work experience students contributed to these efforts, e.g. this PR and this one
  • still work in progress
• discussed whether we still want to address Simplify netcdf loading rules/actions iris#6316
  • a legacy of unnecessary complexity in the netcdf loading code
  • basically "yes" : made it a Dragon
• decided to ignore the small performance drop associated with making the median aggregator lazy Performance Shift(s): 869691f8 iris#6290
• labelled a bunch of legacy warning issues going stale as : to be addressed by [Agentic] Address warnings in tests - part 1 (Phases 1 - 3) iris#7161
  DeprecationWarning in pp file saving iris#5492
  Resolve numpy deprecation: array conversion into scalar. iris#5531
  Resolve numpy deprecation: using matrix subclass instead of recommended ndarray iris#5532
  Resolve runtime warning: invalid value encountered in cast iris#5533
  Resolve runtime warning: divide by zero encountered in divide iris#5534
  Resolve runtime warning: invalid value encountered in sqrt. iris#5535
• agreed to not pursue small performance hit Investigate cases where averages are slightly higher than input data. iris-esmf-regrid#542
• agreed that Change default pole_method when source grid does bounds do not include the pole iris-esmf-regrid#378 is still worth pursuing
• discussed and replied to @arjclark comments on Make lazy data the default iris#6591 about current and future data model re lazy data

[pp-mo]

Other activity, July 2026

• Delivered a seminar on user support at the UK Met Office
• User support Surgery: 22 questions

[trexfeathers]

Peloton devs meeting 2026-07-15

• Discussed how we ought to implement a policy on AI contributions
  • Would be good to have a top level capitalised markdown file in repositories
  • Could include in CODE_OF_CONDUCT.md, but our preferred template has nothing so far
  • Could follow GeoVista's lead with a heading within the docs: Generative AI
  • GitHub does not yet recognise an 'official' file name like it does with CODE_OF_CONDUCT or CONTRIBUTING
• Revisited the confused documentation around constraints and callbacks: Disentangling the order in which constraints and callback are applied iris#3944
• Celebrated a new contributor! @01luyicheng submitted Add unit tests for Cube.coord_dims iris#7198
• Welcome @gavindavies to our developer team 👋

[ESadek-MO]

🚴 Peloton devs meeting 2026-07-22

Discussed

• Implement variable-length chunksize hinting for netcdf loads iris#6358: We might make variable length strings work by default, link the issue to a chardata one
• [pre-commit.ci] pre-commit autoupdate ncdata#224: isort versioning includes rcs and alpha releases. We think this is something specific to isort; perhaps replacing it with Ruff would be the best solution.
• Consider moving requirements/iris.yml to environment.yml iris#7205: We like this idea, and it shouldn't cause any complications either.
• Reorder pre-commit hooks so that ruff-format and ruff-check happens before mypy iris#7204: Ruff checks should be before MyPy checks in case of reformatting. Templates will also need updating for this, with a comment.
• Update RTD server  iris-grib#777: Mamba, miniforge and their kin are all really opaque. Pixi seems to be the best solution here, but would take a lot of work.
• Allow creation of area weights for Lambert Conformal Grids iris#4812: Was solved a month ago, with no further complaint, and so now closed. User can reopen as needed.

Prioritised for Review

• Fix type hints for iris.cube.Cube arguments iris#7202
• ruff compliance for TC iris#7203
• Add unit tests for Cube.coord_dims iris#7198

Flagged for follow up

• docs: fix duplicated words in iris_cubes, s3_io, and release docs iris#7088: Waiting for the CLA to be signed before we can review.

[HGWright]

🚴 Peloton devs meeting 2026-07-29

Discussed

• Gdt40 save iris-grib#813
• Add step argument to Cube.rolling_window iris#7217
• Are _MeshIndexSets still successfully lazy pointers to original arrays? iris#7221
• Understanding and Controlling what area weights are used when doing Area-Weighted regridding iris#7228

Prioritised for Review

• Merge _MeshIndexSet feature branch iris#7210
• Support utf8 by default for saving strings. iris#7230

Other News

• Gave a Met Office internal Demo for the Iris 3.16 release.

[ukmo-ccbunney]

SciTools Releases

• Iris-grib v0.23.0
• Iris v3.16 (Release Candidate)

[ukmo-ccbunney]

🚴 Peloton devs meeting 2026-08-12

General

• "Templating" chores need to be addressed on each repo by primary leads.
• Discussed possibility of an "Open Peloton" in October; see #6991
  • Will be open to both internal and external collaborators and interested parties.
  • Planned for the 30th October 2026.

PRs

• Fix geotiff export geotransform values for numpy >= 2.4 iris#7140 - leaving this to close? No CLA signed.
• Two PRs for same issue raised within 15 mins of each other!
  • Add step to Cube rolling windows iris#7232
  • Add option for steps in Cube rolling windows iris#7236
  • We will likely merge #7236 as the other is still in draft.

Discussion:

• Discussed the Good First Issue label - they are getting jumped on very quickly by AI generated PRs.
  • Numpy have changed their "Good First Issue" label to be less discoverable by Agents, should we?
  • We don't want to put off contributions, but also don't want "good first issues" to be gobbled up by agents - we want to encourage new developers to build their skills.
  • Should we have a policy of declaring whether a PR is AI generated?
  • This could be an entry in AGENTS.md?
  • Could we "hold" an AI generated PR for a period (1 month?) before merging to allow others a chance to contribute?
• Where we will advertise these developer diary entries?
  • Mention in all SciTools repo docs and READMEs; Advertise developer diary #274
  • Release announcements:
    • Internal: Viva Engage, SharePoint, etc.
    • External: Bluesky

Other News

• AVD User Support Surgery on 5th August 2026
  • 8 questions (3 in-person)

[arjclark]

RE: the Good First Issue and Agentic AI problem, we seem to be mitigating that at the moment in the ANTS repos by having a leading emoji in the labels (which ironically is a characteristic of AI generated titles). Thus far we think there's just hard string matching for "good first issue" as a label going on that's being done by people instructing agents. No doubt it will change in future though.

Experiments thus far with .md files on my/our end seem to suggest that a "don't" in one of those files is always overridden by a user instruction to "do" a thing so you could put "don't raise PRs for good first issues" but if someone instructed an agent to seek out and raise PRs for "good first issues" on python repos (for example) it'd likely still happen so little to no protection is offered on this (and I've no idea what happens in the agents instructing pools of agents paradigm).

Definitely interested in following what you end up doing in this space

[ESadek-MO]

🚴 Peloton devs meeting 2026-08-19

Discussed:

Some long discussions, which I won't go into detail here. The highlights are:

• Real world Mesh is not indexable using Iris iris#7246 - We're unsure if this is a problem with the file, or with Iris.
• Incompatibility with xxhash 4.0 iris#7251 - We think this is important to look into, and an easy enough fix.
• Address slow loading with time-varying reference fields iris#7250

Addressed:

• Remove dead urls from docs iris#6387 - Closed
• pre-commit alpha, beta and rc releases are being included in updates. We've agreed to manually edit these PRs and pin them to the latest stable release.

[pp-mo]

🚴 Peloton devs meeting 2026-08-26

Discussed:

• Allow mesh partitioning iris-esmf-regrid#686 : Needs reviewer, assigned to @HGWright
• Varstrs ncdata#254 : proving easier than expected, - but just PoC (needs more testing + docs). Waiting on @pp-mo to finalise + submit
• Address slow loading with time-varying reference fields iris#7250 : discussed. We expect that a time-dependent reference (e.g orography) will also load OK from netCDF, but this needs to be tried (possible for next week's "Surgery" event)
• Bump sphinx-llm>=1.0.0, exclude gallery timing page from build iris#7258 : proposal looks OK, can enable checks, but will fail as we have outstanding lockfile problems
  assigned to @ESadek-MO
• Minimum supported macOS version cf-units#690 under investigation, @HGWright assigned

Addressed:

• repeated timeouts from Iris CI linkchecker on the ecCodes homepage, e.g. https://github.com/SciTools/iris-grib/actions/runs/32623905694
  Currently failing about once a day, but visits fine in browser.
  Agreed to try applying an extended timeout : **@trexfeathers Set link check timeout to 60 iris-grib#838 probably fixed
• Partitioning src_chunks size is severely limited by ESMPy memory demands iris-esmf-regrid#683 : ongoing discussion from user confusion
  • only solution is to set the partition size right, and it seems Dask chunk sizing is not a good hint for this
  • maybe a case for better documentation, but we don't think we can easily address it : closed
• Add step argument to Cube.rolling_window iris#7217 : had a "collision" of proposed solutions, but now resolved
• Computational performance of iris.cube.Cube.aggregated_by with lazy data iris#5455 : @bouweandela has re-asserted need
  we reviewed our options, but it remains: we would need to rework the way aggregations are constructed, and it isn't clear there will be a simple "better way"
  closed as don't have resource
  ( but have also said we can still discuss whether they could implement this, if wanted )