From ab8f91c436c97a631e48cb5675880d9c54a652d1 Mon Sep 17 00:00:00 2001
From: Charles Turner <charles.turner@anu.edu.au>
Date: Thu, 5 Feb 2026 11:09:05 +0800
Subject: [PATCH 1/2] - Add fallback to {} chunking strategy if dataset
 contains non-cftime object variables - Update test to ensure we fall back to
 working chunking strategy in   case above - Change pandas requirements - pin
 to less than 3.0.0

---
 ci/environment.yml   |  2 +-
 intake_esm/source.py | 30 ++++++++++++++++++++++++++++--
 requirements.txt     |  2 +-
 tests/test_source.py |  2 +-
 4 files changed, 31 insertions(+), 5 deletions(-)

diff --git a/ci/environment.yml b/ci/environment.yml
index 98fad538..3c54a113 100644
--- a/ci/environment.yml
+++ b/ci/environment.yml
@@ -16,7 +16,7 @@ dependencies:
   - itables
   - matplotlib
   - netcdf4 >=1.5.5,!=1.6.1
-  - pandas >=2.1.0
+  - pandas >=2.1.0, <3.0.0
   - pip
   - polars>=1.24.0,<1.33.0
   - pooch
diff --git a/intake_esm/source.py b/intake_esm/source.py
index 2104c513..4fb13f2d 100644
--- a/intake_esm/source.py
+++ b/intake_esm/source.py
@@ -104,9 +104,11 @@ def _open_dataset(
         # How should we handle concat_dim, and other xr.open_mfdataset kwargs?
         xarray_open_kwargs.update(preprocess=preprocess)
         xarray_open_kwargs.update(parallel=True)
-        ds = xr.open_mfdataset(url, **xarray_open_kwargs)
+        # ds = xr.open_mfdataset(url, **xarray_open_kwargs)
+        ds = _open_dataset_try_auto(url, xr.open_mfdataset, xarray_open_kwargs)
     else:
-        ds = xr.open_dataset(url, **xarray_open_kwargs)
+        # ds = xr.open_dataset(url, **xarray_open_kwargs)
+        ds = _open_dataset_try_auto(url, xr.open_dataset, xarray_open_kwargs)
         if preprocess is not None:
             ds = preprocess(ds)
 
@@ -137,6 +139,30 @@ def _open_dataset(
     return ds
 
 
+def _open_dataset_try_auto(url, func: typing.Callable, xarray_open_kwargs) -> xr.Dataset:
+    """
+    Try to open a dataset with chunks set to auto. If we fail because dask doesn't know how to chunk
+    it, set chunks to `{}` and retry. Handles cases where datasets contain things like string variables,
+    which can't be autochunked, as that's restricted to cftime arrays only.
+
+    Attempting to autochunk, rather than always using disk chunks (`{}`) is advantageous as it is generally
+    quite a lot more performant. Unfortunately, there is no straightforward way to detect which variables
+    within a dataset can and can't be autochunked without opening it.
+    """
+    try:
+        ds = func(url, **xarray_open_kwargs)
+    except NotImplementedError as exc:
+        if (
+            'Can not use auto rechunking with object dtype. We are unable to estimate the size in bytes of object data'
+            in str(exc)
+        ):
+            xarray_open_kwargs['chunks'] = {}
+            ds = func(url, **xarray_open_kwargs)
+        else:
+            raise exc
+    return ds
+
+
 def _update_attrs(*, additional_attrs, ds):
     additional_attrs = additional_attrs or {}
     if additional_attrs:
diff --git a/requirements.txt b/requirements.txt
index 95039357..70b5c288 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -4,7 +4,7 @@ fsspec>=2024.12
 intake>=2.0.0
 itables
 netCDF4>=1.5.5
-pandas>=2.1.0
+pandas>=2.1.0,<3.0.0
 polars>=1.24.0,<1.33.0
 pydantic>=2.0
 pydap!=3.5.5
diff --git a/tests/test_source.py b/tests/test_source.py
index 32e47ae9..c4e3faa6 100644
--- a/tests/test_source.py
+++ b/tests/test_source.py
@@ -95,7 +95,7 @@ def test_open_dataset_kerchunk(kerchunk_file=kerchunk_file):
     # chunking
     xarray_open_kwargs = _get_xarray_open_kwargs(
         'reference',
-        dict(engine='zarr', consolidated=False, drop_variables='crs'),
+        dict(engine='zarr', consolidated=False),
         storage_options={
             'remote_protocol': 's3',
             'remote_options': {'anon': True, 'asynchronous': _zarr_async()},

From b8a9da8d67f766f5d1adf5a08c014918543d7ea6 Mon Sep 17 00:00:00 2001
From: Charles Turner <charles.turner@anu.edu.au>
Date: Tue, 17 Mar 2026 09:26:06 +0800
Subject: [PATCH 2/2] Pin pandas in `ci/environment-upstream-dev.yml`

---
 ci/environment-upstream-dev.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/ci/environment-upstream-dev.yml b/ci/environment-upstream-dev.yml
index d6c3834d..d8e11285 100644
--- a/ci/environment-upstream-dev.yml
+++ b/ci/environment-upstream-dev.yml
@@ -15,7 +15,7 @@ dependencies:
   - itables
   - matplotlib
   - netcdf4 >=1.5.5,!=1.6.1
-  - pandas >=2.1.0
+  - pandas >=2.1.0,<3.0.0
   - pip
   - polars>=1.24.0,<1.33.0
   - pooch