Update app.py

app.py

@@ -633,59 +633,130 @@ def build_app():
             )
 
         def _rows_to_html(rows):
-            """Convert agent row data → (html_str, json_str) for the custom table."""
-            if not rows:
+            """Convert agent row data → (html_str, json_str).
+            Accepts list-of-lists, list-of-dicts, list-of-tuples, or pd.DataFrame.
+            """
+            if rows is None:
+                rows = []
+
+            # ── Normalise to list-of-lists in TABLE_COLS order ──────────────
+            # Alias map: agent may use slightly different key names
+            _ALIASES = {
+                "top evidence (truncated)": "Top Evidence",
+                "top evidence":             "Top Evidence",
+                "topic label":              "Topic Label",
+                "rename to":                "Rename To",
+                "approve":                  "Approve",
+                "reasoning":                "Reasoning",
+                "sentences":                "Sentences",
+                "papers":                   "Papers",
+                "#":                        "#",
+            }
+
+            if isinstance(rows, pd.DataFrame):
+                # Normalise column names via alias map
+                rows.columns = [_ALIASES.get(c.lower(), c) for c in rows.columns]
+                for col in TABLE_COLS:
+                    if col not in rows.columns:
+                        rows[col] = ""
+                padded = rows[TABLE_COLS].fillna("").values.tolist()
+
+            elif rows and isinstance(rows[0], dict):
+                # Remap keys through alias table then extract in TABLE_COLS order
+                normalised = []
+                for r in rows:
+                    remapped = {_ALIASES.get(k.lower(), k): v for k, v in r.items()}
+                    normalised.append([remapped.get(col, "") for col in TABLE_COLS])
+                padded = normalised
+
+            else:
+                # List of lists / tuples — pad / truncate to exactly 8 columns
+                padded = []
+                for r in rows:
+                    row = list(r) + [""] * max(0, 8 - len(list(r)))
+                    padded.append(row[:8])
+
+            if not padded:
                 html = "<p class='rt-empty'>No topics yet — run topic modelling first.</p>"
                 return html, "[]"
 
-            padded = []
-            for r in rows:
-                row = list(r) + [""] * max(0, 8 - len(r))
-                padded.append(row[:8])
-
-            json_data = json.dumps(padded)
+            # ── Serialise to JSON (drives JS state) ─────────────────────────
+            # Convert non-serialisable types (numpy bools, NaN, etc.) safely
+            def _safe(v):
+                if v is None or (isinstance(v, float) and v != v):   # NaN check
+                    return ""
+                try:
+                    import numpy as np
+                    if isinstance(v, (np.bool_, np.integer, np.floating)):
+                        return v.item()
+                except ImportError:
+                    pass
+                return v
+
+            safe_padded = [[_safe(c) for c in row] for row in padded]
+            json_data = json.dumps(safe_padded, ensure_ascii=False)
+
+            # ── Build HTML rows ──────────────────────────────────────────────
+            def _esc(v, attr=False):
+                s = str(v) if v is not None else ""
+                s = s.replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
+                if attr:
+                    s = s.replace('"', "&quot;").replace("'", "&#39;")
+                return s
 
             rows_html = ""
-            for i, row in enumerate(padded):
+            for i, row in enumerate(safe_padded):
                 num, label, evidence, sentences, papers, approve, rename, reasoning = row
                 checked = "checked" if str(approve).lower() in ("true", "1", "yes", "✓") else ""
-                # Escape for HTML attribute safety
-                rename_esc   = str(rename).replace('"', "&quot;")
-                reason_esc   = str(reasoning).replace("<", "&lt;").replace(">", "&gt;")
-                evidence_esc = str(evidence).replace("<", "&lt;").replace(">", "&gt;")
-                label_esc    = str(label).replace("<", "&lt;").replace(">", "&gt;")
                 rows_html += f"""
         <tr data-row="{i}">
-          <td class="rt-col-num">{num}</td>
-          <td class="rt-col-label">{label_esc}</td>
-          <td class="rt-col-evid">{evidence_esc}</td>
-          <td class="rt-col-sent">{sentences}</td>
-          <td class="rt-col-paper">{papers}</td>
+          <td class="rt-col-num">{_esc(num)}</td>
+          <td class="rt-col-label">{_esc(label)}</td>
+          <td class="rt-col-evid">{_esc(evidence)}</td>
+          <td class="rt-col-sent">{_esc(sentences)}</td>
+          <td class="rt-col-paper">{_esc(papers)}</td>
           <td class="rt-col-appr">
             <input type="checkbox" class="rt-check" {checked}
               onchange="rtUpdate({i},5,this.checked)">
           </td>
           <td class="rt-col-rename">
-            <input type="text" class="rt-input" value="{rename_esc}"
-              oninput="rtUpdate({i},6,this.value)" placeholder="new label…">
+            <input type="text" class="rt-input"
+              value="{_esc(rename, attr=True)}"
+              oninput="rtUpdate({i},6,this.value)"
+              placeholder="new label…">
           </td>
           <td class="rt-col-reason">
             <textarea class="rt-textarea" rows="2"
-              oninput="rtUpdate({i},7,this.value)" placeholder="notes…">{reason_esc}</textarea>
+              oninput="rtUpdate({i},7,this.value)"
+              placeholder="notes…">{_esc(reasoning)}</textarea>
           </td>
         </tr>"""
 
+            # ── Assemble full HTML block ─────────────────────────────────────
+            # Use a unique suffix so re-renders don't collide with stale globals
+            uid = id(safe_padded)
             html = f"""<div id="review-table-wrap">
 <script>
 (function(){{
+  // Initialise or refresh the shared row-data store
   window._rtData = {json_data};
+
   window.rtUpdate = function(idx, col, val) {{
+    if (!window._rtData || !window._rtData[idx]) return;
     window._rtData[idx][col] = val;
-    var el = document.querySelector('#review-json-state textarea');
-    if (!el) return;
-    var setter = Object.getOwnPropertyDescriptor(HTMLTextAreaElement.prototype,'value').set;
+    // Push updated JSON into the hidden Gradio textbox
+    // Try both possible DOM locations Gradio uses across versions
+    var el = (
+      document.querySelector('#review-json-state textarea') ||
+      document.querySelector('[id*="review-json-state"] textarea') ||
+      document.querySelector('textarea[data-testid="textbox"]')
+    );
+    if (!el) {{ console.warn("rtUpdate: hidden textbox not found"); return; }}
+    var setter = Object.getOwnPropertyDescriptor(
+      window.HTMLTextAreaElement.prototype, 'value'
+    ).set;
     setter.call(el, JSON.stringify(window._rtData));
-    el.dispatchEvent(new Event('input',{{bubbles:true}}));
+    el.dispatchEvent(new Event('input', {{bubbles: true}}));
   }};
 }})();
 </script>