[processing] Fix "Split with lines" algorithm can get stuck in

an endless loop if geos difference result is numerically unstable Amend the test to determine whether a split occurred from an exact topological test to instead test with a very small tolerance. This avoids us flagging that a split has occurred when the actual results have only varied by a insignificant amount due to precision of the GEOS difference op. This fixes a hang which can otherwise result where we keep splitting the same pair of lines and think that a change is occurring every time.
qgis · Oct 15, 2020 · 05645b3 · 05645b3
1 parent bf319fd
commit 05645b3
Showing 1 changed file with 6 additions and 4 deletions.
diff --git a/src/analysis/processing/qgsalgorithmsplitwithlines.cpp b/src/analysis/processing/qgsalgorithmsplitwithlines.cpp
@@ -209,12 +209,14 @@ QVariantMap QgsSplitWithLinesAlgorithm::processAlgorithm( const QVariantMap &par
 
               // splitGeometry: If there are several intersections
               // between geometry and splitLine, only the first one is considered.
-              if ( result == QgsGeometry::Success ) // split occurred
+              if ( result == QgsGeometry::Success )
               {
-                if ( inGeom.isGeosEqual( before ) )
+                // sometimes the resultant geometry has changed from the input, but only because of numerical precision issues.
+                // and is effectively indistinguishable from the input. By testing the Hausdorff distance is less than this threshold
+                // we are checking that the maximum "change" between the result and the input is actually significant enough to be meaningful...
+                if ( inGeom.hausdorffDistance( before ) < 1e-12 )
                 {
-                  // bug in splitGeometry: sometimes it returns 0 but
-                  // the geometry is unchanged
+                  // effectively no change!!
                   outGeoms.append( inGeom );
                 }
                 else