Fix: De/Serialize SpanGroups including the SpanGroup keys (#10707)

* fix: De/Serialize `SpanGroups` including the SpanGroup keys

This prevents the loss of `SpanGroup`s that have the same .name as other `SpanGroup`s within the same `SpanGroups` object (upon de/serialization of the `SpanGroups`).

Fixes #10685

* Maintain backwards compatibility for serialized `SpanGroups`

(serialized as: a list of `SpanGroup`s, or b'')

* Add tests for `SpanGroups` deserialization backwards-compatibility

* Move a `SpanGroups` de/serialization test (test_issue10685)
  to tests/serialize/test_serialize_spangroups.py

* Output a warning if deserializing a `SpanGroups` with duplicate .name-d `SpanGroup`s

* Minor refactor

* `SpanGroups.from_bytes` handles only `list` and `dict` types with
`dict` as the expected default
* For lists, keep first rather than last value encountered
* Update error message
* Rename and update tests

* Update to preserve list serialization of SpanGroups

To avoid breaking compatibility of serialized `Doc` and `DocBin` with
earlier versions of spacy v3, revert back to a list-only serialization,
but update the names just for serialization so that the SpanGroups keys
override the SpanGroup names.

* Preserve object identity and current key overwrite

* Preserve SpanGroup object identity
* Preserve last rather than first span group from SpanGroup list
  format without SpanGroups keys

* Update inline comments

* Fix types

* Add type info for SpanGroup.copy

* Deserialize `SpanGroup`s as copies

when a single SpanGroup is the value for more than 1 `SpanGroups` key.
This is because we serialize `SpanGroups` as dicts (to maintain backward-
and forward-compatibility) and we can't assume `SpanGroup`s with the same
bytes/serialization were the same (identical) object, pre-serialization.

* Update spacy/tokens/_dict_proxies.py

* Add more SpanGroups serialization tests

Test that serialized SpanGroups maintain their Span order

* small clarification on older spaCy version

* Update spacy/tests/serialize/test_serialize_span_groups.py

Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>

spacy/errors.py

@@ -204,6 +204,11 @@ class Warnings(metaclass=ErrorsWithCodes):
             "for the corpora used to train the language. Please check "
             "`nlp.meta[\"sources\"]` for any relevant links.")
     W119 = ("Overriding pipe name in `config` is not supported. Ignoring override '{name_in_config}'.")
+    W120 = ("Unable to load all spans in Doc.spans: more than one span group "
+            "with the name '{group_name}' was found in the saved spans data. "
+            "Only the last span group will be loaded under "
+            "Doc.spans['{group_name}']. Skipping span group with values: "
+            "{group_values}")
 
 
 class Errors(metaclass=ErrorsWithCodes):

spacy/tests/serialize/test_serialize_span_groups.py

@@ -0,0 +1,161 @@
+import pytest
+
+from spacy.tokens import Span, SpanGroup
+from spacy.tokens._dict_proxies import SpanGroups
+
+
+@pytest.mark.issue(10685)
+def test_issue10685(en_tokenizer):
+    """Test `SpanGroups` de/serialization"""
+    # Start with a Doc with no SpanGroups
+    doc = en_tokenizer("Will it blend?")
+
+    # Test empty `SpanGroups` de/serialization:
+    assert len(doc.spans) == 0
+    doc.spans.from_bytes(doc.spans.to_bytes())
+    assert len(doc.spans) == 0
+
+    # Test non-empty `SpanGroups` de/serialization:
+    doc.spans["test"] = SpanGroup(doc, name="test", spans=[doc[0:1]])
+    doc.spans["test2"] = SpanGroup(doc, name="test", spans=[doc[1:2]])
+
+    def assert_spangroups():
+        assert len(doc.spans) == 2
+        assert doc.spans["test"].name == "test"
+        assert doc.spans["test2"].name == "test"
+        assert list(doc.spans["test"]) == [doc[0:1]]
+        assert list(doc.spans["test2"]) == [doc[1:2]]
+
+    # Sanity check the currently-expected behavior
+    assert_spangroups()
+
+    # Now test serialization/deserialization:
+    doc.spans.from_bytes(doc.spans.to_bytes())
+
+    assert_spangroups()
+
+
+def test_span_groups_serialization_mismatches(en_tokenizer):
+    """Test the serialization of multiple mismatching `SpanGroups` keys and `SpanGroup.name`s"""
+    doc = en_tokenizer("How now, brown cow?")
+    # Some variety:
+    # 1 SpanGroup where its name matches its key
+    # 2 SpanGroups that have the same name--which is not a key
+    # 2 SpanGroups that have the same name--which is a key
+    # 1 SpanGroup that is a value for 2 different keys (where its name is a key)
+    # 1 SpanGroup that is a value for 2 different keys (where its name is not a key)
+    groups = doc.spans
+    groups["key1"] = SpanGroup(doc, name="key1", spans=[doc[0:1], doc[1:2]])
+    groups["key2"] = SpanGroup(doc, name="too", spans=[doc[3:4], doc[4:5]])
+    groups["key3"] = SpanGroup(doc, name="too", spans=[doc[1:2], doc[0:1]])
+    groups["key4"] = SpanGroup(doc, name="key4", spans=[doc[0:1]])
+    groups["key5"] = SpanGroup(doc, name="key4", spans=[doc[0:1]])
+    sg6 = SpanGroup(doc, name="key6", spans=[doc[0:1]])
+    groups["key6"] = sg6
+    groups["key7"] = sg6
+    sg8 = SpanGroup(doc, name="also", spans=[doc[1:2]])
+    groups["key8"] = sg8
+    groups["key9"] = sg8
+
+    regroups = SpanGroups(doc).from_bytes(groups.to_bytes())
+
+    # Assert regroups == groups
+    assert regroups.keys() == groups.keys()
+    for key, regroup in regroups.items():
+        # Assert regroup == groups[key]
+        assert regroup.name == groups[key].name
+        assert list(regroup) == list(groups[key])
+
+
+@pytest.mark.parametrize(
+    "spans_bytes,doc_text,expected_spangroups,expected_warning",
+    # The bytestrings below were generated from an earlier version of spaCy
+    # that serialized `SpanGroups` as a list of SpanGroup bytes (via SpanGroups.to_bytes).
+    # Comments preceding the bytestrings indicate from what Doc they were created.
+    [
+        # Empty SpanGroups:
+        (b"\x90", "", {}, False),
+        # doc = nlp("Will it blend?")
+        # doc.spans['test'] = SpanGroup(doc, name='test', spans=[doc[0:1]])
+        (
+            b"\x91\xc4C\x83\xa4name\xa4test\xa5attrs\x80\xa5spans\x91\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x04",
+            "Will it blend?",
+            {"test": {"name": "test", "spans": [(0, 1)]}},
+            False,
+        ),
+        # doc = nlp("Will it blend?")
+        # doc.spans['test']  = SpanGroup(doc, name='test', spans=[doc[0:1]])
+        # doc.spans['test2'] = SpanGroup(doc, name='test', spans=[doc[1:2]])
+        (
+            b"\x92\xc4C\x83\xa4name\xa4test\xa5attrs\x80\xa5spans\x91\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x04\xc4C\x83\xa4name\xa4test\xa5attrs\x80\xa5spans\x91\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x02\x00\x00\x00\x05\x00\x00\x00\x07",
+            "Will it blend?",
+            # We expect only 1 SpanGroup to be in doc.spans in this example
+            # because there are 2 `SpanGroup`s that have the same .name. See #10685.
+            {"test": {"name": "test", "spans": [(1, 2)]}},
+            True,
+        ),
+        # doc = nlp('How now, brown cow?')
+        # doc.spans['key1'] = SpanGroup(doc, name='key1', spans=[doc[0:1], doc[1:2]])
+        # doc.spans['key2'] = SpanGroup(doc, name='too', spans=[doc[3:4], doc[4:5]])
+        # doc.spans['key3'] = SpanGroup(doc, name='too', spans=[doc[1:2], doc[0:1]])
+        # doc.spans['key4'] = SpanGroup(doc, name='key4', spans=[doc[0:1]])
+        # doc.spans['key5'] = SpanGroup(doc, name='key4', spans=[doc[0:1]])
+        (
+            b"\x95\xc4m\x83\xa4name\xa4key1\xa5attrs\x80\xa5spans\x92\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x03\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x02\x00\x00\x00\x04\x00\x00\x00\x07\xc4l\x83\xa4name\xa3too\xa5attrs\x80\xa5spans\x92\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03\x00\x00\x00\x04\x00\x00\x00\t\x00\x00\x00\x0e\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x04\x00\x00\x00\x05\x00\x00\x00\x0f\x00\x00\x00\x12\xc4l\x83\xa4name\xa3too\xa5attrs\x80\xa5spans\x92\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x02\x00\x00\x00\x04\x00\x00\x00\x07\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x03\xc4C\x83\xa4name\xa4key4\xa5attrs\x80\xa5spans\x91\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x03\xc4C\x83\xa4name\xa4key4\xa5attrs\x80\xa5spans\x91\xc4(\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x03",
+            "How now, brown cow?",
+            {
+                "key1": {"name": "key1", "spans": [(0, 1), (1, 2)]},
+                "too": {"name": "too", "spans": [(1, 2), (0, 1)]},
+                "key4": {"name": "key4", "spans": [(0, 1)]},
+            },
+            True,
+        ),
+    ],
+)
+def test_deserialize_span_groups_compat(
+    en_tokenizer, spans_bytes, doc_text, expected_spangroups, expected_warning
+):
+    """Test backwards-compatibility of `SpanGroups` deserialization.
+    This uses serializations (bytes) from a prior version of spaCy (before 3.3.1).
+
+    spans_bytes (bytes): Serialized `SpanGroups` object.
+    doc_text (str): Doc text.
+    expected_spangroups (dict):
+        Dict mapping every expected (after deserialization) `SpanGroups` key
+        to a SpanGroup's "args", where a SpanGroup's args are given as a dict:
+          {"name": span_group.name,
+           "spans": [(span0.start, span0.end), ...]}
+    expected_warning (bool): Whether a warning is to be expected from .from_bytes()
+        --i.e. if more than 1 SpanGroup has the same .name within the `SpanGroups`.
+    """
+    doc = en_tokenizer(doc_text)
+
+    if expected_warning:
+        with pytest.warns(UserWarning):
+            doc.spans.from_bytes(spans_bytes)
+    else:
+        # TODO: explicitly check for lack of a warning
+        doc.spans.from_bytes(spans_bytes)
+
+    assert doc.spans.keys() == expected_spangroups.keys()
+    for name, spangroup_args in expected_spangroups.items():
+        assert doc.spans[name].name == spangroup_args["name"]
+        spans = [Span(doc, start, end) for start, end in spangroup_args["spans"]]
+        assert list(doc.spans[name]) == spans
+
+
+def test_span_groups_serialization(en_tokenizer):
+    doc = en_tokenizer("0 1 2 3 4 5 6")
+    span_groups = SpanGroups(doc)
+    spans = [doc[0:2], doc[1:3]]
+    sg1 = SpanGroup(doc, spans=spans)
+    span_groups["key1"] = sg1
+    span_groups["key2"] = sg1
+    span_groups["key3"] = []
+    reloaded_span_groups = SpanGroups(doc).from_bytes(span_groups.to_bytes())
+    assert span_groups.keys() == reloaded_span_groups.keys()
+    for key, value in span_groups.items():
+        assert all(
+            span == reloaded_span
+            for span, reloaded_span in zip(span_groups[key], reloaded_span_groups[key])
+        )

spacy/tokens/_dict_proxies.py

@@ -1,10 +1,11 @@
-from typing import Iterable, Tuple, Union, Optional, TYPE_CHECKING
+from typing import Dict, Iterable, List, Tuple, Union, Optional, TYPE_CHECKING
+import warnings
 import weakref
 from collections import UserDict
 import srsly
 
 from .span_group import SpanGroup
-from ..errors import Errors
+from ..errors import Errors, Warnings
 
 
 if TYPE_CHECKING:
@@ -16,7 +17,7 @@ if TYPE_CHECKING:
 # Why inherit from UserDict instead of dict here?
 # Well, the 'dict' class doesn't necessarily delegate everything nicely,
 # for performance reasons. The UserDict is slower but better behaved.
-# See https://treyhunner.com/2019/04/why-you-shouldnt-inherit-from-list-and-dict-in-python/0ww
+# See https://treyhunner.com/2019/04/why-you-shouldnt-inherit-from-list-and-dict-in-python/
 class SpanGroups(UserDict):
     """A dict-like proxy held by the Doc, to control access to span groups."""
 
@@ -53,20 +54,52 @@ class SpanGroups(UserDict):
         return super().setdefault(key, default=default)
 
     def to_bytes(self) -> bytes:
-        # We don't need to serialize this as a dict, because the groups
+        # We serialize this as a dict in order to track the key(s) a SpanGroup
-        # know their names.
+        # is a value of (in a backward- and forward-compatible way), since
+        # a SpanGroup can have a key that doesn't match its `.name` (See #10685)
         if len(self) == 0:
             return self._EMPTY_BYTES
-        msg = [value.to_bytes() for value in self.values()]
+        msg: Dict[bytes, List[str]] = {}
+        for key, value in self.items():
+            msg.setdefault(value.to_bytes(), []).append(key)
         return srsly.msgpack_dumps(msg)
 
     def from_bytes(self, bytes_data: bytes) -> "SpanGroups":
-        msg = [] if bytes_data == self._EMPTY_BYTES else srsly.msgpack_loads(bytes_data)
+        # backwards-compatibility: bytes_data may be one of:
+        # b'', a serialized empty list, a serialized list of SpanGroup bytes
+        # or a serialized dict of SpanGroup bytes -> keys
+        msg = (
+            []
+            if not bytes_data or bytes_data == self._EMPTY_BYTES
+            else srsly.msgpack_loads(bytes_data)
+        )
         self.clear()
         doc = self._ensure_doc()
+        if isinstance(msg, list):
+            # This is either the 1st version of `SpanGroups` serialization
+            # or there were no SpanGroups serialized
             for value_bytes in msg:
                 group = SpanGroup(doc).from_bytes(value_bytes)
+                if group.name in self:
+                    # Display a warning if `msg` contains `SpanGroup`s
+                    # that have the same .name (attribute).
+                    # Because, for `SpanGroups` serialized as lists,
+                    # only 1 SpanGroup per .name is loaded. (See #10685)
+                    warnings.warn(
+                        Warnings.W120.format(
+                            group_name=group.name, group_values=self[group.name]
+                        )
+                    )
                 self[group.name] = group
+        else:
+            for value_bytes, keys in msg.items():
+                group = SpanGroup(doc).from_bytes(value_bytes)
+                # Deserialize `SpanGroup`s as copies because it's possible for two
+                # different `SpanGroup`s (pre-serialization) to have the same bytes
+                # (since they can have the same `.name`).
+                self[keys[0]] = group
+                for key in keys[1:]:
+                    self[key] = group.copy()
         return self
 
     def _ensure_doc(self) -> "Doc":

spacy/tokens/span_group.pyi

@@ -24,3 +24,4 @@ class SpanGroup:
     def __getitem__(self, i: int) -> Span: ...
     def to_bytes(self) -> bytes: ...
     def from_bytes(self, bytes_data: bytes) -> SpanGroup: ...
+    def copy(self) -> SpanGroup: ...