Here is a more complete list (code-generated):

• apply
• argmax
• argmin
• as_numpy
• assign_attrs
• assign_coords
• astype
• bfill
• broadcast_equals
• broadcast_like
• chunks
• clip
• coarsen
• combine_first
• convert_calendar
• cumulative
• cumulative_integrate
• curvefit
• diff
• differentiate
• drop
• drop_attrs
• drop_dims
• drop_duplicates
• drop_encoding
• drop_indexes
• drop_isel
• drop_sel
• drop_vars
• dropna
• dtypes
• dump_to_store
• eval
• expand_dims
• ffill
• fillna
• filter_by_attrs
• from_dataframe
• get_index
• groupby
• groupby_bins
• head
• idxmax
• idxmin
• imag
• info
• integrate
• interp
• interp_calendar
• interp_like
• interpolate_na
• isin
• isnull
• load_store
• loc
• map
• map_blocks
• merge
• notnull
• pad
• plot
• polyfit
• quantile
• query
• rank
• real
• reindex
• reindex_like
• rename
• rename_dims
• rename_vars
• reorder_levels
• resample
• reset_coords
• reset_encoding
• reset_index
• roll
• rolling
• rolling_exp
• set_coords
• set_index
• set_xindex
• shift
• sortby
• squeeze
• stack
• swap_dims
• tail
• thin
• to_array
• to_dask_dataframe
• to_dataarray
• to_dataframe
• to_pandas
• to_stacked_array
• transpose
• unify_chunks
• unstack
• weighted
• where

I think these are not available because they where defined directly on Dataset, while DataTree uses mixins to define reductions and binary ops...

import xarray as xr

dt_methods = set(m for m in dir(xr.DataTree) if not m.startswith("_"))
ds_methods = set(m for m in dir(xr.Dataset) if not m.startswith("_"))
for m in sorted(ds_methods - dt_methods):
    print(f"- [ ] {m}")

Did I miss that?

I don't think so, we might have forgotten to make an issue for this.

Is there a reason they are missing?

Available time? 🤷

while DataTree uses mixins to define reductions and binary ops...

So does Dataset! Repurposing those mixins for DataTree made it easy to add e.g. all the aggregation methods in one PR.

Here is a more complete list (code-generated):

So I don't think these are all equally important. We deliberately prioritized methods to manipulate and iterate over tree contents above implementing data analysis methods. A large number of the methods on that list could be implemented in one line using map_over_datasets, which to me means it's less urgent to add them as we are only saving users one line of code.

The things that are important to add that are still missing are to do with manipulating and viewing tree contents, such as set/assign_coords, drop*, map. Some of those we do have issues/open PRs for.

Also as an aside I actually feel like some of those methods on the list shouldn't even exist on Dataset (such as curvefit), as they aren't really fundamental to our data structures, and should live in a separate package.