Use the less restrictive datatype when a column is mixed

CHANGELOG.md

@@ -15,6 +15,7 @@ Please see [MIGRATING.md](./MIGRATING.md) for information on breaking changes.
 
 - Endpoints which return system fields before the record list
 - Example python script
+- Columns with mixed and incompatible datatypes
 
 ### Changed
 

src/ldlite/_jsonx.py

@@ -161,7 +161,7 @@ def _compile_attrs(  # noqa: C901, PLR0912, PLR0913
     if depth > max_depth:
         return
     table = _table_name(parents)
-    qkey = {}
+    qkey: dict[str, Attr] = {}
     for k, a in quasikey.items():
         qkey[k] = Attr(a.name, a.datatype, order=1)
     arrays: list[tuple[str, list[JsonValue], str]] = []
@@ -170,6 +170,11 @@ def _compile_attrs(  # noqa: C901, PLR0912, PLR0913
         if k is None or v is None:
             continue
         attr = prefix + k
+        prev_attr = (
+            newattrs[table][attr]
+            if table in newattrs and attr in newattrs[table]
+            else None
+        )
         if isinstance(v, dict):
             if depth == max_depth:
                 newattrs[table][attr] = Attr(
@@ -181,6 +186,11 @@ def _compile_attrs(  # noqa: C901, PLR0912, PLR0913
                 objects.append((attr, v, k))
         elif isinstance(v, list):
             arrays.append((attr, v, k))
+        elif prev_attr and prev_attr.datatype == "varchar":
+            # If a column has already been through this check
+            # and determined to be a varchar it means that
+            # we can't type it as anything more specific later
+            qkey[attr] = prev_attr
         elif isinstance(v, bool):
             a = Attr(decode_camel_case(attr), "boolean", order=3)
             qkey[attr] = a

tests/test_cases/query_cases.py

@@ -604,6 +604,7 @@ def case_null_records(self) -> QueryCase:
             ],
         )
 
+    # https://github.com/library-data-platform/ldlite/issues/52
     def case_erm_keys(self) -> QueryCase:
         return QueryCase(
             json_depth=3,
@@ -638,3 +639,41 @@ def case_erm_keys(self) -> QueryCase:
                 ("prefix__t", "id"),
             ],
         )
+
+    # https://github.com/library-data-platform/ldlite/issues/54
+    def case_mixed_uuid(self) -> QueryCase:
+        return QueryCase(
+            json_depth=3,
+            values={
+                "prefix": [
+                    {
+                        "purchaseOrders": [
+                            {
+                                "id": "aaaa",
+                                "value": "value",
+                            },
+                            {
+                                "id": "b096504a-3d54-4664-9bf5-1b872466fd66",
+                                "value": 15,
+                            },
+                        ],
+                    },
+                ],
+            },
+            expected_tables=["prefix", "prefix__t", "prefix__tcatalog"],
+            expected_values={
+                "prefix__t": (
+                    ["id", "value"],
+                    [
+                        ("aaaa", "value"),
+                        ("b096504a-3d54-4664-9bf5-1b872466fd66", "15"),
+                    ],
+                ),
+                "prefix__tcatalog": (["table_name"], [("prefix__t",)]),
+            },
+            expected_indexes=[
+                ("prefix", "__id"),
+                ("prefix__t", "__id"),
+                ("prefix__t", "id"),
+            ],
+        )