Drawing Borders in Deep Rift 9

I’ve spend a lot of time over the past couple of years working on a multiplayer strategy game called Deep Rift 9. You can read more about the game here:

Deep Rift 9 Web Site

The game is being developed by Norsedale ApS, and the initial code-base was lifted from my failed Hexadome 2 project, but has since evolved quite a bit in every direction. More importantly, DR9 has a depth and a visual appearance far beyond what I could have hoped to cook up myself, so it’s all for the better.

In any case, since we’re now entering into the treacherous minefield of public (alpha) testing 🙂 I figured it would be a good time to release some of my development notes on the project – these posts will also be available on the DR9 web site.

Starting kind-of backwards, here’s the first one on how we calculate and render the dynamic borders between players land areas.

Voronoi Diagrams

I don’t remember when I first heard or read about Voronoi Diagrams, but I do remember these guys: http://www.big-robot.com giving a rather funny presentation of how they used Voronoi diagrams to procedurally generate maps in “Sir, you are being hunted” back at Unite Nordic 2013 (jeez, time flies).

Actually, the talk is on YouTube if you’re interested , I’m in the front row, purple shirt, bald spot 🙂

There’s a good page on WikiPedia if you want the details:

WikiPedia on Voronoi

The TL;DR version: A Voronoi diagram divides an set of seed-points into convex shapes. There is one seed point in each shape and each shape contains all other points that are closer to that specific seed than they are to any other seed point. The edges of the shape naturally becomes the borders to the surrounding shapes.

Sounds perfect.

Implementation

Luckily for me, I did not have to take this all the way from mathematical concept to working algorithm since several talented people have already laid the foundation and generously shared their implementations. Unfortunately, none of them decided to do it in C# which is what I needed, so I did have to port the code.

The implementation I ended up using is a C++ derivation of Steven Fortune’s original C implementation from AT&T, made by Shane O’sullivan:

Shane O’Sullivans C++ Voronoi implementation

The original source is available from Stevens web page, here:

Steven Fortune’s original C implementation

My C# adaption is below if anyone feels like saving a few hours of tedious pointer-to-reference conversion ;).

Performance

Even though performance is not critical to DR9 – the points that serve as Voronoi “seeds” in DR9 are the cities and they don’t move, nor are they created or destroyed at a very high frequency so the Voronoi generator does not need to run at anywhere near full frame-rate speed – I still decided to do some quick performance tests.

I got rid of a bunch of C and C++ legacy annoyances like additional counters and funky size conversions that are not needed in C# and got a small ~20% performance bump for my efforts (probably not because there was anything wrong with the original code, but more likely because I had earned a penalty when porting, trying to be strict rather than fast).

With this out of the way, and not feeling too eager to mess with the core algorithm, I zoned in on the use of hash tables. Even though hash tables are generally good for performance – at least at scale, something about this particular use-case just didn’t taste right.

Going on nothing but a hunch, I did not really expect to find any gains, so as a quick test, I just dumbed the hash tables down to a size of one, effectively turning them into complicated arrays. Somewhat to my surprise performance increased by another 15%. I initially assumed that my dataset was simply too small for the hashtables to be useful, and I was just suffering the overhead without the benefit, so I bumped my seed count from 128 to 2048.

This had the completely unexpected effect of being 25% faster than the hashed version. Not only was hashing slower, it also scaled worse:

Screen Shot 2017-09-11 at 22.38.17

I’m not sure why this is and I have not had time to dig deeper, but there’s a number of things that could be worth investigating:

  1. The nature of the algorithm may not need real random access, or my data set may have been skewed in such a way that it didn’t – If the searches mostly find candidates in close proximity to where it is, a hashtable has very little use. I did try to shuffle the data but that only pushed the results further in the direction of the non-hashed version, if only marginally so.
  2. The hash table implementation may have been broken, or using an inappropriate hashing function. I looked at this briefly and it seemed reasonable, using spatial position for hashing which matches the even distribution of my seeds nicely. I didn’t do any profiling of the actual buckets to see if the fill was uneven, but I doubt it.
  3. The hash table may break optimization done by either the C# compiler or run-time or by the CPU (this could be anything from overflowing the pipeline or missing the cache to fouling up the CPUs branch prediction). Generally, less code, fewer branches, fewer method calls and less messing with shared variables results in better run-time optimization, even if your loops gets a bit longer. Long lesson learned by years of Java programming – you can really mess the hotspot compiler up if you try to be clever on its behalf.

Personally, I have my money on #3 🙂

In any case, once I had the Voronoi structure, connecting neighbouring shapes with the same owner was trivial, and so was traversing the surrounding edges to create a mesh for the border.

The following screenshot shows borders in red, and lines connecting neighbours in gray. Seeds are considered to have the same owner if they have the same color.

Screen Shot 2017-09-11 at 20.50.11

What did turn out to be something of a hassle, was rendering it.

Ideally I just wanted it painted as a flat configurable-width line on top of the map, but the map is not flat and it is not “a” map, but rather a bunch of map-patches that are dynamically created and destroyed as the view changes. Because the shape of the Voronoi cells are highly dependent on neighbouring patches (in fact that’s all it depends on) it was not practical to create Voronoi meshes for each patch – it had to be created independently of the patches.

Creating a mesh that followed the terrain was not trivial for two reasons. The first is that the actual map patches are not a true representation of the procedural height map. The patches are sub-divided and smoothened to create a nicer terrain and also contain objects such as trees and mountains and thus the height at any given location was not readily available. The second problem is that the Voronoi diagram extends towards infinity (or 32000 as infinity is defined in DR9 🙂 ) and creating a mesh with that many vertices would never work.

I had to find a way to draw the lines on the map, preferably by storing only the end-points of each border.

My first attempt involved a rather exotic use of the depth buffer and a custom shader. I build a mesh of vertical planes and rendered their intersection with the map patches by measuring the distance in the depth buffer. It worked ok from certain angles, but not at all for borders that were perfectly vertical compared to the camera (which turned out to be a fairly common case).

Screen Shot 2017-08-25 at 09.31.58

Fiddling with the shader, I was able to make it less horrible, but never really good (notice how lines get thinner as they become vertical):

Screen Shot 2017-08-25 at 09.55.55

My second attempt was to create extruded upside-down U shapes instead of vertical planes. I used the actual procedural height data plus some small distance as the height of the inverted U, but limited to 10 points per edge. This meant that most “real” borders would get subdivided into small enough pieces that they kind of followed the terrain, while the outer ones that extended toward “infinity” would not generate ridiculous amounts of vertices.

The problem with this approach was that the border would sometimes extend high above the terrain, and other times penetrate through it, disappearing all together.

The final fix turned out to be as simple as it comes: I modified the shader to ignore Z-test and reduced the geometry to be only the flat upper face of the inverted U. Because the line kind of follows the terrain, it never gets wildly out of place, and even when it does deviate, nobody can tell because there is no telling where it *should* have been.

This would not have worked if the user was able to tilt or rotate the camera as it would have broken the illusion, but as it happens, they can’t 🙂

Screen Shot 2017-09-11 at 12.18.45

My C# port is below. The way this works is that you create an instance of VoronoidGenerator supplying any class that you want to associate with the seeds – it does not matter what it is, only that you can somehow convert it into a 2D location.

You then call AddSeed with each of the seeds you want included and finally Generate to build the actual Graph. Generate takes a locationResolver as its last parameter – this is a function that given a seed of your declared type must return a 2D position.

When Generate completes, you can get the first edge in the graph with GetFirstEdge and then loop the graph by following edges (edge.next). Each graph edge contains the border end points as well as a reference to the seeds on either side of the border.

/*
* The author of this software is Steven Fortune.  Copyright (c) 1994 by AT&T
* Bell Laboratories.
* Permission to use, copy, modify, and distribute this software for any
* purpose without fee is hereby granted, provided that this entire notice
* is included in all copies of any software which is or includes a copy
* or modification of this software and in all copies of the supporting
* documentation for such software.
* THIS SOFTWARE IS BEING PROVIDED "AS IS", WITHOUT ANY EXPRESS OR IMPLIED
* WARRANTY.  IN PARTICULAR, NEITHER THE AUTHORS NOR AT&T MAKE ANY
* REPRESENTATION OR WARRANTY OF ANY KIND CONCERNING THE MERCHANTABILITY
* OF THIS SOFTWARE OR ITS FITNESS FOR ANY PARTICULAR PURPOSE.
*/

/* 
* This code was originally written by Stephan Fortune in C code.  I, Shane O'Sullivan, 
* have since modified it, encapsulating it in a C++ class and, fixing memory leaks and 
* adding accessors to the Voronoi Edges.
* Permission to use, copy, modify, and distribute this software for any
* purpose without fee is hereby granted, provided that this entire notice
* is included in all copies of any software which is or includes a copy
* or modification of this software and in all copies of the supporting
* documentation for such software.
* THIS SOFTWARE IS BEING PROVIDED "AS IS", WITHOUT ANY EXPRESS OR IMPLIED
* WARRANTY.  IN PARTICULAR, NEITHER THE AUTHORS NOR AT&T MAKE ANY
* REPRESENTATION OR WARRANTY OF ANY KIND CONCERNING THE MERCHANTABILITY
* OF THIS SOFTWARE OR ITS FITNESS FOR ANY PARTICULAR PURPOSE.
*/

/**
 * C# adaption for use with Unity by Niels Jørgensen, BoaNeo AB (http://www.boaneo.com)
 *
 * Permission to use, copy, modify, and distribute this software for any
 * purpose without fee is hereby granted, provided that this entire notice
 * is included in all copies of any software which is or includes a copy
 * or modification of this software and in all copies of the supporting
 * documentation for such software.
 * THIS SOFTWARE IS BEING PROVIDED "AS IS", WITHOUT ANY EXPRESS OR IMPLIED
 * WARRANTY.  IN PARTICULAR, NEITHER THE AUTHORS NOR AT&T MAKE ANY
 * REPRESENTATION OR WARRANTY OF ANY KIND CONCERNING THE MERCHANTABILITY
 * OF THIS SOFTWARE OR ITS FITNESS FOR ANY PARTICULAR PURPOSE.
 */

using System;
using System.Collections.Generic;
using UnityEngine;

namespace BoaNeoTools.Source.MathNoHash
{
  class Pool where T : new()
  {
    private List _memory = new List();
    private int     _free;
    public int length { get { return _free; } }

    public void Reset()
    {
      _free = 0;
    }

    public T Allocate()
    {
      if (_free >= _memory.Count)
        _memory.Add(new T());

      T obj = _memory[_free++];
      return obj;
    }

    public void Claim(T curr)
    {
      int idx = _memory.IndexOf(curr);
      if (idx >= 0 && idx < _free)
      {
        _memory.RemoveAt(idx);
        _free--;
        _memory.Add(curr);
      }
    }
  }

  class Point 
  {
    public float x;
    public float y;
  }

  class Seed : Point
  {
    public object source;
    public int seednbr;
    public int refcnt;

    public Seed()    {    }
    public Seed(object seedSource)
    {
      source = seedSource;
    }
  }

  class Edge
  {
    public float a;
    public float b;
    public float c;
    public Seed[] ep = new Seed[2];
    public Seed[] reg = new Seed[2];
    public int edgenbr;
  }

  public class GraphEdge
  {
    public float x1;
    public float y1;
    public float x2;
    public float y2;
    public T s1;
    public T s2;
    public GraphEdge next;
  }


  class Halfedge
  {
    public Halfedge ELleft;
    public Halfedge ELright;
    public Edge ELedge;
    public int ELrefcnt;
    public int ELpm;
    public Seed vertex;
    public float ystar;
    public Halfedge PQnext;
  }

  public class VoronoiGenerator
  {
    Edge DELETED = new Edge(); // Used for ugly comparison

    const int le = 0;
    const int re = 1;

    private List _userSeeds = new List();
    private int _seedIdx;

    private GraphEdge _allEdges;

    private Pool _pointPool = new Pool();
    private Pool _graphEdgePool = new Pool();
    private Pool _halfedgePool = new Pool();
    private Pool _internalSeedPool = new Pool();
    private Pool _edgePool = new Pool();

    Halfedge ELleftend;
    Halfedge ELrightend;

    float xmin, xmax, ymin, ymax;

    int nvertices;

    Halfedge PQhash;
    int PQcount;

    float _perimeterXmin, _perimeterXmax, _perimeterYmin, _perimeterYmax;

    float _minDistanceBetweenSeeds;

    public GraphEdge GetFirstEdge()
    {
      return _allEdges;
    }

    public void AddSeed( T seedSource )
    {
      _userSeeds.Add( new Seed(seedSource) );
    }

    public void RemoveSeed( T seedSource )
    {
      for (int i = 0; i < _userSeeds.Count; i++)
      {
        if (_userSeeds[i].source == (object)seedSource)
        {
          _userSeeds.RemoveAt(i);
          return;
        }
      }
    }

    public void Generate(float minX, float maxX, float minY, float maxY, float minDist, Func locationResolver)
    {
      if (_userSeeds.Count == 0)
        return;
      
      CleanupEdges();
      int i;

      _minDistanceBetweenSeeds = minDist;

      _graphEdgePool.Reset();
      _internalSeedPool.Reset();
      _pointPool.Reset();

      xmin = float.MaxValue;
      ymin = float.MaxValue;
      xmax = float.MinValue;
      ymax = float.MinValue;

      for (i = 0; i < _userSeeds.Count; i++)       {         Vector2 p0 = locationResolver((T)_userSeeds[i].source);                  _userSeeds[i].x = p0.x;         _userSeeds[i].y = p0.y;         _userSeeds[i].seednbr = i;         _userSeeds[i].refcnt = 0;         if (p0.x  xmax)           xmax = p0.x;         if (p0.y  ymax)           ymax = p0.y;       }       _userSeeds.Sort((s1, s2) =>
      {
        if (s1.y  s2.y) return 1;
        if (s1.x  s2.x) return 1;
        return 0;
      });

      _seedIdx = 0;
      _edgePool.Reset();
      nvertices = 0;

      float temp;
      if (minX > maxX)
      {
        temp = minX;
        minX = maxX;
        maxX = temp;
      }
      if (minY > maxY)
      {
        temp = minY;
        minY = maxY;
        maxY = temp;
      }
      _perimeterXmin = minX;
      _perimeterYmin = minY;
      _perimeterXmax = maxX;
      _perimeterYmax = maxY;

      _seedIdx = 0;
      Voronoi();
    }

    private void ELinitialize()
    {
      _halfedgePool.Reset();
      ELleftend = HEcreate(null, 0);
      ELrightend = HEcreate(null, 0);
      ELleftend.ELleft = null;
      ELleftend.ELright = ELrightend;
      ELrightend.ELleft = ELleftend;
      ELrightend.ELright = null;
    }

    Halfedge HEcreate(Edge e, int pm)
    {
      Halfedge answer = _halfedgePool.Allocate();
      answer.ELedge = e;
      answer.ELpm = pm;
      answer.PQnext = null;
      answer.vertex = null;
      answer.ELrefcnt = 0;
      answer.ELleft = null;
      answer.ELright = null;
      answer.ystar = 0;
      return answer;
    }

    void ELinsert(Halfedge lb, Halfedge newHe)
    {
      newHe.ELleft = lb;
      newHe.ELright = lb.ELright;
      lb.ELright.ELleft = newHe;
      lb.ELright = newHe;
    }

    Halfedge ELleftbnd(Point p)
    {
      // Use hash table to get close to desired halfedge 
      Halfedge he = ELrightend;
      //ntry += 1;
      // Now search linear list of halfedges for the correct one 
      do
      {
        he = he.ELleft;
      } while (he != ELleftend && !RightOf(he, p));

      return he;
    }

    // This delete routine can't reclaim node, since pointers from hash table may be present.   
    void ELdelete(Halfedge he)
    {
      he.ELleft.ELright = he.ELright;
      he.ELright.ELleft = he.ELleft;
      he.ELedge = DELETED;
    }

    Seed LeftReg(Halfedge he)
    {
      if (he.ELedge == null)
        return _userSeeds[0];
      return he.ELpm == le ? he.ELedge.reg[le] : he.ELedge.reg[re];
    }

    Seed RightReg(Halfedge he)
    {
      if (he.ELedge == null) //if this halfedge has no edge, return the bottom seed (whatever that is)
        return _userSeeds[0];

      //if the ELpm field is zero, return the seed 0 that this edge bisects, otherwise return seed number 1
      return(he.ELpm == le ? he.ELedge.reg[re] : he.ELedge.reg[le]);
    }

    Edge Bisect(Seed s1, Seed s2)
    {
      Edge newedge = _edgePool.Allocate();

      newedge.edgenbr = 0;
      newedge.a = newedge.b = newedge.c = 0;
      newedge.reg[0] = s1; //store the seeds that this edge is bisecting
      newedge.reg[1] = s2;
      RefBump(s1);
      RefBump(s2);
      newedge.ep[0] = null; //to begin with, there are no endpoints on the bisector - it goes to infinity
      newedge.ep[1] = null;

      float dx = s2.x - s1.x; //get the difference in x dist between the seeds
      float dy = s2.y - s1.y;
      float adx = dx > 0 ? dx : -dx; //make sure that the difference in positive
      float ady = dy > 0 ? dy : -dy;
      newedge.c = (float) (s1.x * dx + s1.y * dy + (dx * dx + dy * dy) * 0.5); //get the slope of the line

      if (adx > ady)
      {
        newedge.a = 1.0f;
        newedge.b = dy / dx;
        newedge.c /= dx; //set formula of line, with x fixed to 1
      }
      else
      {
        newedge.b = 1.0f;
        newedge.a = dx / dy;
        newedge.c /= dy; //set formula of line, with y fixed to 1
      }

      newedge.edgenbr = _edgePool.length - 1;

      return newedge;
    }

    //create a new seed where the HalfEdges el1 and el2 intersect - note that the Point in the argument list is not used, don't know why it's there
    Seed Intersect(Halfedge el1, Halfedge el2)
    {
      Edge e1 = el1.ELedge;
      Edge e2 = el2.ELedge;
      if (e1 == null || e2 == null)
        return null;

      //if the two edges bisect the same parent, return null
      if (e1.reg[1] == e2.reg[1])
        return null;

      float d = e1.a * e2.b - e1.b * e2.a;
      if (-1.0e-10 < d && d < 1.0e-10)
        return null;

      float xint = (e1.c * e2.b - e2.c * e1.b) / d;
      float yint = (e2.c * e1.a - e1.c * e2.a) / d;

      Halfedge el;
      Edge e;
      if ((e1.reg[1].y < e2.reg[1].y) ||           (e1.reg[1].y == e2.reg[1].y &&            e1.reg[1].x = e.reg[1].x;       if ((right_of_seed && el.ELpm == le) || (!right_of_seed && el.ELpm == re))         return null;       //create a new seed at the point of intersection - this is a new vector event waiting to happen       Seed v = _internalSeedPool.Allocate();       v.refcnt = 0;       v.seednbr = 0;       v.x = xint;       v.y = yint;       return v;     }     // returns 1 if p is to right of halfedge e      bool RightOf(Halfedge el, Point p)     {       Edge e = el.ELedge;       Seed topSeed = e.reg[1];       bool right_of_seed = p.x > topSeed.x;
      if (right_of_seed && el.ELpm == le) return true;
      if (!right_of_seed && el.ELpm == re) return false;

      bool above;
      if (e.a == 1.0)
      {
        float dyp = p.y - topSeed.y;
        float dxp = p.x - topSeed.x;
        bool fast = false;
        if ((!right_of_seed & (e.b = 0.0)))
        {
          above = dyp >= e.b * dxp;
          fast = above;
        }
        else
        {
          above = p.x + p.y * e.b > e.c;
          if (e.b < 0.0) above = !above;
          if (!above) fast = true;
        }
        if (!fast)
        {
          var dxs = topSeed.x - (e.reg[0]).x;
          above = e.b * (dxp * dxp - dyp * dyp) <                   dxs * dyp * (1.0 + 2.0 * dxp / dxs + e.b * e.b);           if (e.b  t2 * t2 + t3 * t3;       }       return el.ELpm == le ? above : !above;     }     void EndPoint(Edge e, int lr, Seed s)     {       e.ep[lr] = s;       RefBump(s);       if (e.ep[re - lr] == null)         return;       ClipLine(e);       DeRef(e.reg[le]);       DeRef(e.reg[re]);       _edgePool.Claim(e);     }     float Distance(Point s, Point t)     {       float dx = s.x - t.x;       float dy = s.y - t.y;       return Mathf.Sqrt(dx * dx + dy * dy);     }     void MakeVertex(Seed v)     {       v.seednbr = nvertices;       nvertices += 1;     }     void DeRef(Seed v)     {       v.refcnt -= 1;       if (v.refcnt == 0)         _internalSeedPool.Claim(v);     }     void RefBump(Seed v)     {       v.refcnt += 1;     }     //push the HalfEdge into the ordered linked list of vertices     void PQinsert(Halfedge he, Seed v, float offset)     {       he.vertex = v;       RefBump(v);       he.ystar = v.y + offset;       Halfedge last = PQhash;       Halfedge next;       while ((next = last.PQnext) != null &&              (he.ystar > next.ystar ||
              (he.ystar == next.ystar && v.x > next.vertex.x)))
      {
        last = next;
      }
      he.PQnext = last.PQnext;
      last.PQnext = he;
      PQcount += 1;
    }

    //remove the HalfEdge from the list of vertices 
    void PQdelete(Halfedge he)
    {
      if (he.vertex != null)
      {
        Halfedge last = PQhash;
        while (last.PQnext != he)
          last = last.PQnext;

        last.PQnext = he.PQnext;
        PQcount -= 1;
        DeRef(he.vertex);
        he.vertex = null;
      }
    }

    bool PQempty()
    {
      return PQcount == 0;
    }

    Point PQ_min()
    {
      Point answer = _pointPool.Allocate();

      answer.x = PQhash.PQnext.vertex.x;
      answer.y = PQhash.PQnext.ystar;

      return answer;
    }

    Halfedge PQextractmin()
    {
      Halfedge curr = PQhash.PQnext;
      PQhash.PQnext = curr.PQnext;
      PQcount -= 1;
      return curr;
    }

    void PQinitialize()
    {
      PQcount = 0;
      if (PQhash == null)
      {
        PQhash = new Halfedge();
      }
    }

    void CleanupEdges()
    {
      _allEdges = null;
    }

    void PushGraphEdge(Seed s1, Seed s2, float x1, float y1, float x2, float y2)
    {
      GraphEdge newEdge = _graphEdgePool.Allocate(); // new GraphEdge();
      newEdge.next = _allEdges;
      _allEdges = newEdge;
      newEdge.s1 = (T)s1.source;
      newEdge.s2 = (T)s2.source;
      newEdge.x1 = x1;
      newEdge.y1 = y1;
      newEdge.x2 = x2;
      newEdge.y2 = y2;
    }

    void out_triple(Seed s1, Seed s2, Seed s3)
    {
  //    Debug.Log("OutTriple ()");
    }

    void ClipLine(Edge e)
    {
      if (e.reg[0] == null)
        return;

      float x1 = e.reg[0].x;
      float x2 = e.reg[1].x;
      float y1 = e.reg[0].y;
      float y2 = e.reg[1].y;

      //if the distance between the two points this line was created from is less than the square root of 2, then ignore it
      if (Mathf.Sqrt(((x2 - x1) * (x2 - x1)) + ((y2 - y1) * (y2 - y1))) = 0.0)
      {
        s1 = e.ep[1];
        s2 = e.ep[0];
      }
      else
      {
        s1 = e.ep[0];
        s2 = e.ep[1];
      }

      if (e.a == 1.0)
      {
        y1 = _perimeterYmin;
        if (s1 != null && s1.y > _perimeterYmin)
        {
          y1 = s1.y;
        }
        if (y1 > _perimeterYmax)
        {
          y1 = _perimeterYmax;
        }
        x1 = e.c - e.b * y1;
        y2 = _perimeterYmax;
        if (s2 != null && s2.y < _perimeterYmax)           y2 = s2.y;         if (y2  _perimeterXmax) & (x2 > _perimeterXmax)) | ((x1 < _perimeterXmin) & (x2  _perimeterXmax)         {           x1 = _perimeterXmax;           y1 = (e.c - x1) / e.b;         }         ;         if (x1  _perimeterXmax)         {           x2 = _perimeterXmax;           y2 = (e.c - x2) / e.b;         }         ;         if (x2  _perimeterXmin)           x1 = s1.x;         if (x1 > _perimeterXmax)
        {
          x1 = _perimeterXmax;
        }
        y1 = e.c - e.a * x1;
        x2 = _perimeterXmax;
        if (s2 != null && s2.x < _perimeterXmax)           x2 = s2.x;         if (x2  _perimeterYmax) & (y2 > _perimeterYmax)) | ((y1 < _perimeterYmin) & (y2  _perimeterYmax)
        {
          y1 = _perimeterYmax;
          x1 = (e.c - y1) / e.a;
        }
        ;
        if (y1  _perimeterYmax)
        {
          y2 = _perimeterYmax;
          x2 = (e.c - y2) / e.a;
        }
        ;
        if (y2 < _perimeterYmin)
        {
          y2 = _perimeterYmin;
          x2 = (e.c - y2) / e.a;
        }
      }

      PushGraphEdge(e.reg[0],e.reg[1], x1, y1, x2, y2);
    }

    private void Voronoi()
    {
      PQinitialize( );
      ELinitialize( );

      NextSeed();
      Seed new_seed = NextSeed();
      
      Point newintstar = null;
      while (true)
      {
        if (!PQempty())
          newintstar = PQ_min();

        //if the lowest seed has a smaller y value than the lowest vector intersection, process the seed otherwise process the vector intersection  
        if (new_seed != null && (PQempty() || new_seed.y < newintstar.y || (new_seed.y == newintstar.y && new_seed.x  top.y) //if the seed to the left of the event is higher than the seed
          {
            //to the right of it, then swap them and set the 'pm' variable to 1
            var temp = bot;
            bot = top;
            top = temp;
            pm = re;
          }
          Edge e = Bisect(bot, top); //create an Edge (or line) that is between the two seeds. This creates
          //the formula of the line, and assigns a line number to it
          Halfedge bisector = HEcreate(e, pm); //create a HE from the Edge 'e', and make it point to that edge with its ELedge field
          ELinsert(llbnd, bisector); //insert the new bisector to the right of the left HE
          EndPoint(e, re - pm, v); //set one endpoint to the new edge to be the vector point 'v'.
          //If the seed to the left of this bisector is higher than the right
          //seed, then this endpoint is put in position 0; otherwise in pos 1
          DeRef(v); //delete the vector 'v'

          //if left HE and the new bisector don't intersect, then delete the left HE, and reinsert it 
          Seed p;
          if ((p = Intersect(llbnd, bisector)) != null)
          {
            PQdelete(llbnd);
            PQinsert(llbnd, p, Distance(p, bot));
          }

          //if right HE and the new bisector don't intersect, then reinsert it 
          if ((p = Intersect(bisector, rrbnd)) != null)
          {
            PQinsert(bisector, p, Distance(p, bot));
          }
        }
        else break;
      }


      for (Halfedge lbnd = ELleftend.ELright; lbnd != ELrightend; lbnd = lbnd.ELright)
      {
        ClipLine(lbnd.ELedge);
      }
    }

    // return a single in-storage seed
    Seed NextSeed()
    {
      Seed s = null;
      if (_seedIdx < _userSeeds.Count)
      {
        s = _userSeeds[_seedIdx];
        _seedIdx += 1;
      }
      return s;
    }
  }
}

License to Code

As a software engineer I don’t often find myself in scary or life threatening situations – at least not as a direct consequence of my career choice.

Not too long ago, however, I was working on a customer project in which some core functions were lifted from a previous version of the same product. I was told to just re-use what was already there – no point in re-inventing the wheel, right?

As it turned out, the old code was not really doing what it should. This was not a revelation per-se, pretty much all code has some bug or another if you dig deep enough, but the one I found was sufficiently odd to cause me to dig a bit deeper.

Eventually it led me to this (names has been changed to protect the innocent):

private void calc(int x, int y, String sl, String l)
{
  ArrayList<Obj> list = new ArrayList<Obj>();
  list.add(new Obj(14.29,70,0));
  list.add(new Obj(10.53,95,0));
  list.add(new Obj(8.33,120,0));
  list.add(new Obj(6.90,145,0));
  list.add(new Obj(11.11,70,20));
  list.add(new Obj(8.70,95,20));
  list.add(new Obj(7.14,120,20));
  list.add(new Obj(6.06,145,20));
  list.add(new Obj(10.00,70,30));
  list.add(new Obj(8.00,95,30));
  list.add(new Obj(6.67,120,30));
  list.add(new Obj(5.71,145,30));
  list.add(new Obj(9.09,70,40));
  list.add(new Obj(7.41,95,40));
  list.add(new Obj(6.25,120,40));
  list.add(new Obj(5.41,145,40));
  list.add(new Obj(8.33,70,50));
  list.add(new Obj(6.90,95,50));
  list.add(new Obj(5.88,120,50));
  list.add(new Obj(5.13,145,50));
  list.add(new Obj(7.69,70,60));
  list.add(new Obj(6.45,95,60));
  list.add(new Obj(5.56,120,60));
  list.add(new Obj(4.88,145,60));

  for(Obj obj : list)
  {
    if(obj.getX() == x && obj.getY() == y)
    {
      Double answer = 
        obj.getValue() * Double.parseDouble(sl) * Double.parseDouble(l);
      someFunction(answer);
    }
  }
}

I did not include the source for Obj because it doesn’t really matter, so I’ll spare you the time to review what this does and just spell it out for you:

answer = sl*l*(1000/(x+y))

If someone had presented me with this out of context, I would have assumed it was a joke, but it’s not. It’s actual production code from a real-world application, and it’s just wrong on so many levels that I’m not even going to bother dissecting it.

Instead, I’d like to return to the part about my job not being scary or life threatening.

In this case the application was benign and, although stupid, the code did (almost) what it should (round-off errors aside), but seeing this piece of code, and realising that there are actually people out there who call them selves “programmers” or “engineers” who’d do something like that, it occurred to me how ridiculously dangerous bad programmers are.

It’s not just about wasting time, or being annoying, but who is to say that this person won’t end up working for the bank that looks after my money? Or some company that writes control code for nuclear submarines?

Considering that I’m hardly allowed to install a lightbulb in my own home without a license, it’s downright scary that people are allowed to work on software when they have no clue, let alone a license.

2016 MacBook PRO review

Update 2017/3/14:

I’ve been meaning to append an update to this post for a while, but have been too busy to do so. Today, I was once again reminded about it for reasons I’ll get back to shortly.

So, immediately after the original blog post, I had two major issues:

  1. It had a HW glitch in the keyboard (Hubris for being too tough on Dell, I’m sure) – basically the ‘O’ key only worked one of three times it was pressed. Apple replaced it overnight, no prob.
  2. Then I was consistently getting a black screen on start up if I had entered sleep mode with my external monitor attached. Apple tried for a short while to blame my external monitor… That stopped when they learned that it’s an Apple Thunderbolt display. There was another feeble attempt at blaming 3rd party software, so I reset the machine to factory and the problem didn’t go away. So, finally convinced, they offered to take it back, but I would then have to wait 3 months for a new one because it’s a custom build. I finally opted to just have it repaired. One more night in the shop, and everything has been fine since. So, early build, monday model, lemon, I don’t know, but I guess no matter what you pay, you can never be sure HW won’t have issues. Lesson learned.

In both cases Apple Support has been top notch (safe the external monitor thing, but that was actually kind of funny 🙂 ). When things do go bad, and they do, good customer support is all I really ask for, and Apple’s is second to none. I leave them my number, 30 seconds later they call back and walk me through the various tests and problems until there’s a solution.

HW glitches aside, there’s a couple of things that I wrote about originally that I now have a 3+ months perspective on:

Weight and size does matter. I was unsure if the modest reduction would make any difference, but it is noticeably nimbler on an every day basis. It also is way more silent than the old one and it does not have the same hotspots around the charging port, for example.

Performance, well, it’s the same story every time you upgrade. It only takes a few days to get used to it, and you then forget what it was like before. Only thing I’ve noticed consistently is that I have much less graphics issues in Unity. Before, I would get weird texture glitches every now and then (lack of GPU memory probably), and had to restart Unity. I haven’t seen that since I got the new laptop.

The USB-C port, much as I predicted, has been a non-issue. Or rather, it has turned out to be more annoying for others because they can’t borrow my charger or my lightning cable, but since that greatly reduces my risk of loosing either, I consider it an overall win 🙂

The O-led display however… Well, it’s the reason I finally decided to write this, because it’s such a ridiculous non-feature that I just had to vent. It has no practical purpose what so ever; it does nothing that I couldn’t do without it. Nothing. But if it was just that, if it was just a party trick, I could try to forget how much it probably added to the BOM and simply ignore it. But that’s exactly the problem, it’s not just useless. It is in fact incredibly annoying on a daily basis.

The bar have no tactile feedback, so even just using it as if it was a row of regular old function keys is a step down compared to a normal keyboard. But it gets worse because of how good the touch is. I usually hover my finger over the F7 and F8 keys when debugging, but now I can’t because it causes random taps when I don’t want them. And don’t get me started on how many times I’ve been asked if I want to turn Siri on. I do not plan to talk to my laptop. Ever. So you can stop asking, but the damn button is right above the backspace key, so if your finger is just a wee bit off, Siri will think you called. Same story in the other side where F1 is so easily activated that I get annoying help dialogs all over the place.

Apple, seriously, what were you thinking? Did this “feature” go straight from Ive’s skull directly into production – nobody actually tested it?

If you absolutely had to change the keyboard, how about you fix the goddamn cursor keys. Yes, I would like full size up/down keys also – I’m a grown up, I have grown-up-size fingers.

Yes, I could use an external keyboard, but at this price, I really shouldn’t have to.

Original post follows below:

Me and my laptop

It’s been 4 years since I bought my trusty MacBook PRO retina. That’s easily a year more than I’ve ever had a laptop before, and the first time ever that I was in doubt as to whether it made sense to get a new one.

All the laptops I’ve had before (Top of the line Dells and Lenovos, typically) has started getting tired after less than 2 years, making the last year of ownership a real drag. Part of that has been due to Windows bloat and build up of crud in the registry and file system in general, so in part I suspect the MacBook has lived longer not just because of its hardware but also because of OsX, but the hardware is certainly part of it.

(On a side note, the Dells have been notorious for needing hardware replacements; from keyboards loosing keys to graphics cards getting bugged colours or weird artefacts, I’ve made extensive use of the next-business-day service. In fact, my wife got a Dell not too long ago and had to replace the HDD after just a few weeks – They might be ok for home computers, but they’re anything but a professional tool and I’ve had my last Dell for sure).

Still, 4 years in with the MacBook, and I feel like it could easily have done a year or two more, but two things were nudging me towards a new laptop – the 256GB SSD was constantly kissing the 95% mark and I had to do frequent cleanups in my temp and trash folders to be able to work (not a bad thing per-se 🙂 ) and the graphics card was sometimes struggling a bit with two large displays running Unity, photoshop and Blender at the same time, causing the fan to be on more than I would have liked.

Enter the 2016 MacBook Pro.

IMG_1954.jpg

First of all, let’s get the price of out of the way: It’s ridiculous – there is no justification for it, other than Apple charging big bucks because they know they can. For me, as for most people in my situation I would guess, it is, however, also somewhat irrelevant.

It’s my primary work horse – the only tool I own (aside from software licenses), and I buy one every 3 (now 4) years and use it 10 hours every day – saving $1000 and risk getting a Dell-esque piece-of-crap Hardware that needs a new HDD in the middle of the week, costing me a couple of days worth of work, just isn’t an option. The fact that it lasts a year or two more than the competition means I would probably pay even more if needed (Don’t tell Apple that though 😉 )

So, with that out of the way, what am I getting for all my cash?

It’s obviously lighter and thinner – I only just got it so can’t say if this is anything I’ll really notice, but since I drag it around all the time, less is definitely better. The limited testing I’ve done suggests that it also stays cooler, which was my only real concern with the reduced size. The old one could get seriously hot around the hinge when Unity was pushing polygons at 60FPS so I had a (seemingly unwarranted) suspicion that a smaller form-factor might make it harder to get rid of the heat and thus make the problem worse.

I love the keyboard. It’s a bit click-y-ti-clack where the old one was dead silent, but the feel is just phenomenal. I actually think I type faster and make less typos, though that could be a euphoria-induced illusion ;).

The touch pad is even more awesome. I loved the old one which was leaps beyond anything I’d ever had on any other laptop, yet it pales next to the new one. Everything from the sound it makes to the very subtle, yet rewardingly haptic, feedback when you press it is just best-of-class – nothing I’ve tried on any other laptop (mine or others) are in the same ballpark – hell, they are not even in the same galaxy. It’s so good that I feel a bit bad for throwing $100 at a new magic mouse 🙂

The O-led display that replaces the function keys is… Well, it’s a gimmick. A very expensive one, I presume. I really don’t see the point of it, and seriously doubt I’ll ever use it, but I’m going to try to keep an open mind. Maybe there some kind of use for it, and I just can’t see it, but this is probably the one thing I would have taken out if I could, just to save a few bucks.

USB-C… I *really* hate that I can no longer use my thunderbolt display as a charger. This, to me, is the only really annoying “feature” of the new MacBook, but I still think it was the right thing to do. My old macbook would get quite hot around the magnetic charging port, so I suspect the connection wasn’t all that fantastic, even if it was a great concept, and having a single shared port for everything just seems like the right thing to do. The world probably isn’t ready for it, but it wasn’t ready for tablets either, and that changed quick enough. Having an adapter for my other legacy peripherals is a minor moan, that I suspect will go away faster than I can possibly imagine. So thumbs up for paving the way for a world of fewer odd cables, though I wish I could have bought a mag-charger adapter.

I now have 512 GB of SSD which should keep me running for the next 4 years (my 256 GB disk has been almost full for 4 years, so it’s not like my needs are skyrocketing, but every new project I start do tend to add a handfull of GB, so it’s nice to not have to worry about it).

The new graphics card is fast. The old one was never a problem, but I can quite clearly feel the difference when working in Unity – projects that were just barely managing 60FPS in the editor are now keeping an almost constant 100FPS. I know this isn’t a cardinal point for everyone, but for me, this upgrade alone was worth the price.

It’s pretty much the same CPU I had in the old one, though the memory is faster, so I suspect that a benchmark would tell me that everything runs a tad quicker, but probably not enough to have any real-world impact on my daily use. If I consider a typical day in my life, the amount of time I spend waiting for CPU bottlenecks is probably less than the time I spend getting coffee (Yes, I drink too much coffee, but that can’t be helped). I much prefer a smaller, cooler, less battery hungry machine over one that is 15% faster in those few cases where it actually matters, so CPU was never a concern for me.

Finally, I love that I can have it in space grey 🙂

Ups and Downs of Apple TestFlight

UPDATE #4 (20151109):

Everything was fine for a while – Even the build-in “Upload to  App Store…” button in XCode worked as it should, but nothing good lasts forever. Recently I’ve been getting the “You’re not authorised to use this service” error, which is of course complete bullocks and not at all related to whatever the problem is.

I know this because using the Export function and uploading the same IPA with the same credentials with Application Loader works most of the time. The only time it doesn’t work is when XCode decides I’m not authorised and refuses to upload, but still feels a need to lock some file that Application Loader also needs access to. Brilliant…

How is this in any way a difficult problem to fix? Take this crazy idea for example: How about letting the “Upload to App Store…” button do an export and pass the file to the Application Loader.

Yeah I know, crazy idea…

UPDATE #3 (20150902):

Out of necessity I recently re-visited external testing in test flight and found that the review process has been greatly simplified with most (if not all?) of the silly requirements removed and time-to-review down to around a day or so.

Bottom line is that this post is no longer relevant – Apple has stepped up and fixed TestFlight. I could still wonder why it took so long, but I’m just happy it works 🙂

UPDATE #2 (20150604):

That’s a funny coincidence.

Last night all my pre-releases suddenly disappeared from iTunes for a couple of hours, and today – all the builds I’ve done have automatically become active without me having to hit “save” on iTunes Connect.

I’d like to think that Apple actually listens to small developers and that my little rant paid of, but chances are I’m not the only one who complained and it just happened that they fixed it a few days after I finally had enough :).

Anyway: thanks 🙂 Now for the second half of the problem …

UPDATE #1 (20150601):

As I am (once again) waiting for iTunes to update, I was thinking of this post and it occurred to me that this whole sorry mess could be fixed with just 2 simple changes:

1. Provide a way to enter the “what to test” text before uploading so I don’t have to log in to iTunes Connect.
2. Remove the ridiculous review process from the test accounts.

Come on Apple, do something good for your developers for a change?

ORIGINAL POST:

I really don’t have time for this, but seeing as I am waiting for TestFlight anyway, this seems like a good time to vent some frustration.

Last year, Apple acquired TestFlight – the awesome service that let you drop new builds in the hands of your clients in minutes with very little fuzz, once it was set up. Which was a bit of a pain (not because of TestFlight, mind you, but because of Apple).

Here is how that worked in my daily life:

  1. Build App
  2. Wait for XCode to complete
  3. TestFlight immediately discovered the new build and opened its wizard
  4. Fill in information, select team, hit publish, done. One single flow with no waiting (Uploading was done in the background)

So… Apple acquired TestFlight and you would think that they would fix the one issue with TF: the setup. Admittedly, setup is marginally simpler than before but even this they’ve manage to screw up. Really it’s just stupid in a different way, but I’ll get back to that later.

In any case, it’s not that big a deal. I go through the setup phase once for each client so I can live with a few bumps on that road. Builds, however, is a different story. Some days I do 5-10 builds for a single client while ironing out odd issues or testing various designs, so the build process is what matters to me.

Here is what that looks like today:

  1. Build
  2. Wait
  3. Open Organizer
  4. Hit Submit To AppStore
  5. Wait
  6. Pick team
  7. Hit Submit (again)
  8. Wait (a very very long time)
  9. Hit Done (Except, we’re not, are we?)
  10. Open iTunes Connect in a browser
  11. Log in
  12. Navigate to the app MyApps->App->Prerelease
  13. Keep hitting “refresh” while you:
  14. WAIT some more for the state to go from Uploaded to Processing to finally appear in the list
  15. Click the build and put in some random crap in the “what to test” field
  16. Save
  17. Hurray, it’s live

You may think I’m making this up, but trust me, this is ridiculous far beyond my imagination.

So let’s return to the registration process. Before you had a limit of 100 registered devices per developer account and you had to manually register each of them, and make sure you cleared them out yearly so you wouldn’t run out. This was annoying, but I could cope, and though I’ve been in the 60+ registered devices range, the limit was never critical.

Now you can invite testers by e-mail rather than device ID which is great. Awesome. The problem: You can have 25 of these.

Twenty-five. It’s so absurd I have no words for it. I’m a small-time developer, and I’m going to break that limit almost immediately. The only way for me to not run into problems is by telling my clients that they have to limit who can test their apps. That’s a great way to start a relationship with a client.

And the fun doesn’t stop there either. These 25 people are “internal testers” which means they must be created as iTunes Connect users with access to my developer account. Why Apple? Seriously?

Now, Apple’s “big news” when they revealed this new (broken) TestFlight was that you could now have 1000 beta testers of your app, which should have sounded too good to be true. Because it is.

In order to get your app to any of these “external” testers, it must first be reviewed and approved, which means creating icons, asset and descriptions and making sure it does not tickle Apple in any inappropriate manner. Setting all this up takes hours, time that I have to bill my clients. All for an app that is just a prototype for a very limited audience.

It’s most fascinating how a company with unlimited financial resources, massive development teams and complete control over all the involved software (and hardware) components can spend almost a year to reduce a perfectly working service to a stinking pile of pooh.

Well done…

SOAK! Released in Denmark and Sweden

Finally…

EveryplayBanner

The first public release of SOAK! is now available in the Danish and Swedish app stores – we decided to do a limited rollout so that we could get a chance to test basic features like app store and GameCenter integration in the wild, but still on a small scale. The rest of the world follows shortly and then comes the Android port.

We actually launched the game almost a month ago but immediately found a critical bug that hadn’t shown up in sandbox mode and had to pull it, so, once bitten 🙂

Please do let us know what you think – we have a huge list of things we’re going to add over the coming months, but more of that in due time, now we need to start spreading the word…

https://itunes.apple.com/dk/app/soak/id876006257?l=sv&ls=1&mt=8

XCode6 and the mystery of the missing rect

Since “upgrading” to XCode6 I’ve had issues with storyboard based UI’s that used to work, but now suddenly starts acting really weird – table cells that scale to double size of the containing table, and buttons that don’t react to input because their parent (non clipping) views are cut in half.

Everything looks fine and dandy in XCode, but breaks horribly when running the app on both iOS7 and 8 and both on real devices and in the simulator.

In two cases I tracked the problem down to missing or incorrectly updated view rects – if you open the storyboard in a text editor (or rename it to .xml and open it in XCode or, even better, AppCode) and locate the view that’s causing problems, there’s a good chance you’ll see a rect with incorrect dimensions or simply won’t see any rects at all.

The structure should be:

...
 <view key="view" contentMode="scaleToFill" id="xMb-eO-e2t">
 <rect key="frame" x="0.0" y="0.0" width="320" height="504"/>
 <autoresizingMask key="autoresizingMask" flexibleMaxX="YES" flexibleMaxY="YES"/>
 <subviews>
...

I.e. a <rect> before the <autoresizemask>.

Correcting or inserting the rect fixes the problems and should be ok as long as you don’t do something silly like, say, edit the storyboard in XCode6’s storyboard editor.

But it’s ok. I understand that Apple, with its limited resources, can’t fix every single little insignificant corner-case in XCode when they need to focus their energy on inventing important new and productive programming languages. //sarcasm…

May I just humbly suggest that they got their priorities ass-backwards?

Smart Watch, Dumb Phone?

So, it seems Apple is finally ready to launch the iWatch – I am probably in the market for one, if for no other reason than the geek factor of it, though I can’t help but think that these things keep doing it ass-backwards.

How is it a “Smart” watch if it is nothing but a dumb accessory to the “Smart” phone. For it to be “Smart” it has to be the other way around. The watch has to be the phone, and the phone has to be the accessory, otherwise it’s just one more gadget I need to remember.

Done right, the Smart Watch is a communication hub with just enough functionality for me to be able to leave the increasingly large, clumsy, battery eating brick of a phone at home.

It should:

  1. Tell the time (wee 🙂 )
  2. Do phone calls
  3. Play music
  4. Check/browse e-mail
  5. Follow twitter and other social media updates
  6. Show my calendar
  7. Maybe even basic GPS / location tracking
  8. Possibly simple home automation
  9. (Oh, and the health thing, for those who find that useful 🙂 )

It does not require a massive CPU or GPU or a big screen but it has to have 3G and WiFi. When I want to play a game, browse the internet, write emails and need the screen real-estate, horsepower, or a semi-proper keyboard, I can turn my watch into a hotspot and connect my (3G-less) “phone” to it.

In fact, I’d probably just connect my laptop or my iPad to the watch and ditch the phone entirely. I don’t really need the phone – it’s a silly device – too small to be useful and too big to drag around all the time.

Why you should NOT advertise on LinkedIn!

I got a “promotional offer” from LinkedIn the other day: 300SEK of free advertising.

I’ve been wanting to promote TempusCura for a while, and figured LinkedIn would be a good place to do so, so as a test I created a campaign, provided my CreditCard details (because as always, free promotions are not really free, there was an administration fee of SEK 35:-.), and set it to run immediately.

I deliberately set the burn-rate high at 300SEK/day because I wanted to test if there was a noticeable difference in downloads or web-site traffic, and opted for impressions rather than clicks to get the widest exposure. As I said, it was a test.

I then went on to do other stuff…

This morning, and by complete accident, I ended up looking at the Campaign page of LinkedIn again, and to my horror saw that my little test had burned not 300, but 3.000SEK in two days. Unlike Facebook and Google, the amount you pay for the campaign is not limited (at least not by default) – if I had not seen this today, it would easily have been 30K by the time I was invoiced. Not cool!

So what did I get for my money?

213.000 impressions.

Wow!

Well, wow-not, much as expected, web adverts are not worth the price of the paper they’re not printed on… 213.000 impressions resulted in 20 clicks – probably from existing customers, friends or family.

It’s hard not to laugh. That’s SEK 150:- per click to advertise an app that I sell for SEK 7:-

I don’t know why there are so many ads on web-sites? I mean, I can certainly see it from the point of view of those who get paid, but for the advertisers? Seriously?

My advice: Draw your ad on the back of a piece of toilet paper, wipe, flush, and save a lot of money.

Tempus Cura Released

iTunesArtworkTempusCura was released on the App Store today.

I’m deliberately not promoting the release yet as I’d like to see the first 100 users or so trickle in slowly in case there are any issues that needs to be resolved, but I’m looking forward to any and all feedback, so please do drop me a note if you decide to try it out :).

The app itself is a free download and comes with 30 days of free subscription, so for the first month you can create or joins as many projects as you like. When the subscription expires you can either extend it at the early introductory price of $1/month or $10/year in which case the app remains unrestricted.

Alternatively you may purchase seat licenses for each of the projects you intend to use at the similarly discounted rate of $1/project. This in turn unlocks those projects once and for all, with no need to ever purchase another license or subscription to access that project again.