5.6.5 Breakpoint Actions

The action part of a breakpoint spec supplies information to the debugger as to what should be done when the breakpoint is activated. This is achieved by setting the three so called debugger action variables. These are listed below, together with their most important values.

For example, the breakpoint below specifies that whenever the Exit port of foo/2 is reached, no trace message should be output, no interaction should take place and the debugger should be switched off. 

     | ?- add_breakpoint([pred(foo/2),port(exit)]-
                         [show(silent),command(proceed),mode(off)], _).

Here, the action part consists of three actions, setting the three action variables. This breakpoint spec can be simplified by omitting the wrappers around the variable values, as the sets of possible values of the variables are all disjoint. If we use spy/2, the pred wrapper goes away, too, resulting in a much more concise, equivalent formulation of the above breakpoint: 

     | ?- spy(foo/2,exit-[silent,proceed,off]).

Let us now revisit the process of breakpoint selection. When the debugger arrives at a port it first initializes the action variables according to the current debugging and leashing modes, as shown below: 

     debugging    leashing           |        Action variables
     mode         mode               |  show        command     mode
     --------------------------------|-------------------------------
     trace        at leashed port    |  print       ask         trace
                                     |
     trace        at unleashed port  |  print       proceed     trace
                                     |
     debug        -                  |  silent      proceed     debug
                                     |
     zip          -                  |  silent      flit        zip

It then considers each breakpoint, most recent first, until it finds a breakpoint whose test part succeeds. If such a breakpoint is found, its action part is evaluated, normally changing the action variable settings. A failure of the action part is ignored, in the sense that the breakpoint is still treated as the selected one. However, as a side-effect, a procedure box will always be built in such cases. More precisely, the failure of the action part causes the flit command value to be changed to proceed, all other command values being left unchanged. This is to facilitate the creation of breakpoints that stop at non-Call ports (see below for an example).

If no applicable breakpoint is found, the action variables remain unchanged.

The debugger then executes the actions specified by the action variables. This process, referred to as the action execution, means the following:

Specifically, if command is ask, the user is prompted for a debugger command, which in turn is converted to new assignments to the action variables. The debugger will then repeat the action execution process, described above. For example, the `c' (creep) interactive command is converted to [silent,proceed,trace], the `d' (display) command to [display,ask] (when command is ask, the mode is irrelevant), etc.

The default values of the action variables correspond to the standard debugger behavior described in Basic Debug. For example, when an unleashed port is reached in trace mode, a trace message is printed and the execution proceeds in trace mode, without stopping. In zip mode, no trace message is shown, and execution continues in zip mode, without building procedure boxes at Call ports.

Note that a spypoint action part that is empty ([] or not present) is actually treated as [print,ask]. Again, this is the standard behavior of spypoints, as described in Basic Debug.

If an action part is nonempty, but it does not set the action variables, the only effect it will have is to hide the remaining older spypoints, as the debugger will behave in the standard way, according to the debugging mode. Still, such breakpoints may be useful if they have side-effects, for example: 

     | ?- spy(foo/2, -[parent_pred(P),
                       goal(G),
                       true(format('~q called from:~w~n',[G,P]))]).
     % The debugger will first zip -- showing spypoints (zip)
     % Conditional spypoint for user:foo/2 added, BID=1
     true
     % zip
     | ?- foo(3,X).
     foo(2,_701) called from:bar/3
     foo(1,_1108) called from:bar/3
     foo(0,_1109) called from:bar/3
     foo(1,_702) called from:bar/3
     X = 2 ? ;
     no

This spypoint produces some output at ports of foo/2, but otherwise will not influence the debugger. Notice that a breakpoint spec with an empty test part can be written -Actions.

Let us look at some simple examples of what other effects can be achieved by appropriate action variable settings: 

     | ?- spy(foo/2, -[print,proceed]).

This is an example of an unleashed spypoint: it will print a trace message passing each port of foo/2, but will not stop there. Note that because of the proceed command a procedure box will be built, even in zip mode, and so the debugger will be activated at non-Call ports of foo/2.

The next example is a variant of the above: 

     | ?- spy(foo/2, -[print,flit]).

This will print a trace message at the Call port of foo/2 and will then continue the execution in the current debugging mode, without building a procedure box for this call. This means that the debugger will not be able to notice any other ports of foo/2.

Now let us address the task of stopping at a specific non-Call port of a predicate. For this to work in zip mode, one has to ensure that a procedure box is built at the Call port. In the following example, the first spypoint causes a box to be built for each call of foo/2, while the second one makes the debugger stop when the Fail port of foo/2 is reached. 

     | ?- spy(foo/2, call-proceed), spy(foo/2, fail).
     % Conditional spypoint for user:foo/2 added, BID=1
     % Conditional spypoint for user:foo/2 added, BID=2

You can achieve the same effect with a single spypoint, by putting the fail condition (which is a shortcut for port(fail)) in the action part, rather than in the test part. 

     | ?- spy(foo/2, -[fail,print,ask]).

Here, when the execution reaches the Call port of foo/2, the test part (which contains the pred(foo/2) condition only) succeeds, so the breakpoint is found applicable. However, the action part fails at the Call port. This has a side-effect in zip mode, as the default flit command value is changed to proceed. In other modes the action variables are unaffected. The net result is that a procedure box is always built for foo/2, which means that the debugger will actually reach the Fail port of this predicate. When this happens, the action part succeeds, and executing the actions print,ask will cause the debugger to stop.

Note that we have to explicitly mention the print,ask actions here, because the action part is otherwise nonempty (contains the fail condition). It is only the empty or missing action part, which is replaced by the default [print,ask]. If you want to include a condition in the action part, you have to explicitly mention all action variable settings you need.

To make this simpler, the debugger handles breakpoint condition macros, which expand to other conditions. For example leash is a macro that expands to [print,ask]. Consequently, the last example can be simplified to: 

     | ?- spy(foo/2, -[fail,leash]).

Similarly, the macro unleash expands to [print,proceed], while hide to [silent,proceed].

We now briefly describe further possible settings to the action variables.

The mode variable can be assigned the values skip(Inv) and qskip(Inv), meaning skipping and quasi-skipping until a port is reached whose invocation number is less or equal to Inv. When the debugger arrives at this port it sets the mode variable to trace.

It may be surprising that skip(...) is a mode, rather than a command. This is because commands are executed and immediately forgotten, but skipping has a lasting effect: the program is to be run with no debugging until a specific point, without creating new procedure boxes, and ignoring the existing ones in the meantime.

Here is an example using the skip mode: 

     | ?- spy(foo/2,call-[print,proceed,inv(Inv),skip(Inv)]).

This breakpoint will be found applicable at Call ports of foo/2. It will print a trace message there and will skip over to the Exit or Fail port without stopping. Notice that the number of the current invocation is obtained in the action part, using the inv condition with a variable argument. A variant of this example follows: 

     | ?- spy(foo/2,-[silent,proceed,
                            (   call -> inv(Inv), skip(Inv)
                            ;   true
                            )]).

This spypoint makes foo/2 invisible in the output of the debugger: at all ports we silently proceed (i.e. display nothing and do not stop). Furthermore, at the Call port we perform a skip, so neither foo/2 itself, nor any predicate called within it will be shown by the debugger.

Notice the use of the true/0 test in the above conditional! This is a breakpoint test that always succeeds. The debugger also recognizes false as a test that always fails. Note that while false and fail are synonyms as built-in predicates, they are completely different as breakpoint conditions: the latter is a shortcut for port(fail).

The show variable has four additional value patterns. Setting it to display, write, or write_term(Options) will result in the debugged goal G being shown using display(G), writeq(G), or write_term(G, Options), respectively. The fourth pattern, Method-Sel, can be used for replacing the goal in the trace message by one of its subterms, the one pointed to by the selector Sel.

For example, the following spypoint instructs the debugger to stop at each port of foo/2, and to only display the first argument of foo/2 in the trace message, instead of the complete goal. 

     | ?- spy(foo/2, -[print-[1],ask]).
     % Conditional spypoint for user:foo/2 added, BID=1
     | ?- foo(5,X).
      *      1      1 Call: ^1 5 ?

The command variable has several further value patterns. The variable can be set to proceed(OldGoal,NewGoal). At a Call port this instructs the debugger to first build a procedure box for the current goal, then to unify it with OldGoal and finally execute NewGoal in its place (cf. the `u' (unify) interactive debugger command). At non-Call ports this command first goes back to the Call port (cf. the `r' (retry) command), and then does the above activities.

A variant of the proceed/2 command is flit(OldGoal,NewGoal). This has the same effect, except for not building a procedure box for OldGoal.

We now just briefly list further command values (for the details, see Action Variables). Setting command to exception(E) will raise an exception E, abort will abort the execution. The values retry(Inv), reexit(Inv), redo(Inv), fail(Inv) will cause the debugger to go back to an earlier Call, Exit, Redo, or Fail port with invocation number Inv (cf. the `j' (jump) interactive debugger command).

Sometimes it may be useful to access the value of an action variable. This can be done with the get condition: e.g. get(mode(M)) will unify M with the current execution mode. The get(...) wrapper can be omitted in the test part, but not in the action part (since there a mode(M) action will set, rather than read, the mode action variable). For example: 

     | ?- spy(foo/2, mode(trace)-show(print-[1])).

This spypoint will be found applicable only in trace mode (and will cause the first argument of foo/2 to appear in the trace message). (The mode and show wrappers can also be omitted in the above example, they are used only to help with interpreting the breakpoint spec.)

Send feedback on this subject.