fixed stuck UI results in Bayesian

Author: rozyczko

EasyReflectometryApp/Backends/Py/analysis.py

@@ -662,35 +662,29 @@ def _on_sample_finished(self, results: list) -> None:
         """Handle successful completion of Bayesian sampling."""
         if self._is_stale_worker_signal():
             return
-        if not results:
-            logger.error('Bayesian sampling finished with empty results list')
-            self._fitting_logic.on_sample_finished()
-            self._fitter_thread = None
-            self.fittingChanged.emit()
-            self.externalFittingChanged.emit()
-            return
-        try:
-            posterior = results[0]  # {'draws', 'param_names', 'internal_bumps_object', 'logp'}
-            self._bayesian_logic.posterior = posterior
-            self._fitting_logic.on_sample_finished()
-            self._fitter_thread = None
-        except Exception:
-            logger.exception('Error storing Bayesian posterior result')
-            self._fitter_thread = None
-            self.fittingChanged.emit()
-            self.externalFittingChanged.emit()
+        # {'draws', 'param_names', 'internal_bumps_object', 'logp'}
+        posterior = results[0] if results else None
+        if not isinstance(posterior, dict) or 'draws' not in posterior:
+            # A malformed worker result is a failed run, not a finished one with no posterior.
+            logger.error('Bayesian sampling returned no usable posterior: %r', results)
+            self._on_fit_failed('Bayesian sampling returned no posterior')
             return
-        # Phase 2: compute posterior predictive, diagnostics, and rendered plots
-        try:
-            self._compute_and_publish_posterior_predictive()
-            self._compute_diagnostics()
-            self._render_corner_plot()
-            self._render_trace_plot()
-        except Exception:
-            logger.exception('Error during posterior computation / rendering')
-        finally:
-            self.fittingChanged.emit()
-            self.externalFittingChanged.emit()
+        self._bayesian_logic.posterior = posterior
+        self._fitting_logic.on_sample_finished()
+        self._fitter_thread = None
+        # Each post-processing step runs on its own, so one failing does not skip the others.
+        for step in (
+            self._compute_and_publish_posterior_predictive,
+            self._compute_diagnostics,
+            self._render_corner_plot,
+            self._render_trace_plot,
+        ):
+            try:
+                step()
+            except Exception:
+                logger.exception('Bayesian post-processing step %s failed', getattr(step, '__name__', step))
+        self.fittingChanged.emit()
+        self.externalFittingChanged.emit()
 
     def _compute_and_publish_posterior_predictive(self) -> None:
         """Compute posterior predictive reflectivity and SLD, publish to plotting."""

tests/test_analysis_bayesian.py

@@ -970,6 +970,30 @@ def test_emits_fitting_signals(self, analysis):
         assert emissions['fitting'] >= 1
         assert emissions['external'] >= 1
 
+    @pytest.mark.parametrize('results', [[], [None], [{'param_names': ['thickness']}]])
+    def test_malformed_result_is_reported_as_failure(self, analysis, results):
+        analysis._fitter_thread = 'some-worker'
+        received = []
+        analysis.fitFailed.connect(received.append)
+
+        analysis._on_sample_finished(results)
+
+        assert received == ['Bayesian sampling returned no posterior']
+        assert analysis._fitting_logic.fit_error_message == 'Bayesian sampling returned no posterior'
+        assert analysis._bayesian_logic.posterior is None
+        assert analysis._fitter_thread is None
+
+    def test_failing_step_does_not_skip_the_others(self, analysis):
+        with patch.object(analysis, '_compute_and_publish_posterior_predictive', side_effect=RuntimeError('boom')):
+            with patch.object(analysis, '_compute_diagnostics') as mock_diag:
+                with patch.object(analysis, '_render_corner_plot') as mock_corner:
+                    with patch.object(analysis, '_render_trace_plot') as mock_trace:
+                        analysis._on_sample_finished([SAMPLE_POSTERIOR_2D])
+                        mock_diag.assert_called_once()
+                        mock_corner.assert_called_once()
+                        mock_trace.assert_called_once()
+        assert analysis._bayesian_logic.posterior is SAMPLE_POSTERIOR_2D
+
 
 # ===================================================================
 # Bayesian param names