Minimal implementation of a ora2 as an uncurl variant

The basic idea here is:

- Once we can describe the structure of a RIA store in an `uncurl` URL
  template, generic implementations can handle actual operations
- We need a dataset ID for this, and take it either from the traditional
  `archive-id` remote parameter, or the DataLad dataset ID in the repo
- With this information and the base URL from the `url` parameter, we
  build the `uncurl` URL template, and write it into the active config
  (as an override). From there `uncurl` picks it up, and we ensure this
  by running its `prepare()` in the `ora2` `prepare()` method last.

An included test confirm the basic principle by copying an annex key
to a local (`file://` RIA store) via a standard `git annex copy`
operation.

This setup is particularly attractive, because now specialized URL
handlers can also be selected via standard `uncurl` means for handler
selection (based on URL matching, which includes protocol switching).
If this is working as expected, we can have things like a
persistent-SSH-shell handler to only be active for particular hosts
(and even only temporarily).

This changeset includes several TODOs that outline further steps that
need to be approached in the future.

datalad_ria/ora_remote.py

@@ -1,3 +1,5 @@
+import uuid
+
 from datalad_next.annexremotes import (
     RemoteError,
     super_main,
@@ -43,6 +45,9 @@ def initremote(self):
             # Adopt git-annex's style of messaging
             raise RemoteError('ria+<scheme>://... URL expected for url=')
 
+        # TODO run _get_ria_dsid() to confirm the validity of the ID
+        # setup
+
         # here we could do all kinds of sanity and version checks.
         # however, in some sense `initremote` is not very special in
         # this regard. Most (or all) of these checks would also run
@@ -52,6 +57,55 @@ def initremote(self):
         # not seeing a relevant remote-specific git-config). Therefore
         # we are not doing any checks here (for now).
 
+    def prepare(self):
+        # UUID to use for this dataset in the store
+        dsid = self._get_ria_dsid()
+
+        # check for a remote-specific uncurl config
+        # self.get_remote_gitcfg() would also consider a remote-type
+        # general default, which is undesirable here
+        tmpl_var = f'remote.{self.remotename}.uncurl-url'
+        url_tmpl = self.repo.config.get(tmpl_var, None)
+        if url_tmpl is None:
+            # pull the recorded ria URL from git-annex
+            ria_url = self.annex.getconfig('url')
+            assert ria_url.startswith('ria+')
+            # TODO check the layout settings of the actual store
+            # to match this template
+            base_url = ria_url[4:]
+            url_tmpl = (
+                # we fill in base url and dsid directly here (not via
+                # uncurl templating), because it is simpler
+                f'{base_url}/{dsid[:3]}/{dsid[3:]}/annex/objects/'
+                # RIA v? uses the "mixed" dirhash
+                '{annex_dirhash}{annex_key}/{annex_key}'
+            )
+        # we set the URL template in the config for the base class
+        # routines to find
+        self.repo.config.set(tmpl_var, url_tmpl, scope='override')
+        # the rest is UNCURL "business as usual"
+        super().prepare()
+
+    #
+    # helpers
+    #
+    def _get_ria_dsid(self):
+        # check if the remote has a particular dataset ID configured
+        # via git-annex
+        dsid = self.annex.getconfig('archive-id')
+        # if not, fall back on the datalad dataset ID
+        if not dsid:
+            dsid = self.repo.config.get('datalad.dataset.id')
+        # under all circumstances this must be a valid UUID
+        try:
+            uuid.UUID(dsid)
+        except ValueError as e:
+            raise RemoteError(
+                'No valid dataset UUID identifier found,'
+                'specify via archive-id='
+            ) from e
+        return dsid
+
 
 def main():
     """CLI entry point installed as ``git-annex-remote-ora2``"""

datalad_ria/tests/test_ora.py

@@ -51,3 +51,20 @@ def test_ora_localops(ria_store_localaccess, populated_dataset):
 
     # smoke test that it can run
     repo.call_annex(ir_cmd + [f'url=ria+{store_path.as_uri()}'])
+
+    cp_cmd = [
+        'copy',
+        '-t', f'test-{ora_external_type}',
+    ]
+
+    repo.call_annex(cp_cmd + ['one.txt'])
+    # the annex key properties (and dirhash) are determined by the
+    # file content and the MD5E backend default.
+    # If neither of those changes, they must not change
+    key_fpath = store_path / \
+        ds.id[:3] / ds.id[3:] / 'annex' / 'objects' / \
+        'X9' / '6J' / \
+        'MD5E-s8--7e55db001d319a94b0b713529a756623.txt' / \
+        'MD5E-s8--7e55db001d319a94b0b713529a756623.txt'
+    assert key_fpath.exists()
+    assert key_fpath.read_text() == 'content1'