From 714b167200ac1a8a95f21bfddc4515a10a3634d0 Mon Sep 17 00:00:00 2001
From: ljgray <liam.j.gray@usask.ca>
Date: Tue, 25 Apr 2023 11:45:23 -0700
Subject: [PATCH] fix(memh5): block axis downselection in shared datasets

---
 caput/memh5.py | 23 +++++++++++++++++++----
 1 file changed, 19 insertions(+), 4 deletions(-)

diff --git a/caput/memh5.py b/caput/memh5.py
index 39e73040..83b3a857 100644
--- a/caput/memh5.py
+++ b/caput/memh5.py
@@ -2543,6 +2543,9 @@ def deep_group_copy(
     >>> list(g2["foo"]["bar"])
     [0, 1]
 
+    It is important to note that axis downselections cannot be applied to
+    shared datasets.
+
     Parameters
     ----------
     g1 : h5py.Group or zarr.Group
@@ -2590,10 +2593,9 @@ def deep_group_copy(
 
     to_file = isinstance(g2, file_format.module.Group)
 
-    # Prepare a dataset for writing out, applying selections and transforming any
-    # datatypes
-    # Returns: dict(dtype, shape, data_to_write)
-    def _prepare_dataset(dset):
+    # Get the selection associated with this dataset
+    # Returns: slice
+    def _get_selection(dset):
         # Look for a selection for this dataset (also try without the leading "/")
         try:
             selection = selections.get(
@@ -2602,6 +2604,14 @@ def _prepare_dataset(dset):
         except AttributeError:
             selection = slice(None)
 
+        return selection
+
+    # Prepare a dataset for writing out, applying selections and transforming any
+    # datatypes
+    # Returns: dict(dtype, shape, data_to_write)
+    def _prepare_dataset(dset):
+        selection = _get_selection(dset)
+
         # Check if this is a distributed dataset and figure out if we can make this work
         # out
         if to_file and isinstance(dset, MemDatasetDistributed):
@@ -2708,6 +2718,11 @@ def _prepare_compression_args(dset):
             stack += [entry[k] for k in sorted(entry, reverse=True)]
 
         elif key in shared:
+            # Make sure that we aren't trying to apply a selection to this dataset
+            if _get_selection(entry) != slice(None):
+                raise ValueError(
+                    f"Cannot apply a selection to a shared dataset ({entry.name})"
+                )
             # Just point to the existing dataset
             parent_name, name = posixpath.split(posixpath.join(g2.name, key))
             parent_name = format_abs_path(parent_name)