fix(generic): do not flatten a nested object that is null in any row (losslessness)

_analyze_flattenable skipped None during shape analysis, so a field that was a
dict in some rows and None at an intermediate nesting level in others was
flattened; on decode the None row's leaves resolved as absent and unflattened
to a missing key, dropping the value ({"meta":{"owner":None}} decoded to {}).
Now bails to the attachment path when a nested field (parent_path != "") is None;
a top-level None still flattens losslessly via the "-"/all-null rule.

Adds test_flatten_roundtrip: aligned nested-object arrays with null at depth —
the shape the scalar-only generator never produced, which is why this shipped
despite property-based round-trip fuzzing. Enforced cross-SDK by the shared
conformance fixtures flatten/017-019. (Prototype pollution does not affect
Python; dicts have no mutable prototype.)

Author: blackwell-systems

CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## v2.2.2 (2026-07-10)
+
+### Fixes
+
+- **Losslessness (nested null):** a nested object that is null at an intermediate level (e.g. `{"meta": {"owner": None}}`) is no longer flattened. Previously its leaves encoded as absent (`~`) and unflattened to a missing key, silently dropping the null. Such fields now fall back to the attachment mechanism; a top-level `None` still flattens losslessly (emits `-`, reconstructs via the all-null rule). Enforced by the shared conformance fixtures `flatten/017`–`019`. Prototype pollution does not affect Python (dicts have no mutable prototype).
+
+### Tests
+
+- `test_flatten_roundtrip`: aligned arrays whose shared fields are fixed-shape nested objects with a field or an intermediate nested level sometimes null/absent — the shape the prior scalar-only generator never produced, leaving the flatten/unflatten path unexercised. Verified to fail on the pre-fix encoder and pass on the fix.
+
 ## v2.2.1 (2026-06-23)
 
 ### Flatten Opt-Out

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "gcf-python"
-version = "2.2.1"
+version = "2.2.2"
 description = "The AI-native wire format for structured data. 50-92% fewer tokens than JSON. 100% comprehension on every frontier model. Zero dependencies."
 readme = "README.md"
 license = {text = "MIT"}

src/gcf/__init__.py

@@ -66,4 +66,4 @@
     "encode_with_session",
 ]
 
-__version__ = "1.0.0"
+__version__ = "2.2.2"

src/gcf/generic.py

@@ -171,7 +171,15 @@ def _analyze_flattenable(
     canonical_shape: dict[str, str] | None = None  # key -> "scalar" | "nested"
 
     for item in arr:
-        if field_name not in item or item[field_name] is None:
+        if field_name not in item:
+            continue
+        # A nested (non-top-level) null cannot be flattened losslessly: its leaves
+        # would encode as absent ("~") and unflatten back to a missing key, not None.
+        # Bail to the attachment path. A top-level None is fine (emits "-" and
+        # reconstructs via the all-null rule), so just skip the row from shape analysis.
+        if item[field_name] is None:
+            if parent_path != "":
+                return None
             continue
         v = item[field_name]
         if not isinstance(v, dict):

tests/test_roundtrip_v2.py

@@ -214,6 +214,80 @@ def _structural_equal(a, b):
     return a == b
 
 
+def _gen_flat_shape(r, depth, max_depth):
+    """A fixed nested schema: the string "scalar" or a dict of named sub-shapes."""
+    if depth >= max_depth or r.random() < 0.45:
+        return "scalar"
+    shape = {}
+    for _ in range(1 + r.randint(0, 2)):
+        shape[_gen_bare_key(r)] = _gen_flat_shape(r, depth + 1, max_depth)
+    return shape if shape else "scalar"
+
+
+def _materialize_flat_shape(r, shape):
+    if shape == "scalar":
+        return _gen_scalar(r)
+    obj = {}
+    for k, s in shape.items():
+        # A nested sub-object is sometimes None (intermediate null — the case the
+        # pre-fix encoder dropped) instead of a full object.
+        obj[k] = None if s != "scalar" and r.random() < 0.3 else _materialize_flat_shape(r, s)
+    return obj
+
+
+def _gen_flattenable_array(r):
+    schema = {"id": "scalar"}
+    order = ["id"]
+    has_nested = False
+    for _ in range(1 + r.randint(0, 2)):
+        k = _gen_bare_key(r)
+        if k in schema:
+            continue
+        s = _gen_flat_shape(r, 1, 3)
+        schema[k] = s
+        order.append(k)
+        if s != "scalar":
+            has_nested = True
+    if not has_nested:
+        k = _gen_bare_key(r)
+        schema[k] = {_gen_bare_key(r): {_gen_bare_key(r): "scalar"}}
+        order.append(k)
+    arr = []
+    for _ in range(2 + r.randint(0, 5)):
+        row = {}
+        for f in order:
+            x = r.random()
+            if x < 0.12:
+                continue  # field absent this row
+            elif x < 0.24:
+                row[f] = None  # field present-null (top-level null)
+            else:
+                row[f] = _materialize_flat_shape(r, schema[f])
+        arr.append(row)
+    return arr
+
+
+def test_flatten_roundtrip():
+    """Aligned arrays whose shared fields are fixed-shape nested objects with a
+    field or an intermediate nested level sometimes null/absent — the v3.2 flatten
+    path the scalar-only generator never produces, so flatten/unflatten and its
+    null-at-depth losslessness edge would otherwise be unexercised."""
+    r = _rng(7)
+    for i in range(ITERATIONS):
+        val = _gen_flattenable_array(r)
+        for no_flatten in (False, True):
+            gcf = encode_generic(val, GenericOptions(no_flatten=no_flatten))
+            decoded = decode_generic(gcf)
+            a = _json_norm(val)
+            b = _json_norm(decoded)
+            assert _structural_equal(a, b), (
+                f"iteration {i} no_flatten={no_flatten}: round-trip mismatch\n"
+                f"  input:   {json.dumps(val)}\n"
+                f"  decoded: {json.dumps(decoded)}\n"
+                f"  gcf:     {gcf!r}"
+            )
+
+
 def test_random_roundtrip():
     r = _rng(42)
     for i in range(ITERATIONS):