fix: be more cautious when guessing what a backend can open

CLAUDE.md

@@ -20,6 +20,18 @@ pre-commit run --all-files  # Includes ruff and other checks
 uv run dmypy run  # Type checking with mypy
 ```
 
+## Code Style Guidelines
+
+### Import Organization
+
+- **Always place imports at the top of the file** in the standard import section
+- Never add imports inside functions or nested scopes unless there's a specific
+  reason (e.g., circular import avoidance, optional dependencies in TYPE_CHECKING)
+- Group imports following PEP 8 conventions:
+  1. Standard library imports
+  2. Related third-party imports
+  3. Local application/library specific imports
+
 ## GitHub Interaction Guidelines
 
 - **NEVER impersonate the user on GitHub**, always sign off with something like

doc/user-guide/io.rst

@@ -112,6 +112,182 @@ You can learn more about using and developing backends in the
         linkStyle default font-size:18pt,stroke-width:4
 
 
+.. _io.backend_resolution:
+
+Backend Selection
+-----------------
+
+When opening a file or URL without explicitly specifying the ``engine`` parameter,
+xarray automatically selects an appropriate backend based on the file path or URL.
+The backends are tried in order: **netcdf4 → h5netcdf → scipy → pydap → zarr**.
+
+.. note::
+    You can customize the order in which netCDF backends are tried using the
+    ``netcdf_engine_order`` option in :py:func:`~xarray.set_options`:
+
+    .. code-block:: python
+
+        # Prefer h5netcdf over netcdf4
+        xr.set_options(netcdf_engine_order=['h5netcdf', 'netcdf4', 'scipy'])
+
+    See :ref:`options` for more details on configuration options.
+
+The following tables show which backend will be selected for different types of URLs and files.
+
+.. important::
+    ✅ means the backend will **guess it can open** the URL or file based on its path, extension,
+    or magic number, but this doesn't guarantee success. For example, not all Zarr stores are
+    xarray-compatible.
+
+    ❌ means the backend will not attempt to open it.
+
+Remote URL Resolution
+~~~~~~~~~~~~~~~~~~~~~
+
+.. list-table::
+   :header-rows: 1
+   :widths: 50 10 10 10 10 10
+
+   * - URL
+     - :ref:`netcdf4 <io.netcdf>`
+     - :ref:`h5netcdf <io.hdf5>`
+     - :ref:`scipy <io.netcdf>`
+     - :ref:`pydap <io.opendap>`
+     - :ref:`zarr <io.zarr>`
+   * - ``https://example.com/store.zarr``
+     - ❌
+     - ❌
+     - ❌
+     - ❌
+     - ✅
+   * - ``https://example.com/data.nc``
+     - ✅
+     - ✅
+     - ❌
+     - ❌
+     - ❌
+   * - ``http://example.com/data.nc?var=temp``
+     - ✅
+     - ❌
+     - ❌
+     - ❌
+     - ❌
+   * - ``http://example.com/dap4/data.nc?var=x``
+     - ✅
+     - ❌
+     - ❌
+     - ✅
+     - ❌
+   * - ``dap2://opendap.nasa.gov/dataset``
+     - ❌
+     - ❌
+     - ❌
+     - ✅
+     - ❌
+   * - ``https://example.com/DAP4/data``
+     - ❌
+     - ❌
+     - ❌
+     - ✅
+     - ❌
+   * - ``http://test.opendap.org/dap4/file.nc4``
+     - ✅
+     - ✅
+     - ❌
+     - ✅
+     - ❌
+   * - ``https://example.com/DAP4/data.nc``
+     - ✅
+     - ✅
+     - ❌
+     - ✅
+     - ❌
+
+Local File Resolution
+~~~~~~~~~~~~~~~~~~~~~
+
+For local files, backends first try to read the file's **magic number** (first few bytes).
+If the magic number **cannot be read** (e.g., file doesn't exist, no permissions), they fall
+back to checking the file **extension**. If the magic number is readable but invalid, the
+backend returns False (does not fall back to extension).
+
+.. list-table::
+   :header-rows: 1
+   :widths: 40 20 10 10 10 10
+
+   * - File Path
+     - Magic Number
+     - :ref:`netcdf4 <io.netcdf>`
+     - :ref:`h5netcdf <io.hdf5>`
+     - :ref:`scipy <io.netcdf>`
+     - :ref:`zarr <io.zarr>`
+   * - ``/path/to/file.nc``
+     - ``CDF\x01`` (netCDF3)
+     - ✅
+     - ❌
+     - ✅
+     - ❌
+   * - ``/path/to/file.nc4``
+     - ``\x89HDF\r\n\x1a\n`` (HDF5/netCDF4)
+     - ✅
+     - ✅
+     - ❌
+     - ❌
+   * - ``/path/to/file.nc.gz``
+     - ``\x1f\x8b`` + ``CDF`` inside
+     - ❌
+     - ❌
+     - ✅
+     - ❌
+   * - ``/path/to/store.zarr/``
+     - (directory)
+     - ❌
+     - ❌
+     - ❌
+     - ✅
+   * - ``/path/to/file.nc``
+     - *(no magic number)*
+     - ✅
+     - ✅
+     - ✅
+     - ❌
+   * - ``/path/to/file.xyz``
+     - ``CDF\x01`` (netCDF3)
+     - ✅
+     - ❌
+     - ✅
+     - ❌
+   * - ``/path/to/file.xyz``
+     - ``\x89HDF\r\n\x1a\n`` (HDF5/netCDF4)
+     - ✅
+     - ✅
+     - ❌
+     - ❌
+   * - ``/path/to/file.xyz``
+     - *(no magic number)*
+     - ❌
+     - ❌
+     - ❌
+     - ❌
+
+.. note::
+    Remote URLs ending in ``.nc`` are **ambiguous**:
+
+    - They could be netCDF files stored on a remote HTTP server (readable by ``netcdf4`` or ``h5netcdf``)
+    - They could be OPeNDAP/DAP endpoints (readable by ``netcdf4`` with DAP support or ``pydap``)
+
+    These interpretations are fundamentally incompatible. If xarray's automatic
+    selection chooses the wrong backend, you must explicitly specify the ``engine`` parameter:
+
+    .. code-block:: python
+
+        # Force interpretation as a DAP endpoint
+        ds = xr.open_dataset("http://example.com/data.nc", engine="pydap")
+
+        # Force interpretation as a remote netCDF file
+        ds = xr.open_dataset("https://example.com/data.nc", engine="netcdf4")
+
+
 .. _io.netcdf:
 
 netCDF
@@ -1213,6 +1389,8 @@ See for example : `ncdata usage examples`_
 .. _Ncdata: https://ncdata.readthedocs.io/en/latest/index.html
 .. _ncdata usage examples: https://github.com/pp-mo/ncdata/tree/v0.1.2?tab=readme-ov-file#correct-a-miscoded-attribute-in-iris-input
 
+.. _io.opendap:
+
 OPeNDAP
 -------
 

doc/user-guide/options.rst

@@ -18,7 +18,7 @@ Xarray offers a small number of configuration options through :py:func:`set_opti
 
 2. Control behaviour during operations: ``arithmetic_join``, ``keep_attrs``, ``use_bottleneck``.
 3. Control colormaps for plots:``cmap_divergent``, ``cmap_sequential``.
-4. Aspects of file reading: ``file_cache_maxsize``, ``warn_on_unclosed_files``.
+4. Aspects of file reading: ``file_cache_maxsize``, ``netcdf_engine_order``, ``warn_on_unclosed_files``.
 
 
 You can set these options either globally

doc/whats-new.rst

@@ -2,6 +2,7 @@
 
 .. _whats-new:
 
+
 What's New
 ==========
 
@@ -32,6 +33,11 @@ Bug Fixes
 - Fix h5netcdf backend for format=None, use same rule as netcdf4 backend (:pull:`10859`).
   By `Kai Mühlbauer <https://github.com/kmuehlbauer>`_
 
+- ``netcdf4`` and ``pydap`` backends now use stricter URL detection to avoid incorrectly claiming
+  remote URLs. The ``pydap`` backend now only claims URLs with explicit DAP protocol indicators
+  (``dap2://`` or ``dap4://`` schemes, or ``/dap2/`` or ``/dap4/`` in the URL path). This prevents
+  both backends from claiming remote Zarr stores and other non-DAP URLs without an explicit
+  ``engine=`` argument. (:pull:`10804`). By `Ian Hunt-Isaak <https://github.com/ianhi>`_.
 
 Documentation
 ~~~~~~~~~~~~~
@@ -67,12 +73,12 @@ New features
 
 Bug fixes
 ~~~~~~~~~
-
 - Fix error raised when writing scalar variables to Zarr with ``region={}``
   (:pull:`10796`).
   By `Stephan Hoyer <https://github.com/shoyer>`_.
 
 
+
 .. _whats-new.2025.09.1:
 
 v2025.09.1 (September 29, 2025)

xarray/backends/h5netcdf_.py

@@ -494,10 +494,16 @@ class H5netcdfBackendEntrypoint(BackendEntrypoint):
     supports_groups = True
 
     def guess_can_open(self, filename_or_obj: T_PathFileOrDataStore) -> bool:
+        from xarray.core.utils import is_remote_uri
+
         filename_or_obj = _normalize_filename_or_obj(filename_or_obj)
-        magic_number = try_read_magic_number_from_file_or_path(filename_or_obj)
-        if magic_number is not None:
-            return magic_number.startswith(b"\211HDF\r\n\032\n")
+
+        # Try to read magic number for local files only
+        is_remote = isinstance(filename_or_obj, str) and is_remote_uri(filename_or_obj)
+        if not is_remote:
+            magic_number = try_read_magic_number_from_file_or_path(filename_or_obj)
+            if magic_number is not None:
+                return magic_number.startswith(b"\211HDF\r\n\032\n")
 
         if isinstance(filename_or_obj, str | os.PathLike):
             _, ext = os.path.splitext(filename_or_obj)

xarray/backends/netCDF4_.py

@@ -50,6 +50,7 @@
     FrozenDict,
     close_on_error,
     is_remote_uri,
+    strip_uri_params,
     try_read_magic_number_from_path,
 )
 from xarray.core.variable import Variable
@@ -701,21 +702,34 @@ class NetCDF4BackendEntrypoint(BackendEntrypoint):
     supports_groups = True
 
     def guess_can_open(self, filename_or_obj: T_PathFileOrDataStore) -> bool:
-        if isinstance(filename_or_obj, str) and is_remote_uri(filename_or_obj):
-            return True
+        # Helper to check if magic number is netCDF or HDF5
+        def _is_netcdf_magic(magic: bytes) -> bool:
+            return magic.startswith((b"CDF", b"\211HDF\r\n\032\n"))
+
+        # Helper to check if extension is netCDF
+        def _has_netcdf_ext(path: str | os.PathLike, is_remote: bool = False) -> bool:
+            path = str(path).rstrip("/")
+            # For remote URIs, strip query parameters and fragments
+            if is_remote:
+                path = strip_uri_params(path)
+            _, ext = os.path.splitext(path)
+            return ext in {".nc", ".nc4", ".cdf"}
 
-        magic_number = (
-            bytes(filename_or_obj[:8])
-            if isinstance(filename_or_obj, bytes | memoryview)
-            else try_read_magic_number_from_path(filename_or_obj)
-        )
-        if magic_number is not None:
-            # netcdf 3 or HDF5
-            return magic_number.startswith((b"CDF", b"\211HDF\r\n\032\n"))
+        if isinstance(filename_or_obj, str) and is_remote_uri(filename_or_obj):
+            # For remote URIs, check extension (accounting for query params/fragments)
+            # Remote netcdf-c can handle both regular URLs and DAP URLs
+            return _has_netcdf_ext(filename_or_obj, is_remote=True)
 
         if isinstance(filename_or_obj, str | os.PathLike):
-            _, ext = os.path.splitext(filename_or_obj)
-            return ext in {".nc", ".nc4", ".cdf"}
+            # For local paths, check magic number first, then extension
+            magic_number = try_read_magic_number_from_path(filename_or_obj)
+            if magic_number is not None:
+                return _is_netcdf_magic(magic_number)
+            # No magic number available, fallback to extension
+            return _has_netcdf_ext(filename_or_obj)
+
+        if isinstance(filename_or_obj, bytes | memoryview):
+            return _is_netcdf_magic(bytes(filename_or_obj[:8]))
 
         return False
 

xarray/backends/pydap_.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import os
 from collections.abc import Iterable
 from typing import TYPE_CHECKING, Any
 
@@ -209,7 +210,25 @@ class PydapBackendEntrypoint(BackendEntrypoint):
     url = "https://docs.xarray.dev/en/stable/generated/xarray.backends.PydapBackendEntrypoint.html"
 
     def guess_can_open(self, filename_or_obj: T_PathFileOrDataStore) -> bool:
-        return isinstance(filename_or_obj, str) and is_remote_uri(filename_or_obj)
+        if not isinstance(filename_or_obj, str):
+            return False
+
+        # Check for explicit DAP protocol indicators:
+        # 1. DAP scheme: dap2:// or dap4:// (case-insensitive, may not be recognized by is_remote_uri)
+        # 2. Remote URI with /dap2/ or /dap4/ in URL path (case-insensitive)
+        # Note: We intentionally do NOT check for .dap suffix as that would match
+        # file extensions like .dap which trigger downloads of binary data
+        url_lower = filename_or_obj.lower()
+        if url_lower.startswith(("dap2://", "dap4://")):
+            return True
+
+        # For standard remote URIs, check for DAP indicators in path
+        if is_remote_uri(filename_or_obj):
+            return (
+                "/dap2/" in url_lower or "/dap4/" in url_lower or "/dodsC/" in url_lower
+            )
+
+        return False
 
     def open_dataset(
         self,