SynapseIndia dotnet debugging development process

Outline
 Teams and Processes
 Select Languages
 Debugging
 Misc (as time allows)
 AI
 Multiplayer

Introduction
 Used to be programmers created games
 But many great programmers not great game makers
 With budget shift, emphasis has shifted
 Game content creators are artist and designers
 Programmers can be thought of as providing
services for content
 But fate of entire game rests in their hands
Based on Chapter 3.1, Introduction to Game Development

Programming Areas – Game Code
 Everything directly related to game itself
 How camera behaves, score is kept, AI for bots, etc.
 Often in scripting language (rest is in C++,
more on languages next)
 Produce faster iterations
 Allow technical designers/artists to change
behaviors
 More appropriate language for domain (ex: AI
probably not easiest in C++)

Programming Areas – Game Engine
 Support code that is not game specific
 More than just drawing pretty 3d graphics (that is
actually the graphics engine, part of the game
engine)
 Isolate game code from hardware
 ex: controller, graphics, sound
 Allows designers to concentrate on game
 Common functionality needed across game
 Serialization, network communication, pathfinding,
collision detection

Programming Areas – Tools
 Most involve content creation
 Level editors, particle effect editors, sound editors
 Some to automate repetitive tasks (ex: convert content to
game format)
 These usually have no GUI
 Sometimes written as plug-ins for off-the-shelf tools
 Ex: extensions to Maya or 3dStudio or Photoshop
 If no such extension available, build from scratch

Programming Team Organization
 Programmers often specialize
 Graphics, networking, AI
 May be generalists, know something about everything
 Often critical for “glue” to hold specialists together
 Make great lead programmers
 More than 3 or 4, need some organization
 Often lead programmer, much time devoted to management
 More than 10 programmers, several leads (graphics lead,
AI lead, etc.)

Software Methodologies
 Code and Fix
 Waterfall
 Iterative
 Agile
 (Take cs3733, Software Engineering)

Methodologies – Code and Fix
 Really, lack of a methodology
 And all too common
 Little or no planning, diving straight into implementation
 Reactive, no proactive
 End with bugs. If bugs faster than can fix, “death spiral”
and may be cancelled
 Even those that make it, must have “crunch time”
 viewed after as badge of honor, but results in burnout

Methodologies - Waterfall
 Plan ahead
 Proceed through various planning steps before
implementation
 requirements analysis, design, implementation,
testing (validation), integration, and maintenance
 The waterfall loops back as fixes required
 Can be brittle to changing functionality,
unexpected problems in implementation
 Going back to beginning

Methodologies - Iterative
 Develop for a period of time (1-2 months), get
working game, add features
 Periods can coincide with publisher milestones
 Allows for some planning
 Time period can have design before implementation
 Allows for some flexibility
 Can adjust (to new technical challenges or producer
demands)

Methodologies - Agile
 Admit things will change, avoid looking too far
in the future
 Value simplicity and the ability to change
 Can scale, add new features, adjust
 Relatively new for game development
 Big challenge is hard to convince publishers

Common Practices – Version Control
 Database containing files and past history of
them
 Central location for all code
 Allows team to work on related files without
overwriting each other’s work
 History preserved to track down errors
 Branching and merging for platform specific
parts

Common Practices – Quality (1 of 2)
 Code reviews – walk through code by other
programmer(s)
 Formal or informal
 “Two eyes are better than one”
 Value is programmer aware others read
 Asserts
 Force program to crash to help debugging
 Ex: Check condition is true at top of code, say pointer
not NULL before following
 Removed during release

Common Practices – Quality (2 of 2)
 Unit tests
 Low level test of part of game (Ex: see if physics computations correct)
 Tough to wait until very end and see if bug
 Often automated, computer runs through combinations
 Verify before assembling
 Acceptance tests
 Verify high-level functionality working correctly (Ex: see if levels load
correctly)
 Note, above are programming tests (ie- code, technical). Still turned over to
testers that track bugs, do gameplay testing.
 Bug database
 Document and track bugs
 Can be from programmers, publishers, customers
 Classify by severity
 Keeps bugs from falling through cracks
 Helps see how game is progressing

Outline
 Teams and Processes (done)
 Select Languages (next)
 Debugging
 AI
 Multiplayer

C++ (1 of 3)
 Mid-late 1990’s, C was language of choice
 Since then, C++ language of choice for games
 First commercial release in 1985 (AT&T)
 List pros (+) and cons (-)
 (Take cs2102 OO Design Concepts or cs4233 OOAD)
+ C Heritage
 Learning curve easier
 Compilers wicked fast
+ Performance
 Used to be most important, but less so (but still for core parts)
 Maps closely to hardware (can “guess” what assembly
instructions will be)
 Can not use features to avoid cost, if want (ie- virtual function
have extra step but don’t have to use)
 Memory management controlled by user

C++ (2 of 3)
+ High-level
 Classes (objects), polymorphism, templates, exceptions
 Especially important as code-bases enlarge
 Strongly-typed (helps reduce errors)
 ex: declare before use, and const
+ Libraries
 C++ middleware readily available
 OpenGL, DirectX, Standard Template Library (containers,
like “vectors”, and algorithms, like “sort”)

C++ (3 of 3)
- Too Low-level
 Still force programmer to deal with low-level issues
 ex: memory management, pointers
- Too complicated
 Years of expertise required to master (other languages seek to
overcome, like Java and C#)
- Lacking features
 No built-in way to look at object instances
 No built-in way to serialize
 Forces programmer to build such functionality (or learn custom or
3rd party library)
- Slow iteration
 Brittle, hard to try new things
 Code change can take a looong time as can compile

C++ (Summary)
 When to use?
 Any code where performance is crucial
 Used to be all, now game engine such as graphics and
AI
 Game-specific code often not C++
 Legacy code base, expertise
 When also use middle-ware libraries in C++
 When not to use?
 Tool building (GUI’s tough)
 High-level game tasks (technical designers)

Java (1 of 3)
 Java popular, but only recently so for games
 Invented in 1990 by Sun Microsystems
+ Concepts from C++ (objects, classes)
 Powerful abstractions
+ Cleaner language
 Memory management built-in
 Templates not as messy
 Object functions, such as virtualization
+ Code portability (JVM)
(Hey, draw picture)
+ Libraries with full-functionality built-in

Java (2 of 3)
- Performance
 Interpreted, garbage collection, security
 So take 4x to 10x hit
+ Can overcome with JIT compiler, Java Native
Interface (not interpreted)
- Platforms
 JVM, yeah, but not all games (most PC games not,
nor consoles)
+ Strong for browser-games, mobile

Java (3 of 3)
 Used in:
 Downloadable/Casual games
 PopCap games
 Mummy Maze, Seven Seas, Diamond Mine
 Yahoo online games (WorldWinner)
 Poker, Blackjack
 PC
 Star Wars Galaxies uses Java (and simplified Java for
scripting language)
 You Don’t Know Jack and Who Wants to be a Millionaire all
Java

Scripting Languages (1 of 3)
 Not compiled, rather specify (script) sequence of actions
 Most games rely upon some
 Trigger a few events, control cinematic
 Others games may use it lots more
 Control game logic and behavior (Game Maker has GML)
+ Ease of development
 Low-level things taken care of
 Fewer errors by programmer
- But script errors tougher, often debuggers worse
 Less technical programming required
 Still, most scripting done by programmers
 Iteration time faster (don’t need to re-compile all code)
 Can be customized for game (ex: just AI tasks)

+ Code as an asset
 Ex: consider Peon in C++, with behavior in C++, maybe art as an asset.
Script would allow for behavior to be an asset also
 Can be easily modified, even by end-user in “mod”
- Performance
 Parsed and executed “on the fly”
 Hit could be 10x or more over C++
 Less efficient use of instructions, memory management
-Tool support
 Not as many debuggers, IDEs
 Errors harder to catch
- Interface with rest of game
 Core in C++, must “export” interface
 Can be limiting way interact
 (Hey, draw picture)

 Python
 Interpreted, OO, many libraries, many tools
 Quite large (bad when memory constrained)
 Ex: Blade of Darkness, Earth and Beyond, Eve Online,
Civilization 4 (Table 3.2.1 full list)
 Lua (pronounced: Loo-ah)
 Not OO, but small (memory). Embed in other programs. Doesn’t
scale well.
 Ex: Grim Fandango, Baldur’s Gate, Far Cry (Table 3.2.2 full
list)
 Others:
 Ruby, Perl, JavaScript
 Custom: GML, QuakeC, UnrealScript
 Implementing own tough, often performs poorly so careful!

Macromedia Flash (1 of 2)
 More of a platform and IDE (ala Game Maker) than a language (still,
has ActionScript)
 “Flash” refers authoring environment, the player, or the
application files
 Released 1997, popular with Browser bundles by 2000
 Advantages
 Wide audience (nearly all platforms have Flash player)
 Easy deployment (embed in Web page)
 Rapid development (small learning curve, for both artists and
programmers)
 Disadvantages
 3D games
 Performance (interpreted, etc.)

Macromedia
Flash (2 of 2)
 Timeline Based
 Frames and Frame rate (like animations)
 Programmers indicate when (time) event occurs (can occur
across many frames)
 Vector Engine
 Lines, vertices, circles
 Can be scaled to any size, still looks crisp
 Scripting
 ActionScript similar to JavaScript
 Classes (as of Flash v2.0)
 Backend connectivity (load other Movies, URLs)

Outline
 Select Languages (done)
 Debugging (next)
 AI
 Multiplayer

Debugging Introduction
 New Integrated Development Environments (IDEs) have
debugging tools
 Trace code, print values, profile
 But debugging frustrating
 Beginners not know how to proceed
 Even advanced can get “stuck”
 Don’t know how long takes to find
 Variance can be high
 Mini-outline
 5-step debugging process
 Debugging tips
 Touch scenarios and patterns
 Prevention

Step 1: Reproduce the Problem
Consistently
 Find case where always occurs
 “Sometimes game crashes after kill boss” doesn’t
help much
 Identify steps to get to bug
 Ex: start single player, skirmish map 44, find enemy
camp, use projectile weapon …
 Produces systematic way to reproduce

Step 2: Collect Clues
 Collect clues as to bug
 But beware that some clues are false
 Ex: if bug follows explosion may think they are related, but
may be from something else
 Ex: if crash using projectile, what about that code that
makes it possible to crash?
 Don’t spend too long, get in and observe
 Ex: see reference pointer from arrow to unit that shot arrow
should get experience points, but it is may be NULL
 That’s the bug, but why is it NULL?

Step 3: Pinpoint Error
 Propose a hypothesis and prove or disprove
 Ex: suppose arrow pointer corrupted during flight. Add code to
print out values of arrow in air. But equals same value that
crashes. Wrong.
 Ex: suppose unit deleted before experience point. Print out values
of all in camp before fire and all deleted. Yep, that’s it.
 Or, divide-and-conquer method (note, can use in
conjunction with hypo-test above, too)
 Sherlock Holmes “when you have eliminated the impossible,
whatever remains, however improbably, must be the truth”
 Setting breakpoints, look at all values, until discover bug
 The “divide” part means break it into smaller sections
 Ex: if crash, put breakpoint ½ way. Is it before or after? Repeat
 Look for anomalies, NULL or NAN values

Step 4: Repair the Problem
 Propose solution. Exact solution depends upon stage of
problem.
 Ex: late in code cannot change data structures. Too many
other parts use.
 Worry about “ripple” effects.
 Ideally, want original coder to fix. At least, talk with
original coder for insights.
 Consider other similar cases, even if not yet reported
 Ex: other projectiles may cause same problem as arrows did

Step 5: Test Solution
 Obvious, but can be overlooked if programmer
is sure they have fix (but programmer can be
wrong!)
 So, test that fix repairs bug
 Best by independent tester
 Test if other bugs introduced (beware “ripple”
effect)

Debugging Tips (1 of 3)
 Question your assumptions – don’t even assume
simple stuff works, or “mature” products
 Ex: libraries can have bugs
 Minimize interactions – systems can interfere,
make slower so isolate the bug to avoid
complications
 Minimize randomness – ex, can be caused by
random seed or player input. Fix input (script
player) so reproducible

 Break complex calculations into steps – may be
equation that is fault or “cast” badly
 Check boundary conditions – classic “off by
one” for loops, etc.
 Disrupt parallel computations – “race
conditions” if happen at same time (cs3013)
 Use debugger – breakpoints, memory watches,
stack …
 Check code recently changed – if bug appears,
may be in latest code (not even yours!)

 Take a break – too close, can’t see it. Remove
to provide fresh prospective
 Explain bug to someone else – helps retrace
steps, and others provide alternate hypotheses
 Debug with partner – provides new techniques
 Get outside help – tech support for consoles,
libraries, …

Tough Debugging Scenarios and Patterns
(1 of 2)
 Bug in Release but not in Debug
 Often in initialized code
 Or in optimized code
 Turn on optimizations one-by-one
 Bug in Hardware but not in Dev Kit
 Usually dev kit has extra memory (for tracing, etc.).
Suggest memory problem (pointers), stack overflow, not
checking memory allocation
 Bug Disappears when Changing Something Innocuous
 Likely timing problem (race condition) or memory problem
 Even if looks like gone, probably just moved. So keep
looking

Tough Debugging Scenarios and Patterns
(2 of 2)
 Truly Intermittent Problems
 Maybe best you can do is grab all data values (and stack,
etc) and look at (“Send Error Report”)
 Unexplainable Behavior
 Ex: values change without touching. Usually memory
problem. Could be from supporting system. Retry, rebuild,
reboot, re-install.
 Bug in Someone Else’s Code
 “No it is not”. Be persistent with own code first.
 It’s not in hardware. (Ok, very, very rarely, but expect it
not to be) Download latest firmware, drivers
 If really is, best bet is to help isolate to speed them in fixing
it.

Debugging Prevention (1 of 2)
 Understand underlying system
 Knowing language not enough
 Must understand underlying system
 At least one level down
 Engine for scripters
 OS for engine
 Maybe two layers down (hardware, assembly)
 Add infrastructure, tools to assist
 Make general
 Alter game variables on fly (speed up)
 Visual diagnostics (maybe on avatars)
 Log data (events, units, code, time stamps)
 Record and playback capability

Debugging Prevention (2 of 2)
 Set compiler on highest level warnings
 Don’t ignore warnings
 Compile with multiple compilers
 See if platform specific
 Write own memory manager (for console games,
especially, since tools worse)
 Use asserts
 Always initialize when declared
 Indent code, use comments
 Use consistent style, variable names
 Avoid identical code – harder to fix if bug
 Avoid hard-coded (magic numbers) – makes brittle
 Verify coverage (test all code) when testing

Outline
 Select Languages (done)
 Debugging (done)
 AI (next)
 Multiplayer

Introduction to AI
 Opponents that are challenging, or allies that are
helpful
 Unit that is credited with acting on own
 Human-level intelligence too hard
 But under narrow circumstances can do pretty well
(ex: chess and Deep Blue)
 Artificial Intelligence (around in CS for some
time)

AI for CS different than AI for Games
 Must be smart, but purposely flawed
 Loose in a fun, challenging way
 No unintended weaknesses
 No “golden path” to defeat
 Must not look dumb
 Must perform in real time (CPU)
 Configurable by designers
 Not hard coded by programmer
 “Amount” and type of AI for game can vary
 RTS needs global strategy, FPS needs modeling of
individual units at “footstep” level
 RTS most demanding: 3 full-time AI programmers
 Puzzle, street fighting: 1 part-time AI programmer

AI for Games – Mini Outline
 Introduction (done)
 Agents (next)
 Finite State Machines
 Common AI Techniques
 Promising AI Techniques

Game Agents (1 of 2)
 Most AI focuses around game agent
 think of agent as NPC, enemy, ally or neutral
 Loops through: sense-think-act cycle
 Acting is event specific, so talk about sense+think
 Sensing
 Gather current world state: barriers, opponents, objects
 Needs limitations : avoid “cheating” by looking at game
data
 Typically, same constraints as player (vision, hearing
range)
 Often done simply by distance direction (not computed as per
actual vision)
 Model communication (data to other agents) and reaction
times (can build in delay)

Game Agents (2 of 2)
 Thinking
 Evaluate information and make decision
 As simple or elaborate as required
 Two ways:
 Precoded expert knowledge, typically hand-crafted if-
then rules + randomness to make unpredictable
 Search algorithm for best (optimal) solution

Game Agents – Thinking (1 of 3)
 Expert Knowledge
 finite state machines, decision trees, … (FSM most popular,
details next)
 Appealing since simple, natural, embodies common sense
 Ex: if you see enemy weaker than you, attack. If you see
enemy stronger, then go get help
 Often quite adequate for many AI tasks
 Trouble is, often does not scale
 Complex situations have many factors
 Add more rules, becomes brittle

 Search
 Look ahead and see what move to do next
 Ex: piece on game board, pathfinding (ch 5.4)
 Machine learning
 Evaluate past actions, use for future
 Techniques show promise, but typically too slow
 Need to learn and remember

 Making agents stupid
 Many cases, easy to make agents dominate
 Ex: bot always gets head-shot
 Dumb down by giving “human” conditions, longer reaction
times, make unnecessarily vulnerable
 Agent cheating
 Ideally, don’t have unfair advantage (such as more
attributes or more knowledge)
 But sometimes might to make a challenge
 Remember, that’s the goal, AI lose in challenging way
 Best to let player know

 Agents (done)
 Finite State Machines (next)
 Common AI Techniques

Finite State Machines (1 of 2)
 Abstract model of computation
 Formally:
 Set of states
 A starting state
 An input vocabulary
 A transition function that maps inputs and the current state
to a next state
Wander Attack
Flee
SeeEnemy
LowHealth
NoEnemy
NoEnemy

Finite State Machines (2 of 2)
 Most common game AI software pattern
 Natural correspondence between states and behaviors
 Easy to diagram
 Easy to program
 Easy to debug
 Completely general to any problem
 Problems
 Explosion of states
 Often created with ad hoc structure

Finite-State Machine: Approaches
 Three approaches
 Hardcoded (switch statement)
 Scripted
 Hybrid Approach

Finite-State Machine:
Hardcoded FSM
void RunLogic( int * state ) {
switch( state )
{
case 0: //Wander
Wander();
if( SeeEnemy() ) { *state = 1; }
break;
case 1: //Attack
Attack();
if( LowOnHealth() ) { *state = 2; }
if( NoEnemy() ) { *state = 0; }
break;
case 2: //Flee
Flee();
if( NoEnemy() ) { *state = 0; }
break;
}
}

Problems with switch FSM
1. Code is ad hoc
 Language doesn’t enforce structure
2. Transitions result from polling
 Inefficient – event-driven sometimes better
3. Can’t determine 1st time state is entered
4. Can’t be edited or specified by game designers
or players

Scripted with alternative language
AgentFSM
{
State( STATE_Wander )
OnUpdate
Execute( Wander )
if( SeeEnemy ) SetState( STATE_Attack )
OnEvent( AttackedByEnemy )
SetState( Attack )
State( STATE_Attack )
OnEnter
Execute( PrepareWeapon )
OnUpdate
Execute( Attack )
if( LowOnHealth ) SetState( STATE_Flee )
if( NoEnemy ) SetState( STATE_Wander )
OnExit
Execute( StoreWeapon )
State( STATE_Flee )
OnUpdate
Execute( Flee )
if( NoEnemy ) SetState( STATE_Wander )
}

Scripting Advantages
1. Structure enforced
2. Events can be handed as well as polling
3. OnEnter and OnExit concept exists
4. Can be authored by game designers
 Easier learning curve than straight C/C++

Scripting Disadvantages
 Not trivial to implement
 Several months of development
 Custom compiler
 With good compile-time error feedback
 Bytecode interpreter
 With good debugging hooks and support
 Scripting languages often disliked by users
 Can never approach polish and robustness of commercial
compilers/debuggers

Hybrid Approach
 Use a class and C-style macros to approximate a scripting language
 Allows FSM to be written completely in C++ leveraging existing
compiler/debugger
 Capture important features/extensions
 OnEnter, OnExit
 Timers
 Handle events
 Consistent regulated structure
 Ability to log history
 Modular, flexible, stack-based
 Multiple FSMs, Concurrent FSMs
 Can’t be edited by designers or players

Extensions
 Many possible extensions to basic FSM
 OnEnter, OnExit
 Timers
 Global state, substates
 Stack-Based (states or entire FSMs)
 Multiple concurrent FSMs
 Messaging

 Agents (done)
 Finite State Machines (done)
 Common AI Techniques (next)

Common Game AI Techniques
 Whirlwind tour of common techniques
 (See book chapters)

SynapseIndia dotnet debugging development process

More Related Content

Similar to SynapseIndia dotnet debugging development process (20)

More from Synapseindiappsdevelopment (20)

Recently uploaded (20)

SynapseIndia dotnet debugging development process