Use explicit stack in DouglasPeuckerLineSimplifier (#1127)

modules/core/src/main/java/org/locationtech/jts/simplify/DouglasPeuckerLineSimplifier.java

@@ -95,30 +95,58 @@ private void simplifyRingEndpoint(CoordinateList pts) {
 
   private LineSegment seg = new LineSegment();
 
-  private void simplifySection(int i, int j)
+  /**
+   * Douglas-Peucker section simplification using an explicit stack.
+   * <p>
+   * The classic recursive form can throw {@link StackOverflowError} on long,
+   * highly oscillatory linework where the furthest vertex is always adjacent
+   * to one endpoint (depth O(n)). An explicit stack keeps the same left-then-right
+   * split order while bounding memory by heap rather than the JVM call stack
+   * (see JTS #1127).
+   */
+  private void simplifySection(int i0, int j0)
   {
-    if((i+1) == j) {
-      return;
-    }
-    seg.p0 = pts[i];
-    seg.p1 = pts[j];
-    double maxDistance = -1.0;
-    int maxIndex = i;
-    for (int k = i + 1; k < j; k++) {
-      double distance = seg.distance(pts[k]);
-      if (distance > maxDistance) {
-        maxDistance = distance;
-        maxIndex = k;
+    // Pair stack: [i, j, i, j, ...]. Worst-case depth is O(n).
+    int[] stack = new int[Math.max(4, pts.length * 2)];
+    int top = 0;
+    stack[top++] = i0;
+    stack[top++] = j0;
+
+    while (top > 0) {
+      int j = stack[--top];
+      int i = stack[--top];
+      if ((i + 1) == j) {
+        continue;
       }
-    }
-    if (maxDistance <= distanceTolerance) {
-      for(int k = i + 1; k < j; k++) {
-        usePt[k] = false;
+      seg.p0 = pts[i];
+      seg.p1 = pts[j];
+      double maxDistance = -1.0;
+      int maxIndex = i;
+      for (int k = i + 1; k < j; k++) {
+        double distance = seg.distance(pts[k]);
+        if (distance > maxDistance) {
+          maxDistance = distance;
+          maxIndex = k;
+        }
+      }
+      if (maxDistance <= distanceTolerance) {
+        for (int k = i + 1; k < j; k++) {
+          usePt[k] = false;
+        }
+      }
+      else {
+        // Grow if needed (should not for size 2*n, but stay safe).
+        if (top + 4 > stack.length) {
+          int[] grown = new int[stack.length * 2];
+          System.arraycopy(stack, 0, grown, 0, top);
+          stack = grown;
+        }
+        // Push right then left so left is processed first (same as recursion).
+        stack[top++] = maxIndex;
+        stack[top++] = j;
+        stack[top++] = i;
+        stack[top++] = maxIndex;
       }
-    }
-    else {
-      simplifySection(i, maxIndex);
-      simplifySection(maxIndex, j);
     }
   }
 

modules/core/src/test/java/org/locationtech/jts/simplify/DouglasPeuckerSimplifierTest.java

@@ -165,6 +165,25 @@ public void testPolygonEndpointCollapse() {
         "POLYGON EMPTY"
         );
   }
+
+  /**
+   * High-frequency zig-zag forces O(n) recursion depth in classic recursive
+   * Douglas-Peucker (each split peels one vertex). Matches the failure mode
+   * in JTS #1127 (noisy GPS tracks). Explicit-stack implementation must
+   * complete without StackOverflowError and keep every vertex at tolerance 0.
+   */
+  public void testLongUnbalancedLineDoesNotStackOverflow() {
+    int n = 5000;
+    StringBuilder sb = new StringBuilder("LINESTRING (0 0");
+    for (int i = 1; i < n; i++) {
+      // alternate height so the furthest vertex is always adjacent to an endpoint
+      sb.append(", ").append(i).append(i % 2 == 0 ? " 0" : " 1");
+    }
+    sb.append(")");
+    Geometry geom = read(sb.toString());
+    Geometry result = DouglasPeuckerSimplifier.simplify(geom, 0.0);
+    assertEquals(n, result.getNumPoints());
+  }
 
   private void checkDP(String wkt, double tolerance, String wktExpected) {
     Geometry geom = read(wkt);