chore: fix experimental blob operations json string values (#1299)

Author: GarrettWu

bigframes/core/compile/scalar_op_compiler.py

@@ -1220,6 +1220,11 @@ def to_json_string_op_impl(json_obj: ibis_types.Value):
     return to_json_string(json_obj=json_obj)
 
 
+@scalar_op_compiler.register_unary_op(ops.JSONValue, pass_op=True)
+def json_value_op_impl(x: ibis_types.Value, op: ops.JSONValue):
+    return json_value(json_obj=x, json_path=op.json_path)
+
+
 # Blob Ops
 @scalar_op_compiler.register_unary_op(ops.obj_fetch_metadata_op)
 def obj_fetch_metadata_op_impl(obj_ref: ibis_types.Value):
@@ -1931,6 +1936,13 @@ def to_json_string(  # type: ignore[empty-body]
     """Convert JSON to STRING."""
 
 
+@ibis_udf.scalar.builtin(name="json_value")
+def json_value(  # type: ignore[empty-body]
+    json_obj: ibis_dtypes.JSON, json_path: ibis_dtypes.String
+) -> ibis_dtypes.String:
+    """Retrieve value of a JSON field as plain STRING."""
+
+
 @ibis_udf.scalar.builtin(name="ML.DISTANCE")
 def vector_distance(vector1, vector2, type: str) -> ibis_dtypes.Float64:  # type: ignore[empty-body]
     """Computes the distance between two vectors using specified type ("EUCLIDEAN", "MANHATTAN", or "COSINE")"""

bigframes/dataframe.py

@@ -743,16 +743,16 @@ def _repr_html_(self) -> str:
 
         df = self.copy()
         if bigframes.options.experiments.blob:
-            import bigframes.bigquery as bbq
-
             blob_cols = [
                 col
                 for col in df.columns
                 if df[col].dtype == bigframes.dtypes.OBJ_REF_DTYPE
             ]
             for col in blob_cols:
                 df[col] = df[col]._apply_unary_op(ops.ObjGetAccessUrl(mode="R"))
-                df[col] = bbq.json_extract(df[col], "$.access_urls.read_url")
+                df[col] = df[col]._apply_unary_op(
+                    ops.JSONValue(json_path="$.access_urls.read_url")
+                )
 
         # TODO(swast): pass max_columns and get the true column count back. Maybe
         # get 1 more column than we have requested so that pandas can add the
@@ -770,8 +770,7 @@ def _repr_html_(self) -> str:
             if bigframes.options.experiments.blob:
 
                 def url_to_image_html(url: str) -> str:
-                    # url is a json string, which already contains double-quotes ""
-                    return f"<img src={url}>"
+                    return f'<img src="{url}">'
 
                 formatters = {blob_col: url_to_image_html for blob_col in blob_cols}
 

bigframes/operations/__init__.py

@@ -86,6 +86,7 @@
     JSONExtractArray,
     JSONExtractStringArray,
     JSONSet,
+    JSONValue,
     ParseJSON,
     ToJSONString,
 )
@@ -298,6 +299,7 @@
     "JSONExtractArray",
     "JSONExtractStringArray",
     "JSONSet",
+    "JSONValue",
     "ParseJSON",
     "ToJSONString",
     # Bool ops

bigframes/operations/blob.py

@@ -47,7 +47,7 @@ def metadata(self) -> bigframes.series.Series:
         )
         import bigframes.bigquery as bbq
 
-        return bbq.json_extract(details_json, "$.gcs_metadata")
+        return bbq.json_extract(details_json, "$.gcs_metadata").rename("metadata")
 
     def content_type(self) -> bigframes.series.Series:
         """Retrive the content type of the Blob.
@@ -57,11 +57,11 @@ def content_type(self) -> bigframes.series.Series:
 
         Returns:
             BigFrames Series: json-string of the content type."""
-        import bigframes.bigquery as bbq
-
         metadata = self.metadata()
 
-        return bbq.json_extract(metadata, "$.content_type")
+        return metadata._apply_unary_op(
+            ops.JSONValue(json_path="$.content_type")
+        ).rename("content_type")
 
     def display(self, n: int = 3, *, content_type: str = ""):
         """Display the blob content in the IPython Notebook environment. Only works for image type now.
@@ -73,14 +73,12 @@ def display(self, n: int = 3, *, content_type: str = ""):
             n (int, default 3): number of sample blob objects to display.
             content_type (str, default ""): content type of the blob. If unset, use the blob metadata of the storage. Possible values are "image", "audio" and "video".
         """
-        import bigframes.bigquery as bbq
-
         # col name doesn't matter here. Rename to avoid column name conflicts
         df = bigframes.series.Series(self._block).rename("blob_col").head(n).to_frame()
 
         obj_ref_runtime = df["blob_col"]._apply_unary_op(ops.ObjGetAccessUrl(mode="R"))
-        df["read_url"] = bbq.json_extract(
-            obj_ref_runtime, json_path="$.access_urls.read_url"
+        df["read_url"] = obj_ref_runtime._apply_unary_op(
+            ops.JSONValue(json_path="$.access_urls.read_url")
         )
 
         if content_type:
@@ -104,11 +102,7 @@ def display_single_url(read_url: str, content_type: str):
                 ipy_display.display(response.content, raw=True)
 
         for _, row in df.iterrows():
-            # both are JSON-formated strings
-            read_url = str(row["read_url"]).strip('"')
-            content_type = str(row["content_type"]).strip('"')
-
-            display_single_url(read_url, content_type)
+            display_single_url(row["read_url"], row["content_type"])
 
     def image_blur(
         self,

bigframes/operations/json_ops.py

@@ -120,3 +120,18 @@ def output_type(self, *input_types):
 
         # After JSON type implementation, ONLY return JSON data.
         return left_type
+
+
+@dataclasses.dataclass(frozen=True)
+class JSONValue(base_ops.UnaryOp):
+    name: typing.ClassVar[str] = "json_value"
+    json_path: str
+
+    def output_type(self, *input_types):
+        input_type = input_types[0]
+        if not dtypes.is_json_like(input_type):
+            raise TypeError(
+                "Input type must be an valid JSON object or JSON-formatted string type."
+                + f" Received type: {input_type}"
+            )
+        return dtypes.STRING_DTYPE