Use primary_id_attribute for updates, return typed empty DF, redact LLM logs by default, fix string concat

- prodev: use primary_id_attribute from tables.create() instead of guessing
  the ID column name (e.g., TABLE_TASKid)
- get(): return empty DataFrame with select columns (not columnless) for
  consistent downstream access on empty results
- LLM log: redact prompts/responses by default (include_prompts=False) to
  avoid logging sensitive data; metadata (timing, provider) always logged
- Fix implicit string concatenation in error messages

Author: Saurabh Badenkal

examples/advanced/datascience_risk_assessment.py

@@ -561,8 +561,13 @@ def _summarize_with_template(flagged_df):
     return summaries
 
 
-def _export_llm_log(llm_complete):
-    """Export LLM interaction log (prompts, responses, timing) to a text file."""
+def _export_llm_log(llm_complete, include_prompts=False):
+    """Export LLM interaction log (timing, provider metadata) to a text file.
+
+    By default, prompt and response content is not included to avoid logging
+    sensitive data (PII, customer data). Set include_prompts=True to include
+    full content for debugging.
+    """
     log_path = OUTPUT_DIR / "llm_interactions.txt"
     with open(log_path, "w", encoding="utf-8") as f:
         f.write("LLM Interaction Log\n")
@@ -576,9 +581,12 @@ def _export_llm_log(llm_complete):
 
         for i, entry in enumerate(llm_complete.log, 1):
             f.write(f"--- Call {i} ({entry['elapsed_seconds']:.2f}s) ---\n\n")
-            f.write(f"[System Prompt]\n{entry['system_prompt']}\n\n")
-            f.write(f"[User Prompt]\n{entry['user_prompt']}\n\n")
-            f.write(f"[Response]\n{entry['response']}\n\n")
+            if include_prompts:
+                f.write(f"[System Prompt]\n{entry['system_prompt']}\n\n")
+                f.write(f"[User Prompt]\n{entry['user_prompt']}\n\n")
+                f.write(f"[Response]\n{entry['response']}\n\n")
+            else:
+                f.write(f"[Response length: {len(entry['response'])} chars]\n\n")
 
     print(f"[OK] LLM interaction log saved to {log_path}")
 

examples/advanced/prodev_quick_start.py

@@ -106,7 +106,7 @@ def run_demo(client):
     print(f"[INFO] Output folder: {OUTPUT_DIR.resolve()}")
 
     # -- Step 1: Create 4 tables --
-    primary_name_col = step1_create_tables(client)
+    primary_name_col, primary_id_col = step1_create_tables(client)
 
     # -- Step 2: Create relationships --
     step2_create_relationships(client)
@@ -118,7 +118,7 @@ def run_demo(client):
     step4_query_and_analyze(client, customer_ids, primary_name_col)
 
     # -- Step 5: Update and delete --
-    step5_update_and_delete(client, task_ids, primary_name_col)
+    step5_update_and_delete(client, task_ids, primary_name_col, primary_id_col)
 
     # -- Step 6: Cleanup --
     cleanup(client)
@@ -148,10 +148,11 @@ def step1_create_tables(client):
             f"{TABLE_CUSTOMER}_Revenue": "money",
         },
     )
-    # The primary name column logical name is returned by tables.create()
-    # so we know exactly what key to use in create payloads.
+    # The primary column logical names are returned by tables.create()
+    # so we know exactly what keys to use in payloads and queries.
     primary_name_col = result.primary_name_attribute
-    print(f"[OK] Created table: {TABLE_CUSTOMER} (primary column: {primary_name_col})")
+    primary_id_col = result.primary_id_attribute
+    print(f"[OK] Created table: {TABLE_CUSTOMER} (name: {primary_name_col}, id: {primary_id_col})")
 
     # Project table
     client.tables.create(
@@ -186,9 +187,9 @@ def step1_create_tables(client):
     )
     print(f"[OK] Created table: {TABLE_TIMEENTRY}")
     print(f"[OK] All 4 tables created (suffix: {SUFFIX})")
-    print(f"[INFO] Primary name column: '{primary_name_col}'")
+    print(f"[INFO] Primary name column: '{primary_name_col}', ID column: '{primary_id_col}'")
 
-    return primary_name_col
+    return primary_name_col, primary_id_col
 
 
 # ================================================================
@@ -451,7 +452,7 @@ def step4_query_and_analyze(client, customer_ids, primary_name_col):
 # ================================================================
 
 
-def step5_update_and_delete(client, task_ids, primary_name_col):
+def step5_update_and_delete(client, task_ids, primary_name_col, primary_id_col):
     """Demonstrate update and delete with DataFrames."""
     print("\n" + "-" * 60)
     print("STEP 5: Update and delete records")
@@ -460,13 +461,14 @@ def step5_update_and_delete(client, task_ids, primary_name_col):
     status_col = f"{TABLE_TASK}_Status"
 
     # Update: mark first two tasks as "Complete"
+    # Use primary_id_col (from tables.create metadata) as the ID column name
     update_df = pd.DataFrame(
         {
-            f"{TABLE_TASK}id": [task_ids.iloc[0], task_ids.iloc[1]],
+            primary_id_col: [task_ids.iloc[0], task_ids.iloc[1]],
             status_col: ["Complete", "Complete"],
         }
     )
-    client.dataframe.update(TABLE_TASK, update_df, id_column=f"{TABLE_TASK}id")
+    client.dataframe.update(TABLE_TASK, update_df, id_column=primary_id_col)
     print(f"[OK] Updated 2 tasks to 'Complete'")
 
     # Delete: remove the last task

src/PowerPlatform/Dataverse/operations/dataframe.py

@@ -142,7 +142,7 @@ def get(
             rows.extend(row.data for row in batch)
 
         if not rows:
-            return pd.DataFrame()
+            return pd.DataFrame(columns=select) if select else pd.DataFrame()
         return pd.DataFrame.from_records(rows)
 
     # ----------------------------------------------------------------- create
@@ -287,7 +287,7 @@ def update(
         change_columns = [column for column in changes.columns if column != id_column]
         if not change_columns:
             raise ValueError(
-                "No columns to update. The DataFrame must contain at least one column " "besides the id_column."
+                "No columns to update. The DataFrame must contain at least one column besides the id_column."
             )
         change_list = dataframe_to_records(changes[change_columns], na_as_null=clear_nulls)
 
@@ -348,7 +348,7 @@ def delete(
         invalid = [ids.index[i] for i, v in enumerate(raw_list) if not isinstance(v, str) or not v.strip()]
         if invalid:
             raise ValueError(
-                f"ids Series contains invalid values at index(es) {invalid}. " "All IDs must be non-empty strings."
+                f"ids Series contains invalid values at index(es) {invalid}. " f"All IDs must be non-empty strings."
             )
         id_list = [v.strip() for v in raw_list]