Manual: OTcl Linkage

From nsnam
Revision as of 01:45, 24 July 2008 by Lachlan (Talk | contribs) (Class TclCommand: 231 -> 2^31)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Previous chapter: "Undocumented Facilities" | index | Next chapter: "The Class Simulator"

ns is an object oriented simulator, written in C++, with an OTcl interpreter as a frontend. The simulator supports a class hierarchy in C++ (also called the compiled hierarchy in this document), and a similar class hierarchy within the OTcl interpreter (also called the interpreted hierarchy in this document). The two hierarchies are closely related to each other; from the user’s perspective, there is a one-to-one correspondence between a class in the interpreted hierarchy and one in the compiled hierarchy. The root of this hierarchy is the class TclObject. Users create new simulator objects through the interpreter; these objects are instantiated within the interpreter, and are closely mirrored by a corresponding object in the compiled hierarchy. The interpreted class hierarchy is automatically established through methods defined in the class TclClass. user instantiated objects are mirrored through methods defined in the class TclObject. There are other hierarchies in the C++ code and OTcl scripts; these other hierarchies are not mirrored in the manner of TclObject.

Concept Overview

Why two languages? ns uses two languages because simulator has two different kinds of things it needs to do. On one hand, detailed simulations of protocols requires a systems programming language which can efficiently manipulate bytes, packet headers, and implement algorithms that run over large data sets. For these tasks run-time speed is important and turn-around time (run simulation, find bug, fix bug, recompile, re-run) is less important.

On the other hand, a large part of network research involves slightly varying parameters or configurations, or quickly exploring a number of scenarios. In these cases, iteration time (change the model and re-run) is more important. Since configuration runs once (at the beginning of the simulation), run-time of this part of the task is less important.

ns meets both of these needs with two languages, C++ and OTcl. C++ is fast to run but slower to change, making it suitable for detailed protocol implementation. OTcl runs much slower but can be changed very quickly (and interactively), making it ideal for simulation configuration. ns (via tclcl) provides glue to make objects and variables appear on both langauges. For more information about the idea of scripting languages and split-language programming, see Ousterhout’s article in IEEE Computer [26]. For more information about split level programming for network simulation, see the ns paper [2].

Which language for what? Having two languages raises the question of which language should be used for what purpose.

Our basic advice is to use OTcl:

  • for configuration, setup, and “one-time” stuff
  • if you can do what you want by manipulating existing C++ objects

and use C++:

  • if you are doing anything that requires processing each packet of a flow
  • if you have to change the behavior of an existing C++ class in ways that weren’t anticipated

For example, links are OTcl objects that assemble delay, queueing, and possibly loss modules. If your experiment can be done with those pieces, great. If instead you want do something fancier (a special queueing dicipline or model of loss), then you’ll need a new C++ object.

There are certainly grey areas in this spectrum: most routing is done in OTcl (although the core Dijkstra algorithm is in C++). We’ve had HTTP simulations where each flow was started in OTcl and per-packet processing was all in C++. This approache worked OK until we had 100s of flows starting per second of simulated time. In general, if you’re ever having to invoke Tcl many times per second, you problably should move that code to C++.

Code Overview

In this document, we use the term “interpreter” to be synonymous with the OTcl interpreter. The code to interface with the interpreter resides in a separate directory, tclcl. The rest of the simulator code resides in the directory, ns-2. We will use the notation ~tclcl/hfilei to refer to a particular hfilei in the Tcl directory. Similarly, we will use the notation, ~ns/hfilei to refer to a particular hfilei in the ns-2 directory.

There are a number of classes defined in ~tclcl/. We only focus on the six that are used in ns: The Class Tcl contains the methods that C++ code will use to access the interpreter. The Class TclObject is the base class for all simulator objects that are also mirrored in the compiled hierarchy. The Class TclClass defines the interpreted class hierarchy, and the methods to permit the user to instantiate TclObjects. The Class TclCommand is used to define simple global interpreter commands. The Class EmbeddedTcl contains the methods to load higher level builtin commands that make configuring simulations easier. Finally, the Class InstVar containsmethods to access C++ member variables as OTcl instance variables.

The procedures and functions described in this chapter can be found in ~tclcl/Tcl.{cc, h}, ~tclcl/, ~tclcl/tcl-object.tcl, and, ~tclcl/tracedvar.{cc, h}. The file ~tclcl/tcl2c++.c is used in building ns, and is mentioned briefly in this chapter.

Class Tcl

The class Tcl encapsulates the actual instance of the OTcl interpreter, and provides the methods to access and communicate with that interpreter. The methods described in this section are relevant to the ns programmer who is writing C++ code. The class provides methods for the following operations:

  • obtain a reference to the Tcl instance;
  • invoke OTcl procedures through the interpreter;
  • retrieve, or pass back results to the interpreter;
  • report error situations and exit in an uniform manner; and
  • store and lookup “TclObjects”.
  • acquire direct access to the interpreter.

We describe each of the methods in the following subsections.

Obtain a Reference to the class Tcl instance

A single instance of the class is declared in ~tclcl/ as a static member variable; the programmer must obtain a reference to this instance to access other methods described in this section. The statement required to access this instance is: Tcl& tcl = Tcl::instance();

Invoking OTcl Procedures

There are four different methods to invoke an OTcl command through the instance, tcl. They differ essentially in their calling arguments. Each function passes a string to the interpreter, that then evaluates the string in a global context. These methods will return to the caller if the interpreter returns TCL_OK. On the other hand, if the interpreter returns TCL_ERROR, the methods will call tkerror{}. The user can overload this procedure to selectively disregard certain types of errors. Such intricacies of OTcl programming are outside the scope of this document. The next section describes methods to access the result returned by the interpreter.

  • tcl.eval(char* s) invokes Tcl_GlobalEval() to execute s through the interpreter.
  • tcl.evalc(const char* s) preserves the argument string s. It copies the string s into its internal buffer; it then invokes the previous eval(char* s) on the internal buffer.
  • tcl.eval() assumes that the command is already stored in the class’ internal bp_; it directly invokes tcl.eval(char* bp_). A handle to the buffer itself is available through the method tcl.buffer(void).
  • tcl.evalf(const char* s, . . . ) is a Printf(3) like equivalent. It uses vsprintf(3) internally to create the input string.

As an example, here are some of the ways of using the above methods:

Tcl& tcl = Tcl::instance();
char wrk[128];
strcpy(wrk, "Simulator set NumberInterfaces_ 1");

sprintf(tcl.buffer(), "Agent/SRM set requestFunction_ %s", "Fixed");

tcl.evalc("puts stdout hello world");
tcl.evalf("%s request %d %d", name_, sender, msgid);

Passing Results to/from the Interpreter

When the interpreter invokes a C++ method, it expects the result back in the privatemember variable, tcl_->result. Two methods are available to set this variable.

  • tcl.result(const char* s) Pass the result string s back to the interpreter.
  • tcl.resultf(const char* fmt, . . . ) varargs(3) variant of above to format the result using vsprintf(3), pass the result string back to the interpreter.
if (strcmp(argv[1], "now") == 0) {
   tcl.resultf("%.17g", clock());
   return TCL_OK;
tcl.result("Invalid operation specified");
return TCL_ERROR;

Likewise, when a C++ method invokes an OTcl command, the interpreter returns the result in tcl_->result.

  • tcl.result(void) must be used to retrieve the result. Note that the result is a string, that must be converted into an

internal format appropriate to the type of result.

tcl.evalc("Simulator set NumberInterfaces_");
char* ni = tcl.result();
if (atoi(ni) != 1)
   tcl.evalc("Simulator set NumberInterfaces_ 1");

Error Reporting and Exit

This method provides a uniform way to report errors in the compiled code.

  • tcl.error(const char* s) performs the following functions: write s to stdout; write tcl_->result to stdout; exit with error code 1.
tcl.resultf("cmd = %s", cmd);
tcl.error("invalid command specified");

Note that there are minor differences between returning TCL_ERROR as we did in the previous subsection, and calling Tcl::error(). The former generates an exception within the interpreter; the user can trap the exception and possibly recover from the error. If the user has not specified any traps, the interpreter will print a stack trace and exit. However, if the code invokes error(), then the simulation user cannot trap the error; in addition, ns will not print any stack trace.

Hash Functions within the Interpreter

ns stores a reference to every TclObject in the compiled hierarchy in a hash table; this permits quick access to the objects. The hash table is internal to the interpreter. ns uses the name of the TclObject as the key to enter, lookup, or delete the TclObject in the hash table.

  • tcl.enter(TclObject* o) will insert a pointer to the TclObject o into the hash table. It is used by TclClass::create_shadow() to insert an object into the table, when that object is created.
  • tcl.lookup(char* s) will retrieve the TclObject with the name s. It is used by TclObject::lookup().
  • tcl.remove(TclObject* o) will delete references to the TclObject o from the hash table. It is used by TclClass::delete_shadow() to remove an existing entry from the hash table, when that object is deleted.

These functions are used internally by the class TclObject and class TclClass.

Other Operations on the Interpreter

If the above methods are not sufficient, then we must acquire the handle to the interpreter, and write our own functions.

  • tcl.interp(void) returns the handle to the interpreter that is stored within the class Tcl.

Class TclObject

class TclObject is the base class for most of the other classes in the interpreted and compiled hierarchies. Every object in the class TclObject is created by the user fromwithin the interpreter. An equivalent shadow object is created in the compiled hierarchy. The two objects are closely associated with each other. The class TclClass, described in the next section, contains the mechanisms that perform this shadowing.

In the rest of this document, we often refer to an object as a TclObject.<ref>In the latest release of ns and ns/tclcl, this object has been renamed to SplitObjefct, which more accurately reflects its nature of existence. However, for the moment, we will continue to use the term TclObject to refer to these objects and this class.</ref> By this, we refer to a particular object that is either in the class TclObject, or in a class that is derived from the class TclObject. If it is necessary, we will explicitly qualify whether that object is an object within the interpreter, or an object within the compiled code. In such cases, we will use the abbreviations “interpreted object”, and “compiled object” to distinguish the two. and within the compiled code respectively.

Differences from ns v1 Unlike ns v1, the class TclObject subsumes the earlier functions of the NsObject class. It therefore stores the interface variable bindings that tie OTcl instance variables in the interpreted object to corresponding C++ member variables in the compiled object. The binding is stronger than in ns v1 in that any changes to the OTcl variables are trapped, and the current C++ and OTcl values are made consistent after each access through the interpreter. The consistency is done through the Class InstVar. Also unlike ns v1, objects in the class TclObject are no longer stored as a global link list. Instead, they are stored in a hash table in the class Tcl.

Example configuration of a TclObject The following example illustrates the configuration of an SRM agent (class Agent/SRM/Adaptive).

set srm [new Agent/SRM/Adaptive]
$srm set packetSize_ 1024
$srm traffic-source $s0

By convention in ns, the class Agent/SRM/Adaptive is a subclass of Agent/SRM, is a subclass of Agent, is a subclass of TclObject. The corresponding compiled class hierarchy is the ASRMAgent, derived from SRMAgent, derived from Agent, derived from TclObject respectively. The first line of the above example shows how a TclObject is created (or destroyed); the next line configures a [#Variable Bindings|bound variable]]; and finally, the last line illustrates the interpreted object invoking a C++ method as if they were an instance procedure.

Creating and Destroying TclObjects

When the user creates a new TclObject, using the procedures new{} and delete{}; these procedures are defined in ~tclcl/tcl-object.tcl. They can be used to create and destroy objects in all classes, including TclObjects.<ref>As an example, the classes Simulator, Node, Link, or rtObject, are classes that are not derived from the class TclObject. Objects in these classes are not, therefore, TclObjects. However, a Simulator, Node, Link, or route Object is also instantiated using the new procedure in ns.</ref> In this section, we describe the internal actions executed when a TclObject is created.

Creating TclObjects By using new{}, the user creates an interpreted TclObject. the interpreter will execute the constructor for that object, init{}, passing it any arguments provided by the user. ns is responsible for automatically creating the compiled object. The shadow object gets created by the base class TclObject’s constructor. Therefore, the constructor for the new TclObject must call the parent class constructor first. new{} returns a handle to the object, that can then be used for further operations upon that object.

The following example illustrates the Agent/SRM/Adaptive constructor:

Agent/SRM/Adaptive instproc init args {
   eval $self next $args
   $self array set closest_ "requestor 0 repairor 0"
   $self set eps_ [$class set eps_]

The following sequence of actions are performed by the interpreter as part of instantiating a new TclObject. For ease of exposition, we describe the steps that are executed to create an Agent/SRM/Adaptive object. The steps are:

  1. Obtain an unique handle for the new object from the TclObject name space. The handle is returned to the user. Most handles in ns have the form _ohNNNi, where hNNNi is an integer. This handle is created by getid{}. It can be retrieved from C++ with the name(){} method.
  2. Execute the constructor for the new object. Any user-specified arguments are passed as arguments to the constructor. This constructor must invoke the constructor associated with its parent class. In our example above, the Agent/SRM/Adaptive calls its parent class in the very first line. Note that each constructor, in turn invokes its parent class’ constructor ad nauseum. The last constructor in ns is the TclObject constructor. This constructor is responsible for setting up the shadow object, and performing other initializations and bindings, as we explain below. It is preferable to call the parent constructors first before performing the initializations required in this class. This allows the shadow objects to be set up, and the variable bindings established.
  3. The TclObject constructor invokes the instance procedure create-shadow{} for the class Agent/SRM/Adaptive.
  4. When the shadow object is created, ns calls all of the constructors for the compiled object, each of which may establish variable bindings for objects in that class, and perform other necessary initializations. Hence our earlier injunction that it is preferable to invoke the parent constructors prior to performing the class initializations.
  5. After the shadow object is successfully created, create_shadow(void)
    1. adds the new object to the hash table of TclObjects described earlier.
    2. makes cmd{} an instance procedure of the newly created interpreted object. This instance procedure invokes the command() method of the compiled object. In a later subsection, we describe how the command method is defined, and invoked.

Note that all of the above shadowing mechanisms only work when the user creates a new TclObject through the interpreter. It will not work if the programmer creates a compiled TclObject unilaterally. Therefore, the programmer is enjoined not to use the C++ new method to create compiled objects directly.

Deletion of TclObjects The delete operation destroys the interpreted object, and the corresponding shadow object. For example, use-scheduler{hscheduleri} uses the delete procedure to remove the default list scheduler, and instantiate an alternate scheduler in its place.

Simulator instproc use-scheduler type {
   $self instvar scheduler_
   delete scheduler_ ;                   # first delete the existing list scheduler
   set scheduler_ [new Scheduler/$type]

As with the constructor, the object destructor must call the destructor for the parent class explicitly as the very last statement of the destructor. The TclObject destructor will invoke the instance procedure delete-shadow, that in turn invokes the equivalent compiled method to destroy the shadow object. The interpreter itself will destroy the interpreted object.

Variable Bindings

In most cases, access to compiled member variables is restricted to compiled code, and access to interpretedmember variables is likewise confined to access via interpreted code; however, it is possible to establish bi-directional bindings such that both the interpreted member variable and the compiled member variable access the same data, and changing the value of either variable changes the value of the corresponding paired variable to same value.

The binding is established by the compiled constructor when that object is instantiated; it is automatically accessible by the interpreted object as an instance variable. ns supports five different data types: reals, bandwidth valued variables, time valued variables, integers, and booleans. The syntax of how these values can be specified in OTcl is different for each variable type.

  • Real and Integer valued variables are specified in the “normal” form. For example,
$object set realvar 1.2e3
$object set intvar 12
  • Bandwidth is specified as a real value, optionally suffixed by a ‘k’ or ‘K’ to mean kilo-quantities, or ‘m’ or ‘M’ to mean mega-quantities. A final optional suffix of ‘B’ indicates that the quantity expressed is in Bytes per second. The default is bandwidth expressed in bits per second. For example, all of the following are equivalent:
$object set bwvar 1.5m
$object set bwvar 1.5mb
$object set bwvar 1500k
$object set bwvar 1500kb
$object set bwvar .1875MB
$object set bwvar 187.5kB
$object set bwvar 1.5e6
  • Time is specified as a real value, optionally suffixed by a ‘m’ to express time in milli-seconds, ‘n’ to express time in nano-seconds, or ‘p’ to express time in pico-seconds. The default is time expressed in seconds. For example, all of the following are equivalent:
$object set timevar 1500m
$object set timevar 1.5
$object set timevar 1.5e9n
$object set timevar 1500e9p

Note that we can also safely add a s to reflect the time unit of seconds. ns will ignore anything other than a valid real number specification, or a trailing ‘m’, ‘n’, or ‘p’. • Booleans can be expressed either as an integer, or as ‘T’ or ‘t’ for true. Subsequent characters after the first letter are ignored. If the value is neither an integer, nor a true value, then it is assumed to be false. For example,

$object set boolvar t ;           # set to true
$object set boolvar true
$object set boolvar 1 ;           # or any non-zero value
$object set boolvar false ;       # set to false
$object set boolvar junk
$object set boolvar 0

The following example shows the constructor for the ASRMAgent.<ref>Note that this constructor is embellished to illustrate the features of the variable binding mechanism.</ref>

ASRMAgent::ASRMAgent() {
   bind("pdistance_", &pdistance_);        /* real variable */
   bind("requestor_", &requestor_);        /* integer variable */
   bind_time("lastSent_", &lastSessSent_); /* time variable */
   bind_bw("ctrlLimit_", &ctrlBWLimit_);   /* bandwidth variable */
   bind_bool("running_", &running_);       /* boolean variable */

Note that all of the functions above take two arguments, the name of an OTcl variable, and the address of the corresponding compiled member variable that is linked. While it is often the case that these bindings are established by the constructor of the object, it need not always be done in this manner. We will discuss such alternate methods when we describe the [[#Class InstVar|class InstVar]] in detail later.

Each of the variables that is bound is automatically initialised with default values when the object is created. The default values are specified as interpreted class variables. This initialisation is done by the routing init-instvar{}, invoked by methods in the class InstVar. init-instvar{} checks the class of the interpreted object, and all of the parent class of that object, to find the first class in which the variable is defined. It uses the value of the variable in that class to initialise the object. Most of the bind initialisation values are defined in ~ns/tcl/lib/ns-default.tcl.

For example, if the following class variables are defined for the ASRMAgent:

Agent/SRM/Adaptive set pdistance_ 15.0
Agent/SRM set pdistance_ 10.0
Agent/SRM set lastSent_ 8.345m
Agent set ctrlLimit_ 1.44M
Agent/SRM/Adaptive set running_ f

Therefore, every new Agent/SRM/Adaptive object will have pdistance_ set to 15.0; lastSent_ is set to 8.345m from the setting of the class variable of the parent class; ctrlLimit_ is set to 1.44M using the class variable of the parent class twice removed; running is set to false; the instance variable pdistance_ is not initialised, because no class variable exists in any of the class hierarchy of the interpreted object. In such instance, init-instvar{} will invoke warn-instvar{}, to print out a warning about such a variable. The user can selectively override this procedure in their simulation scripts, to elide this warning.

Note that the actual binding is done by instantiating objects in the class InstVar. Each object in the class InstVar binds one compiled member variable to one interpreted member variable. A TclObject stores a list of InstVar objects corresponding to each of its member variable that is bound in this fashion. The head of this list is stored in its member variable instvar_ of the TclObject.

One last point to consider is that ns will guarantee that the actual values of the variable, both in the interpreted object and the compiled object, will be identical at all times. However, if there are methods and other variables of the compiled object that track the value of this variable, they must be explicitly invoked or changed whenever the value of this variable is changed. This usually requires additional primitives that the user should invoke. One way of providing such primitives in ns is through the command() method described in the next section.

Variable Tracing

In addition to variable bindings, TclObject also supports tracing of both C++ and Tcl instance variables. A traced variable can be created and configured either in C++ or Tcl. To establish variable tracing at the Tcl level, the variable must be visible in Tcl, which means that it must be a bound C++/Tcl or a pure Tcl instance variable. In addition, the object that owns the traced variable is also required to establish tracing using the Tcl trace method of TclObject. The first argument to the trace method must be the name of the variable. The optional second argument specifies the trace object that is responsible for tracing that variable. If the trace object is not specified, the object that owns the variable is responsible for tracing it.

For a TclObject to trace variables, it must extend the C++ trace method that is virtually defined in TclObject. The Trace class implements a simple trace method, thereby, it can act as a generic tracer for variables.

class Trace : public Connector {
   virtual void trace(TracedVar*);

Below is a simple example for setting up variable tracing in Tcl:

# $tcp tracing its own variable cwnd_
$tcp trace cwnd_
# the variable ssthresh_ of $tcp is traced by a generic $tracer
set tracer [new Trace/Var]
$tcp trace ssthresh_ $tracer

For a C++ variable to be traceable, it must belong to a class that derives from TracedVar. The virtual base class TracedVar keeps track of the variable’s name, owner, and tracer. Classes that derive from TracedVar must implement the virtual method value(), that takes a character buffer as an argument and writes the value of the variable into that buffer.

class TracedVar {
   virtual char* value(char* buf) = 0;
   TracedVar(const char* name);
   const char* name_; // name of the variable
   TclObject* owner_; // the object that owns this variable
   TclObject* tracer_; // callback when the variable is changed

The TclCL library exports two classes of TracedVar: TracedInt and TracedDouble. These classes can be used in place of the basic type int and double respectively. Both TracedInt and TracedDouble overload all the operators that can change the value of the variable such as assignment, increment, and decrement. These overloaded operators use the assign method to assign the new value to the variable and call the tracer if the new value is different from the old one. TracedInt and TracedDouble also implement their value methods that output the value of the variable into the string. The width and precision of the output can be pre-specified. (How??)

Command Methods: Definition and Invocation

For every TclObject that is created, ns establishes the instance procedure, cmd{}, as a hook to executing methods through the compiled shadow object. The procedure cmd{} invokes the method command() of the shadow object automatically, passing the arguments to cmd{} as an argument vector to the command() method.

The user can invoke the cmd{} method in one of two ways: by explicitly invoking the procedure, specifying the desired operation as the first argument, or implicitly, as if there were an instance procedure of the same name as the desired operation. Most simulation scripts will use the latter form, hence, we will describe that mode of invocation first.

Consider the that the distance computation in SRMis done by the compiled object; however, it is often used by the interpreted object. It is usually invoked as:

$srmObject distance? hagentAddressi

If there is no instance procedure called distance?, the interpreter will invoke the instance procedure unknown{}, defined in the base class TclObject. The unknown procedure then invokes

$srmObject cmd distance? hagentAddressi

to execute the operation through the compiled object’s command() procedure.

Of course, the user could explicitly invoke the operation directly. One reason for this might be to overload the operation by using an instance procedure of the same name. For example,

Agent/SRM/Adaptive instproc distance? addr {
   $self instvar distanceCache_
   if ![info exists distanceCache_($addr)] {
       set distanceCache_($addr) [$self cmd distance? $addr]
   set distanceCache_($addr)

We now illustrate how the command() method using ASRMAgent::command() as an example.

int ASRMAgent::command(int argc, const char*const*argv) {
   Tcl& tcl = Tcl::instance();
   if (argc == 3) {
      if (strcmp(argv[1], "distance?") == 0) {
          int sender = atoi(argv[2]);
          SRMinfo* sp = get_state(sender);
          tcl.tesultf("%f", sp->distance_);
          return TCL_OK;
   return (SRMAgent::command(argc, argv));

We can make the following observations from this piece of code:

  • The function is called with two arguments:

The first argument (argc) indicates the number of arguments specified in the command line to the interpreter. The command line arguments vector (argv) consists of

    • argv[0] contains the name of the method, “cmd”.
    • argv[1] specifies the desired operation.
    • If the user specified any arguments, then they are placed in argv[2...(argc - 1)].

The arguments are passed as strings; they must be converted to the appropriate data type.

  • If the operation is successfully matched, the match should return the result of the operation using methods described earlier (Section 3.3.3).
  • command() itself must return either TCL_OK or TCL_ERROR to indicate success or failure as its return code.
  • If the operation is notmatched in thismethod, itmust invoke its parent’s commandmethod, and return the corresponding result. This permits the user to concieve of operations as having the same inheritance properties as instance procedures or compiled methods. In the event that this command method is defined for a class with multiple inheritance, the programmer has the liberty to choose one of two implementations:
    1. Either they can invoke one of the parent’s command method, and return the result of that invocation, or
    2. They can each of the parent’s command methods in some sequence, and return the result of the first invocation that is successful. If none of them are successful, then they should return an error.

In our document, we call operations executed through the command() instproc-likes. This reflects the usage of these operations as if they were OTcl instance procedures of an object, but can be very subtly different in their realisation and usage.

Class TclClass

This compiled class (class TclClass) is a pure virtual class. Classes derived from this base class provide two functions: construct the interpreted class hierarchy to mirror the compiled class hierarchy; and provide methods to instantiate new TclObjects. Each such derived class is associated with a particular compiled class in the compiled class hierarchy, and can instantiate new objects in the associated class.

As an example, consider a class such as the class RenoTcpClass. It is derived from class TclClass, and is associated with the class RenoTcpAgent. It will instantiate new objects in the class RenoTcpAgent. The compiled class hierarchy for RenoTcpAgent is that it derives from TcpAgent, that in turn derives from Agent, that in turn derives (roughly) from TclObject. RenoTcpClass is defined as

static class RenoTcpClass: public TclClass {
   RenoTcpClass() : TclClass("Agent/TCP/Reno") {}
   TclObject* create(int argc, const char*const* argv) {
       return (new RenoTcpAgent());
} class_reno;

We can make the following observations from this definition:

  1. The class defines only the constructor, and one additional method, to create instances of the associated TclObject.
  2. ns will execute the RenoTcpClass constructor for the static variable class_reno, when it is first started. This sets up the appropriate methods and the interpreted class hierarchy.
  3. The constructor specifies the interpreted class explicitly as Agent/TCP/Reno. This also specifies the interpreted class hierarchy implicitly. Recall that the convention in ns is to use the character slash (’/’) is a separator. For any given class A/B/C/D, the class A/B/C/D is a sub-class of A/B/C, that is itself a sub-class of A/B, that, in turn, is a sub-class of A. A itself is a sub-class of TclObject. In our case above, the TclClass constructor creates three classes, Agent/TCP/Reno sub-class of Agent/TCP subclass of Agent sub-class of TclObject.
  4. This class is associated with the class RenoTcpAgent; it creats new objects in this associated class.
  5. The RenoTcpClass::create method returns TclObjects in the class RenoTcpAgent.
  6. When the user specifies new Agent/TCP/Reno, the routine RenoTcpClass::create is invoked.
  7. The arguments vector (argv) consists of
    1. argv[0] contains the name of the object.
    2. argv[1...3] contain $self, $class, and $proc.Since create is called through the instance procedure create-shadow, argv[3] contains create-shadow.
    3. argv[4] contain any additional arguments (passed as a string) provided by the user.

The class Trace illustrates argument handling by TclClass methods.

class TraceClass : public TclClass {
   TraceClass() : TclClass("Trace") {}
   TclObject* create(int args, const char*const* argv) {
       if (args >= 5)
           return (new Trace(*argv[4]));
           return NULL;
} trace_class;

A new Trace object is created as

new Trace "X"

Finally, the nitty-gritty details of how the interpreted class hierarchy is constructed:

  1. The object constructor is executed when ns first starts.
  2. This constructor calls the TclClass constructor with the name of the interpreted class as its argument.
  3. The TclClass constructor stores the name of the class, and inserts this object into a linked list of the TclClass objects.
  4. During initialization of the simulator, Tcl_AppInit(void) invokes TclClass::bind(void)
  5. For each object in the list of TclClass objects, bind() invokes register{}, specifying the name of the interpreted class as its argument.
  6. register{} establishes the class hierarchy, creating the classes that are required, and not yet created.
  7. Finally, bind() defines instance procedures create-shadow and delete-shadow for this new class.

How to Bind Static C++ Class Member Variables

In the previous section, we have seen how to expose member variables of a C++ object into OTcl space. This, however, does not apply to static member variables of a C++ class. Of course, one may create an OTcl variable for the static member variable of every C++ object; obviously this defeats the whole meaning of static members.

We cannot solve this binding problem using a similar solution as binding in TclObject, which is based on InstVar, because InstVars in TclCL require the presence of a TclObject. However, we can create a method of the corresponding TclClass and access static members of a C++ class through the methods of its corresponding TclClass. The procedure is as follows:

  1. Create your own derived TclClass as described above;
  2. Declare methods bind() and method() in your derived class;
  3. Create your binding methods in the implementation of your bind() with add_method("your_method"), then implement the handler in method() in a similar way as you would do in TclObject::command(). Notice that the number of arguments passed to TclClass::method() are different fromthose passed to TclObject::command(). The former has two more arguments in the front.

As an example, we show a simplified version of PacketHeaderClass in ~ns/ Suppose we have the following class Packet which has a static variable hdrlen_ that we want to access from OTcl:

class Packet {
   static int hdrlen_;

Then we do the following to construct an accessor for this variable:

class PacketHeaderClass : public TclClass {
   PacketHeaderClass(const char* classname, int hdrsize);
   TclObject* create(int argc, const char*const* argv);
       /* These two implements OTcl class access methods */
   virtual void bind();
   virtual int method(int argc, const char*const* argv);

void PacketHeaderClass::bind()
       /* Call to base class bind() must precede add_method() */

int PacketHeaderClass::method(int ac, const char*const* av)
   Tcl& tcl = Tcl::instance();
       /* Notice this argument translation; we can then handle them as if in TclObject::command() */
   int argc = ac - 2;
   const char*const* argv = av + 2;
   if (argc == 2) {
       if (strcmp(argv[1], "hdrlen") == 0) {
           tcl.resultf("%d", Packet::hdrlen_);
           return (TCL_OK);
   } else if (argc == 3) {
       if (strcmp(argv[1], "hdrlen") == 0) {
           Packet::hdrlen_ = atoi(argv[2]);
           return (TCL_OK);
   return TclClass::method(ac, av);

After this, we can then use the following OTcl command to access and change values of Packet::hdrlen_:

PacketHeader hdrlen 120
set i [PacketHeader hdrlen]

Class TclCommand

This class (class TclCommand) provides just the mechanism for ns to export simple commands to the interpreter, that can then be executed within a global context by the interpreter. There are two functions defined in ~ns/ ns-random and ns-version. These two functions are initialized by the function init_misc(void), defined in ~ns/; init_misc is invoked by Tcl_AppInit(void) during startup.

  • class VersionCommand defines the command ns-version. It takes no argument, and returns the current ns version string.
% ns-version ;# get the current version
  • class RandomCommand defines the command ns-random. With no argument, ns-random returns an integer, uniformly distributed in the interval [0, 231 − 1]. When specified an argument, it takes that argument as the seed. If this seed value is 0, the command uses a heuristic seed value; otherwise, it sets the seed for the random number generator to the specified value.
% ns-random ;# return a random number
% ns-random 0 ;#set the seed heuristically
% ns-random 23786 ;#set seed to specified value

Note that, it is generally not advisable to construct top-level commands that are available to the user. We now describe how to define a new command using the example class say_hello. The example defines the command hi, to print the string “hello world”, followed by any command line arguments specified by the user. For example,

% hi this is ns [ns-version]
hello world, this is ns 2.0a12

1. The command must be defined within a class derived from the class TclCommand. The class definition is:

class say_hello : public TclCommand {
   int command(int argc, const char*const* argv);

2. The constructor for the class must invoke the TclCommand constructor with the command as argument; i.e.,

say_hello() : TclCommand("hi") {}

The TclCommand constructor sets up "hi" as a global procedure that invokes TclCommand::dispatch_cmd(). 3. The method command() must perform the desired action. The method is passed two arguments. The first argument, argc, contains the number of actual arguments passed by the user.

The actual arguments passed by the user are passed as an argument vector (argv) and contains the following:

  • argv[0] contains the name of the command (hi).
  • argv[1...(argc - 1)] contains additional arguments specified on the command line by the user.

command() is invoked by dispatch_cmd().

#include <streams.h> /* because we are using stream I/O */
int say_hello::command(int argc, const char*const* argv) {
   cout << "hello world:";
   for (int i = 1; i < argc; i++)
       cout << ’ ’ << argv[i];
   cout << ’\ n’;
   return TCL_OK;

4. Finally, we require an instance of this class. TclCommand instances are created in the routine init_misc(void).

new say_hello;

Note that there used to be more functions such as ns-at and ns-now that were accessible in this manner. Most of these functions have been subsumed into existing classes. In particular, ns-at and ns-now are accessible through the scheduler TclObject. These functions are defined in ~ns/tcl/lib/ns-lib.tcl.

% set ns [new Simulator] ;# get new instance of simulator
% $ns now ;# query simulator for current time
% $ns at ... ;# specify at operations for simulator

Class EmbeddedTcl

ns permits the development of functionality in either compiled code, or through interpreter code, that is evaluated at initialization. For example, the scripts ~tclcl/tcl-object.tcl or the scripts in ~ns/tcl/lib. Such loading and evaluation of scripts is done through objects in the class EmbeddedTcl.

The easiest way to extend ns is to add OTcl code to either ~tclcl/tcl-object.tcl or through scripts in the ~ns/tcl/lib directory. Note that, in the latter case, ns sources ~ns/tcl/lib/ns-lib.tcl automatically, and hence the programmermust add a couple of lines to this file so that their script will also get automatically sourced by ns at startup. As an example, the file ~ns/tcl/mcast/srm.tcl defines some of the instance procedures to run SRM. In ~ns/tcl/lib/ns-lib.tcl, we have the lines:

source tcl/mcast/srm.tcl

to automatically get srm.tcl sourced by ns at startup.

Three points to note with EmbeddedTcl code are that firstly, if the code has an error that is caught during the eval, then ns will not run. Secondly, the user can explicitly override any of the code in the scripts. In particular, they can re-source the entire script after making their own changes. Finally, after adding the scripts to ~ns/tcl/lib/ns-lib.tcl, and every time thereafter that they change their script, the user must recompile ns for their changes to take effect. Of course, in most cases<ref>The few places where this might not work are when certain variables might have to be defined or undefined, or otherwise the script contains code other than procedure and variable definitions and executes actions directly that might not be reversible.</ref>, the user can source their script to override the embedded code.

The rest of this subsection illustrate how to integrate individual scripts directly into ns. The first step is convert the script into an EmbeddedTcl object. The lines below expand ns-lib.tcl and create the EmbeddedTcl object instance called et_ns_lib:

tclsh bin/tcl-expand.tcl tcl/lib/ns-lib.tcl | \
     ../Tcl/tcl2c++ et_ns_lib > gen/

The script, ~ns/bin/tcl-expand.tcl expands ns-lib.tcl by replacing all source lines with the corresponding source files. The program, ~tclcl/tcl2cc.c, converts the OTcl code into an equivalent EmbeddedTcl object, et_ns_lib.

During initialization, invoking the method EmbeddedTcl::load explicitly evaluates the array.

    • ~tclcl/tcl-object.tcl is evaluated by the method Tcl::init(void); Tcl_AppInit() invokes Tcl::Init(). The exact command syntax for the load is:
    • Similarly, ~ns/tcl/lib/ns-lib.tcl is evaluated directly by Tcl_AppInit in ~ns/

Class InstVar

This section describes the internals of the class InstVar. This class defines the methods and mechanisms to bind a C++ member variable in the compiled shadow object to a specified OTcl instance variable in the equivalent interpreted object. The binding is set up such that the value of the variable can be set or accessed either from within the interpreter, or from within the compiled code at all times.

There are five instance variable classes: class InstVarReal, class InstVarTime, class InstVarBandwidth, class InstVarInt, and class InstVarBool, corresponding to bindings for real, time, bandwidth, integer, and boolean valued variables respectively.

We now describe the mechanism by which instance variables are set up. We use the class InstVarReal to illustrate the concept. However, this mechanism is applicable to all five types of instance variables.

When setting up an interpreted variable to access a member variable, the member functions of the class InstVar assume that they are executing in the appropriate method execution context; therefore, they do not query the interpreter to determine the context in which this variable must exist.

In order to guarantee the correct method execution context, a variable must only be bound if its class is already established within the interpreter, and the interpreter is currently operating on an object in that class. Note that the former requires that when a method in a given class is going to make its variables accessible via the interpreter, there must be an associated class TclClass defined that identifies the appropriate class hierarchy to the interpreter. The appropriate method execution context can therefore be created in one of two ways.

An implicit solution occurs whenever a new TclObject is created within the interpreter. This sets up the method execution context within the interpreter. When the compiled shadow object of the interpreted TclObject is created, the constructor for that compiled object can bind its member variables of that object to interpreted instance variables in the context of the newly created interpreted object.

An explicit solution is to define a bind-variables operation within a command function, that can then be invoked via the cmd method. The correct method execution context is established in order to execute the cmd method. Likewise, the compiled code is now operating on the appropriate shadow object, and can therefore safely bind the required member variables.

An instance variable is created by specifying the name of the interpreted variable, and the address of the member variable in the compiled object. The constructor for the base class InstVar creates an instance of the variable in the interpreter, and then sets up a trap routine to catch all accesses to the variable through the interpreter.

Whenever the variable is read through the interpreter, the trap routine is invoked just prior to the occurrence of the read. The routine invokes the appropriate get function that returns the current value of the variable. This value is then used to set the value of the interpreted variable that is then read by the interpreter.

Likewise, whenever the variable is set through the interpreter, the trap routine is invoked just after to the write is completed. The routine gets the current value set by the interpreter, and invokes the appropriate set function that sets the value of the compiled member to the current value set within the interpreter.

Previous chapter: "Undocumented Facilities" | index | Next chapter: "The Class Simulator"