Getting Soar

Where are the latest Soar releases?

Check the downloads page for pointers to downloads and release notes. Alternatively you can check out the development trunk from git. See the build documents for more information.

Running Soar

How do I run Soar?

If you downloaded a Soar release, navigate inside the extracted archive and run the shell scripts or batch files to run the various Soar components. The scripts set essential environment variables so the different Soar components can find their libraries and resources.

Java: What JREs/JDKs can I use?

Use Sun JDK 6+. Other versions and vendors may work but we do not support them.

Using the Soar Java Debugger

What command line options does the debugger accept?

Command line options:
  • -remote => use a remote connection (with default ip and port values)
  • -ip xxx => use this IP value (implies remote connection)
  • -port ppp => use this port (implies remote connection)
Without any remote options we start a local kernel
  • -agent => on a remote connection select this agent as initial agent
  • -agent => on a local connection use this as the name of the initial agent
  • -source "" => load this file of productions on launch (only valid for local kernel)
  • -listen port => use this port to listen for remote connections (only valid for a local kernel)
  • -maximize => start with maximized window
  • -width => start with this window width
  • -height => start with this window height
  • -x -y => start with this window position (Providing width/height/x/y => not a maximized window)
If I run the debugger for a while, it starts to slow down, stutter, and then crashes, or runs out of memory

The problem here is that the trace window in the debugger is using more and more memory over time, since it doesn't get rid of old stuff until you clear the window. Especially at watch 5, there's a lot of text being stored (even if it's mostly hidden). Long before your OS runs out of memory, though, Java runs out of heap space. You can allocate more heap space to most Java implementations using the -Xmxm flag when executing Java. For example:

java -Xmx512m -jar SoarJavaDebugger.jar

This will allocate 512 megs of memory for Java's heap (the default is 64 megs), and should allow you to run the debugger significantly longer.

How do I use the debugger with the graphical demos like JavaTOH and JavaMissionaries? (or any other Soar application)?
  • Start the application
  • Start the debugger (in Linux the debugger must be started after the application--order doesn't matter in Windows)
  • In the debugger, on the top menuBar, pull down the "Kernel" selection and choose "Connect to Remote Soar..."
  • In the popup window "Would you like to shutdown the local kernel now" enter "OK"
  • In the next popup window, if the application is on your local machine, press "OK" to use the default settings. If your application is running on another machine, enter the IP addr and press "OK"
  • If the connection succeeds, then you can use the debugger and the application interchangeably to control the agent. If the connection fails, then the application either is not properly configured for SML, or no agent currently exists in the application.
How can I copy/paste a production into the debugger?

There are several ways to do this:
  • You can paste a production into the trace window.
  • You can use the "edit_production" window in the lower-right corner. If you supply the name of an existing production, it will fill in the window with that production. You can then edit it and load the new version using the "Load Production" button.
  • If your production is in Visual Soar, you can do "Soar Runtime" -> "Connect" to connect to the debugger. Then, open the file with your production and do "Runtime" -> "Send Production" or "Send File" to load your production(s) into Soar.
Developing Soar

SCons is adding a lot of ridiculous include directories and the build fails

Seeing something like this means you should define JAVA_HOME:

g++ -o Core/CLI/CommandLineInterface.o -c -DSCONS -fvisibility=hidden -g3 -Wall -Werror -O3 -m64 -fPIC -ICore/CLI/src -ICore/CLI/include -ICore/SoarKernel/include -ICore/ElementXML/include -ICore/ConnectionSML/include -ICore/KernelSML/include -ICore/shared -I/usr/include -I/usr/include/netatalk -I/usr/include/netinet -I/usr/include/protocols -I/usr/include/dbus-1.0 -I/usr/include/blkid -I/usr/include/netax25 -I/usr/include/nfs -I/usr/include/rdma -I/usr/include/netpacket -I/usr/include/X11 -I/usr/include/c++ -I/usr/include/neteconet -I/usr/include/xen -I/usr/include/rpc -I/usr/include/gnu -I/usr/include/netash -I/usr/include/asm -I/usr/include/sound -I/usr/include/mtd -I/usr/include/asm-generic -I/usr/include/netrom -I/usr/include/compiz -I/usr/include/arpa -I/usr/include/net -I/usr/include/rpcsvc -I/usr/include/netiucv -I/usr/include/linux -I/usr/include/netrose -I/usr/include/video -I/usr/include/bits -I/usr/include/netipx -I/usr/include/uuid -I/usr/include/python2.5 -I/usr/include/sys -I/usr/include/python2.6

See the build document for more details.

How do I get the trace for the initial S1 creation?

The initial state is created right after the agent is created but before the agent pointer is passed back to the client. Therefore, if you create the client and then register for print (or xml) output, you do not ever see the initial S1 creation.

To get these initial print callbacks, you need to register for the after-agent-created event and register your print handlers in that function. This callback fires right after the agent is created but before S1 is created.

How can I look up a wme on the input link if I know its attribute?

Use the Indentifier::FindByAttribute method like this:

pAgent->GetInputLink()->FindByAttribute("location", 0);

This works for any identifier, not just at the top level of the input link.

How do I increase the performance of my SML application?

It is often desirable to maximize the performance of your SML application. This section assumes that you just want to make things as fast as possible after you have finished debugging your application. Debugging is an inherently slow process, so these tips will be less helpful while you’re still debugging.
  • Compile with optimizations turned on. In Visual Studio this means doing a release build. On Linux and OS X, the default settings are probably sufficient, but you can experiment with new settings if you want (let us know if you find better settings).
  • Put primary application and Soar in the same process. That is, use CreateInNewThread? or CreateInCurrentThread?, not CreateRemoteConnection?. Using a remote kernel means socket communication is used, which is slow.
  • Don’t register for unnecessary events. Every event that is registered for causes extra work to be done. Try to find an appropriate event to register for so you don’t end up getting more event calls than you actually need – that is, try to avoid registering for events which occur more frequently than you need and then filtering them on the application side.
  • Don’t connect the debugger. Connecting the debugger creates a remote connection and also registers for several events.
Set watch level 0. Even if you don’t have a client registered for any of the print or XML events, work is still done internally to generate some of the information that would have been sent out. Setting watch level 0 avoids this work.
  • Disable monitor productions. Again, even if no client is registered to print out the text of monitor productions, work is still done internally to prepare the text. Monitor productions can be disabled by excising them or commenting them out, but an easier method is to have each monitor production test a debug flag in working memory which is set by some initialization production or operator. Thus all of the monitor productions can be turned on or off by changing one line of code.
  • Disable timers. Soar uses timers internally to generate the output of the stats command. If you don’t need this information, you can use the “timers –off” command to disable this bookkeeping. This can make a significant difference in the watch 0 case.
  • Avoid running agents separately. Instead of calling RunSelf? or RunSelfTilOutput? on each agent, just call RunAllAgents? on the kernel itself. This runs all agents together and avoids the overhead of running them separately. The absolute best you can do is to call RunAllAgentsForever? as described in section 2.4 – this avoids repeatedly calling the run functions at all and will make it easier to stop and restart your application from the debugger (or other clients).
  • In the case where the absolute best performance under SML is desired, use CreateKernelInCurrentThread? instead of CreateKernelInNewThread? and set the “optimized” flag to true in the parameters passed to CreateKernelInCurrentThread?. This means Soar will execute in the same thread as your application. Without this each call to and from the Soar kernel requires a context switch (assuming a single processor machine). This method also eliminates the thread which polls for new events. This means you must poll for the events yourself by periodically calling CheckForIncommingCommands?, which is a little more work for the programmer.
  • Turn off autocommits. By default, SML sends WME changes to the kernel as soon as they are requested. Performance can be improved by telling SML to buffer all WME changes until an explicit call is made to commit the changes. Turning off autocommits is done via a call to mykernel->SetAutoCommit?(false). Explicitly committing WME changes is done via a call to myagent->Commit(). Sending all the changes at once will give a small performance boost in cases where Soar and the environment are in the same process, and a large performance boost when they are communicating over sockets. Be careful, though -- the agent won't see any wmes until they have been committed (and then not until the next input phase, as usual), and all WMEs must be committed before doing an init-soar, which generally means before giving the user control (since someone could call init-soar from an attached debugger). This typically means committing all WMEs after updating an agent's input-link.
When can I safely make changes to the input-link?

A Soar agent only receives input during the input phase and it does this through an input phase callback while the agent is running. SML allows the environment to change the input-link at other times and those changes are buffered until the next input phase. This means you have several options for handling input:
  • You can register for smlEVENT_BEFORE_INPUT_PHASE and make changes to the input link at that time. This is very close to the way the kernel naturally handles input but will often be relatively slow if Soar is running a lot faster than the environment is changing (a common situation) as this event needs to be sent each decision cycle for each agent, generating a lot of communications traffic.
  • Another option is to register for an update event (smlEVENT_AFTER_ALL_OUTPUT_PHASES and smlEVENT_AFTER_ALL_GENERATED_OUTPUT), check for output at that time and create new inputs immediately. These events are called after the output phase has completed and the new input link changes will be buffered until the next input phase. This is the most common choice in existing SML environments.
  • A third option is to register an output event handler, which looks for a particular attribute to be added to the output link and only then calls the registered function. This handler will be called during the output phase and again, new input will be collected and buffered until the next input phase of the agent.
  • In general, any run event (smlRunEventId) or update event (smlUpdateEventId) is a good candidate to use for changing the input-link. Other events may not be appropriate to use. One particular example is smlEVENT_BEFORE_AGENT_REINITIALIZED and smlEVENT_AFTER_AGENT_REINITIALIZED. These events fire during an init-soar call and are dangerous to use because the system is actively destroying the entire input link and then recreating it to match the last structure defined by the environment. This happens automatically and gives the agent its best chance to continue executing again. However, if you register for these events yourself and try to change the input link during those events, the resulting behavior is likely to depend on whether the system callback occurs before or after your callback. With care this can probably all work out correctly, but you should be aware of what's going on. Similarly, making changes to I/O in response to a production firing or other such events is potentially dangerous as you might be changing input within a given execution phase rather than before or after it, leading to unpredictable results. These events are best used for monitoring Soar's behavior rather than changing it.
What do I need to know about threads in my SML applications?

You generally want to keep all interactions with the kernel in a single thread. You can make calls from other threads, but they will block if Soar is busy doing something else--like running--until the command completes. This requirement raises a few issues:
  • If you're working with a GUI, then the handler for a "run button" shouldn't just call run on the kernel. If it does the GUI will wait for the run to complete before responding. This is usually not acceptable as requests to repaint the screen, other button presses etc. will be ignored. The solution is to create a new thread and run Soar in that thread.
  • Now that Soar is running on a separate thread, when the user presses a "stop" button if you try to call Stop on the kernel from the UI thread it'll block. What you need to do instead is set a flag within the thread that is running Soar and use an event to check whether that flag has been set and if so call Stop then. The smlEVENT_INTERRUPT_CHECK event is a good candidate for this as it fires infrequently and generates little overhead.
  • Similarly, as the environment changes you'll want to update the input link in the thread that is running Soar. The normal way to do this is again inside an event handler. The update family of events (smlEVENT_AFTER_ALL_OUTPUT_PHASES and smlEVENT_AFTER_ALL_GENERATED_OUTPUT) are good candidates to consider.
If all of this seems a bit confusing, take a look at the Java Towers of Hanoi example that's included in the release. It demonstrates all of this behavior and is a good model to follow. If you're working on a command line application without a GUI, then the TestCommandLine? application is a good reference as it demonstrates how to support interruption in a single-threaded application.
How do I properly manage memory in my SML application?

Memory management is actually really easy. Generally, the only objects you should explicitly delete are the kernel object and any objects you directly allocated through a call to new. In Java and Tcl, this generally means you can just let things go out of scope when you’re done with them. There are a couple special cases you should be aware of, though:
  • Agent objects are automatically deleted when the owning Kernel object is deleted (actually, when the call to Kernel::Shutdown is made, which you should always make before deleting the kernel). If you want to destroy an agent earlier, you can by making a call to Kernel:estroyAgent?. Under no circumstances should you delete (in the C++ sense) an Agent object.
  • In Java if you create a ClientXML object through "xml = new ClientXML()" you should call xml.delete() on it when you're done. This isn't strictly required (the garbage collector will get it eventually) but is good practice and will avoid messages about leaked memory when the application shuts down. As per the general rule, in C++ if you create it with new you’re responsible for destroying it with delete.
  • Since there can be multiple clients interacting with the same kernel and agents, your application needs to be listening for the appropriate events so if some other client deletes/destroys a kernel or agent your application is using, you don’t crash. Specifically, listen for the BEFORE_AGENT_DESTOYED and BEFORE_SHUTDOWN events so you can clean things up as needed in your application.
What are the differences in the SML interface between C++ and Java (and other languages)?

For the most part, the exact same classes and methods are available in all languages. In some cases there are differences with obvious mappings -- for example, in C++ a method might take a char whereas in Java it takes a String.
One exception to this rule is in the way event callbacks are handled. The callbacks look very similar from language to language, but some languages have constraints that make a direct conversion of the C++ style impossible. In C++, event callbacks are global functions. In Java, it's not possible to have global functions, so the callback must belong to some class. Thus, for Java we provide an interface for each event type that the class containing the callback method must implement, and then the handling object is passed in as part of the registration. You can look at any of the example Java code to see examples of this (e.g. TestJavaSML or JavaTOH are good places to start).

Another important difference is that in C++ and Tcl, the kernel object must be explicitly deleted after Shutdown(). In C++ this is just calling "delete mykernel;" whereas in Tcl it takes the form "$kernel -delete". In Java and C#, deletion is handled internally by the Shutdown() method (this isn't done in C++ to make debugging easier).

How should I get started on my first C++/Java SML application?
  • See the HelloWorld document for simple example environments
Start by building and running these and then replace the code with your code. Also don't forget to read the SMLQuickStartGuide.
The complete documentation for the SML interface is provided in the ClientSML C++ headers (which are thoughly commented). Using a different language -- don't worry, the interface is virtually the same.

How do I make a copy of an existing project?
  • Go into the SoarSuite\Tools folder and copy the TestClientSML folder as MyProject
  • Within that folder, rename TestClientSML.vcproj as MyProject.vcproj
  • Open the SoarSuite\SML.sln solution file (or make a copy of this file and open that)
  • Select "File | Add Project | Existing Project" and add the MyProject.vcproj to the solution
  • It will be added as "TestClientSML". Select it in Solution Explorer, right click and choose Rename to rename it as MyProject. (Make sure you're not accidentally renaming the real TestClientSML project--you can open the TestClientSML.cpp file in the editor and then move your mouse over the tab at the top of the editor window for the open file to see the path just to be sure).
  • At this point you should be able to build the project to create MyProject.exe
  • Remove TestClientSML.cpp from the project and start adding your own code.
What are the required Visual Studio project settings for my SML application?

Your project needs to reference the !ClientSML, !ConnectionSML, and !ElementXML projects. Additionally, add the include directories for the src folder of each of those projects.

The Java Virtual Machine segfaults when running Soar

See the next question.

SML WMElement/Identifier/StringElement/IntElement/FloatElement pointers become invalid without my knowledge

First, usage of WMElement in this answer refers to any of WMElement/Identifier/StringElement/IntElement/FloatElement (all of which are or extend WMElement).

A few things to remember about WMElement pointers returned by SML methods such as CreateIdentifier:
  • WMElement objects returned by pointers are owned by SML (don't delete them),
  • Pointers to WMElement objects may be destroyed without calling DestroyWME,
  • Pointers to WMElement objects may be destroyed by SML triggered by the agent as it runs.
A general rule: you need to keep track of whether or not a WMElement is valid before you use it. It will stay valid if
  • You do not call DestroyWME on it or on a WME that causes it to get disconnected,
  • You do not return control back to Soar letting it run.
If you call DestroyWME, you need to be sure that you don't hang on to any WMElement objects that the WME you destroyed was keeping valid. Most of the time your working memory structure is a tree so you only need to keep track of the children of any Identifier you destroy.

In Java/Python/Tcl/CSharp, WMElement ConvertToIdentifier or other Convert- function returns something that should be equal to something else but isn't

In our language interfaces covered by SWIG, such as Java, Python, Tcl, and CSharp, ConvertToIdentifier, ConvertToStringElement, ConvertToIntElement, and ConvertToFloatElement issue references to objects that wrap pointers. If these pointers were compared, they would be equal. Unfortunately, comparing references to these objects wrapping equivalent pointers returns false--that they are unequal.

A workaround is to not compare references, or use other language supported reference equality in this case. Instead, compare the time tags by calling GetTimeTag on the reference.