Support for copernicusmarine package in Pyodide/JupyterLite environment (import failing due to incompatible native dependencies)

[dbazin-eo]

Summary

I would like to request support for installing and importing the Copernicus Marine Toolbox (copernicusmarine) (https://github.com/mercator-ocean/copernicus-marine-toolbox) within the JupyterLite/Pyodide-based environment provided by climet-eu/lab.

This toolbox is essential for subsetting and accessing Copernicus Marine datasets directly within browser-based training notebooks. However, attempts to install or import the package currently fail due to missing/unsupported dependencies in the Pyodide runtime.

---

Context

I attempted installation using both:

• micropip.install("copernicusmarine")
• %pip install copernicusmarine

In both cases, the installation step completes (or appears to), but the import fails due to underlying binary dependencies such as:

• pydantic_core
• orjson
• jsonschema
• Dynamic libraries (*.so) not supported or failing to load in WASM

Typical errors include:

ImportError: Could not load dynamic lib: ...
Error: bad export type for '_ZN64...'

and:

Error: bad export type for '__rust_no_alloc_shim_is_unstable': undefined

These errors indicate that the dependencies required by copernicusmarine rely on native extensions or Rust-based modules that are not yet available in a Pyodide-compatible build.

---

Reasons, motivations and use cases

• The copernicusmarine client is the official Python interface for downloading, subsetting, and searching datasets from the Copernicus Marine Service.
• It is a key dependency for browser-based training notebooks, especially for large-scale training (e.g., thousands of participants) where zero-install environments such as climet-eu lab are essential.
• Without it, users cannot interact directly with Copernicus Marine data from within the browser notebook.

---

Required Dependencies

The package currently depends on libraries known to be problematic in Pyodide unless precompiled for WASM, though most of the packages are already available in climet-eu lab and it's already a very nice work from you 🙏

• pydantic>=2
• pydantic_core
• orjson
• jsonschema
• pystac
• h5py, zarr, numcodecs
• numpy, pandas, xarray, fsspec

The full list is available from the official documentation: https://toolbox-docs.marine.copernicus.eu/en/stable/installation.html#dependencies

Providing Pyodide-compatible builds for these appears necessary for successful import.

---

Request

Could you please:

Add Pyodide-compatible builds for:

• copernicusmarine
• Its core dependencies (pydantic_core, orjson, jsonschema, pystac, etc.)

Or

Provide a curated pyodide-lock.json including the full dependency graph precompiled for WASM.

Or

Document whether supporting this package is currently blocked by Rust-based dependencies.

---

Reproduction Steps

Using %pip:

%pip install copernicusmarine
import copernicusmarine

Using micropip:

import micropip
await micropip.install("copernicusmarine")
import copernicusmarine

Both result in import errors due to incompatible shared objects.

---

Environment

• JupyterLite via https://lab.climet.eu/
• Pyodide kernel
• Python 3.13 (Pyodide)
• Client-side execution (WASM)

---

Expected Outcome

Be able to import and use:

import copernicusmarine
temperature_climatology = copernicusmarine.open_dataset(
  dataset_id="cmems_mod_glo_phy_my_0.083deg-climatology_P1M-m",
  variables=["thetao", "zos", "so"],
  minimum_longitude=4.14,
  maximum_longitude=8.29,
  minimum_latitude=54.69,
  maximum_latitude=60.46,
  minimum_depth=0,
  maximum_depth=5000,
)

and work directly with Copernicus Marine datasets:

print(temperature_climatology)

---

Thank You

Adding support for this package would greatly benefit browser-based training workflows.

[juntyr]

Thank you for the report @dbazin-eo!

I've spent a bit of time looking into the issue. The problem doesn't seem to be related to specific packages but seems to be caused during the loading of dynamic library dependencies. The linking errors suggest that something has gone drastically wrong. I unfortunately don't know what yet exactly. I'll keep looking :)

[dbazin-eo]

many thanks to you @juntyr and for the awesome work made at climet-eu/lab! feel free to let keep me posted, this use case would be very interesting for teaching and research activities 🙏

[juntyr]

One of the issues was with the orjson package, which seems to have been fixed in upstream Pyodide with this patch https://github.com/pyodide/pyodide-recipes/blob/dc9e34110f81e04e3e91436de92b576d175d6d88/packages/orjson/patches/0001-Fix-_PyLong_AsByteArray-function-signature.patch.

I'm currently starting the upgrade to the latest Pyodide and the fix will then be included there.

[dbazin-eo]

Hey @juntyr 👋
Many thanks for the update and most importantly for the work being done! So great to read you found the root cause 💪 I'll run test/validation whenever it's appropriate, feel free to keep me posted 🙏

[juntyr]

I wanted to give a brief update. I've been working on upgrading to Pyodide v29. The orjson package now loads correctly, which is promising. Unfortunately, I'm experiencing some issues with the pyarrow dependency, but once that is fixed I will inform you so that you can then test copernicusmarine inside the development version of the online laboratory.

[dbazin-eo]

hey @juntyr 👋 oh sounds great! looking forward to giving it a try! 🙏 I'll report here once you send me the green light 🫡

[juntyr]

I managed to make some more progress and on https://lab.climet.eu/main it's now possible to run

%pip install copernicusmarine
import copernicusmarine

For loading the dataset you need a username and password, maybe you could try out if it works for you?

[juntyr]

copernicusmarine is now part of the pre-resolved venv on https://lab.climet.eu/main, so you can directly import copernicusmarine without needing to install it first

[dbazin-eo]

Hi @juntyr 👋 many thanks for that! I was not sure you could achieve this! It's awesome 👏

I do confirm the import is working as expected, you did a great job!

I've ran a test case in a notebook just to simulate fetching data from the Marine Data Store and for the moment:

• (of course and thanks to you🙏 ) import works ✅
• login works ✅
• describe catalog metadata does not work ❌
  (hits a RuntimeError: can't start new thread)
• subset and get data does not work ❌
  (hits a OSError: [Errno 50] Protocol not available)

By any chance, do you have any idea about what could be NOK when looking at the (more detailed) tracebacks from the attached notebook? 🙏
(I've tried some troubleshoots to exclude hypothesis)

climet-eu_copernicusmarine.ipynb

To reproduce the notebook on your end, you could use these that burns after first read: https://pastebin.com/TXaYp6yW

[juntyr]

Thank you for your help debugging!

I've opened mercator-ocean/copernicus-marine-toolbox#459 to avoid the thread pool, which simply doesn't work inside Pyodide.

The second problem is that urllib3 hasn't been patched fully yet for the online lab, I suppose I need to get to that next :) [for context, requests and aiohttp have already been patched].

I'll get back to you once I've made some progress with urllib3, perhaps you can push the above PR forward and see that we get a release with it?

[dbazin-eo]

Alright sounds great to me @juntyr 🙏
Quick (noob) question, would making progress with urllib3 help in any way about the error raised by botocore?
I'm asking you this because during training, users are encouraged to use the .open_dataset() function of the Toolbox for its simplicity (no file manipulation) and low bandwidth consumption (lazy loading and download only what the users needs).
Many thanks for your insight in advance

[juntyr]

My understanding of the error so far is that botocore uses some advanced features of urllib3 that are not yet supported with Pyodide. So I need to see if botocore can be configured to use the simpler options or if urllib3 support could be extended