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

Peloton devs meeting 2026-07-01

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

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

Other activity, July 2026

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

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 👋

🚴 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.

🚴 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.

SciTools Releases

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

🚴 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.

🚴 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 )

🚴 Peloton devs meeting 2026-09-02

Definitely quality discussions over quantity today!

Discussed:

• Avoid variable name true, which is now a keyword in C23 libmo_unpack#25 : Asked external raiser if they could look into fixing it, as it falls outside of the teams current breadth of experience.
• Fixes based on early testing. iris#7216 : An issue about iplt.contourf workaround that gives different and not expected results. We think that this looks like styling and see the data changes possibly as rounding changes.
• Zarr I/O iris#6961 (comment) : Discussed and reply generated. Check out the issue for that info.

Addressed:

• Following discussion of Link Checker Report iris-grib#837 and its many duplicates at a previous Peloton, we generated this fix Set link check timeout to 60 iris-grib#838 and this discussion reviewed that we had indeed fixed the problem and didn't get the same volume of failing linkcheck reports

We also have been looking at our Dragons 🐉. With the number of these that we have we want to focus on minimising WIP and closing issues as not resourced/not planned it we think it is very unlikely that we will be able to get to these within a reasonable timeframe.

Next update will be after Wednesday 9th September.