Chapter 6 Reactis Coverage Metrics
Reactis uses a number of different coverage metrics on Simulink / Stateflow models to measure how thoroughly a test or set of tests exercises a model. In general, coverage metrics record how many of a given class of syntactic constructs, or coverage targets, that appear in a model have been executed at least once. Some of the metrics supported by Reactis involve only Simulink, some are specific to Stateflow, and the remaining are generic in the sense that they include targets within both the Simulink and the Stateflow portions of a model. When using Reactis for C Plugin, coverage targets are also tracked in the C code portions of models (S-Functions and Stateflow custom C code). The metrics discussed in this chapter may be visualized using Simulator and are central to test generation and model validation using Tester and Validator.
6.1 Simulink-Specific Metrics
The Simulink-specific metrics include the following: conditional subsystem coverage, branch coverage, and lookup table coverage.
6.1.1 Conditional Subsystem Coverage
A (conditional) subsystem is deemed covered if has been executed at least once in some simulation step. In general, every subsystem within a Simulink diagram executes during every simulation step. However, conditional subsystems may be disabled during a simulation step and hence not execute.
6.1.2 Branch Coverage
A number of different Simulink blocks contain targets included in the branch coverage measure. Each block in this group has the characteristic that the set of all possible outcomes of evaluating the block may be partitioned into well-defined sets of mutually exclusive outcomes. For example, the possible outcomes of evaluating a Logical Operator block are true or false. Therefore, for each Logical Operator block, we consider the true outcome and the false outcome to be the two branches associated with the block. Each branch becomes a coverage target in the branch-coverage metric.
The following discusses branch coverage in more detail. Reactis Tester aims to exercise all such branches in the test suites it generates. Reactis Simulator includes several ways to track the branches that have been exercised. In particular, in model diagrams, every block that includes branches is drawn in a way to show each branch. Simulator may be configured to render uncovered branches in red. We now list the blocks included in the branch-coverage metric; for each block, we describe its associated branches and how coverage information for the branches is drawn in Simulator.
6.1.3 Lookup Table Coverage
Reactis supports coverage tracking for the Lookup Table, Lookup Table (2-D) blocks, Pre-Lookup, Interpolation-Nd and some Lookup Table (n-D) blocks. Intuitively a coverage target will be allocated for each interval specified by the input settings of a table. To view the coverage information for a table in Simulator, after selecting Coverage -> Show Details, right-click on the table and select View Coverage Details.
If a Lookup Table block (one-dimensional) has inport u and “Vector of input values” [ u1, u2, ... , un ] then the block will have the following targets:
If a Lookup Table (2-D) block has inputs u1 and u2,
“Row index input values” [ r1, r2, ... , rm ], and
“Column index input values” [ c1, c2, ... , cn ] then the block
will have the following targets:
6.2 Stateflow-Specific Metrics
The Stateflow metrics are defined with respect to the graphical syntax of Stateflow. Among other things, Stateflow diagrams contain states, transition segments, and junctions. Each transition segment in turn may have a label that includes all or some of the following: an event, a condition, a condition action, and a transition action. A segment’s condition action is executed whenever the segment’s condition evaluates to true. A transition consists of a sequence of segments leading from one state to another. The transition will fire when the condition on each segment in the sequence evaluates to true. A segment’s transition action is executed only when it is included in such a firing transition. Highlighting of coverage in Stateflow diagrams is shown in Figure 6.12. Note that in addition to the Stateflow-specific metrics, Decision, Condition, MC/DC and MCC coverage targets may be associated with transition segments.
6.2.1 State Coverage
The targets in this metric are states; a state is covered if it has been entered at least once.
6.2.2 Condition Action Coverage
The targets in this metric are transition segments; a segment is deemed to be covered if its condition action has been executed at least once, or, if the segment has no condition action, if its condition has evaluated to “true” at least once. Note that a segment with an empty condition is assumed to have a condition that always evaluates to “true” when the segment is considered for inclusion in a firing transition during model execution.
6.2.3 Transition Action Coverage
The targets in this metric are transition segments that have transition actions; such a segment is deemed to be covered if its transition action has been executed at least once. Note that if a segment has no transition action, it is ignored by this metric.
6.2.4 Child State Exit via Parent Transition
Consider the Stateflow diagram in Figure 6.12. The state On has three child states: Inactive, Active, and Init. Child State Exit via Parent Transition (CSEPT) coverage tracks whether the transition from On to Off has fired while On is in each of its child states. So in this case there are three CSEPT targets. Namely the transition from On to Off fires when:
More precisely, a Stateflow state S has the following CSEPT targets:
If your model has states nested deeper than one level (e.g. Parent(S) has a parent for some state S) or a transition causing Parent(S) to exit has multiple transition segments, then you can adjust the definition of CSEPT targets with two settings that you specify in the Coverage tab of the Info File Editor (see Section 5.10). These settings specify which states and transitions will be paired with a state to form CSEPT targets. In the following:
The two settings you control indicate the following definitions for PairStates and ExitTrans:
These options yield a parameterized definition of CSEPT targets. A Stateflow state S has the following CSEPT targets:
6.3 Generic Metrics
Generic coverage metrics define targets which may appear in either the Simulink, the Stateflow, or the C code 2 portions of a model. There are four generic coverage metrics supported by Reactis: (1) Decision coverage, (2) Condition coverage, (3) Modified Condition/Decision Coverage (MC/DC), and (4) Boundary coverage. These four metrics are all are based on well-known coverage metrics of the same names developed for measuring coverage of source code.
6.3.1 Decision, Condition, MC/DC and MCC Metrics
Reactis includes facilities for generating tests to meet the Decision, Condition, Modified Condition/Decision Coverage (MC/DC) and Multiple Condition Coverage (MCC) requirements. Each of these metrics involve boolean-valued structures (conditions, decisions) within a model; understanding these metrics requires that the notions of conditions and decisions be defined precisely. As the notions were first developed in the context of traditional programming languages (e.g. C, Ada), the next paragraphs first review their traditional definitions. The adaptations of these notions to Simulink and Stateflow are then discussed.
In software testing, a decision is a boolean-valued expression used to determine which execution path to follow, and a condition is a boolean-valued subexpression of a decision which cannot be broken into smaller boolean subexpressions because it does not contain any boolean operators. Decisions are typically constructed by using boolean operators to combine several conditions. For example, consider the following statement from a C program:
if ((x > 0) && (y > 0)) z = 1; else z = 2;
Traditional Decision coverage may now be defined as follows. Each decision in a program gives rise to two targets: the evaluation of the decision to true, and to false; a program is fully covered by a test suite when each target has been covered, i.e. each decision has evaluated to both true and false. Condition coverage is defined very similarly; the difference is that each condition evaluates to both true and false, rather than each decision.
MC/DC is somewhat more complex to define. It was introduced by John J. Chilenski of Boeing in the early 90s; the definitive research paper was published by Chilenski and Steve Miller, of Rockwell-Collins, in 1994. MC/DC is the level of testing mandated by the Federal Aviation Administration (FAA) in its DO-178/B guidelines for the “most safety-critical” components of aviation software. The MC/DC targets in a program are the conditions; a condition C in decision D is covered by a test suite if there are two test steps X and Y (not necessarily consecutive) in the suite such that:
In other words, each condition must be shown to independently affect the outcome of its enclosing decision.
Multiple Condition Coverage (MCC, also known as Condition
Combination Coverage) tracks one target for each combination of
values to which the conditions in a decision evaluate. For a decision
with N conditions this will create 2N MCC targets. A condition
Short-circuiting of boolean expressions can affect the difficulty of
achieving full MC/DC or MCC coverage of a program. A short-circuited
operator avoids evaluating all conditions of a decision if the outcome
is determined after evaluating only a subset of the conditions. For
example, a short-circuited “and” operator would not evaluate its
second argument if the first evaluates to false. Virtually all
programming languages use short-circuiting for reasons of efficiency.
Without going into the technical details, it can be said that
achieving full MC/DC coverage is easier if short-circuited boolean
operators are used. For MCC coverage, short-circuited expressions
generate a reduced number of coverage targets compared to
non-short-circuited expressions. The
Section 4.8.1 explains how Reactis may be instructed to treat Simulink / Stateflow boolean operators as short-circuited.
In order to adapt the notions of Decision, Condition, MC/DC and MCC coverage to Simulink / Stateflow, it suffices to define what the conditions and decisions in a model are. These are as follows.
Coverage information for Decision, Condition, and MC/DC coverage is rendered as shown in Figure 6.13. Right-clicking on an outport of a Logical Operator block and selecting View Coverage Details gives additional information.
6.3.2 Boundary Value Coverage
Boundary Value Coverage tracks whether a data item assumes values of interest for a particular block. In the case of top-level model inports, test points, and configuration variables, these are the boundary values of the item’s domain of possible values. In the case of a Relational Operator block, the values of interest occur when the two inputs to the block are equal or very close.
The boundary values tracked for an inport, test point, or configuration variable are determined by its associated type constraint 3 as shown in Table 6.1. If an inport or configuration variable has a type not shown in the table, then it has no boundary value targets.
Reactis also tracks boundary value coverage for the Relational Operator block. For a relational operator block with integer inputs a and b, the boundary value targets are:
For a relational operator block with floating point inputs a and b, the boundary value targets are:
where reltol is the relative tolerance specified in the Coverage tab of the Info File Editor. Note that boundary value coverage for relational operators has the potential to introduce a large number of very hard to cover targets. Since some users might prefer to not track such targets, this functionality is disabled by default and can be enabled from the Coverage tab of the Info File Editor.
Reactis can be configured to track whether or not saturation has occurred for blocks which are set to saturate on integer overflow. Reactis will create up to three boundary value targets for each block with saturate on integer overflow enabled:
The tracking of integer saturation is controlled from the Coverage tab of the Info File Editor. When enabled, integer saturation is tracked for the following blocks:
6.4 Validator-Related Targets
See Chapter 9 for a description of the two Validator-related targets: assertions and user-defined targets.
6.5 Excluding Coverage Targets
Reactis supports three different ways to disable coverage tracking for a subset of targets. When a target is excluded using any of the three mechanisms, Reactis Tester will not try to generate a test to exercise the target and Reactis Simulator will not report the target as uncovered.
6.5.1 Disabling a Coverage Metric
In some cases, tracking a certain metric may not be desirable. For example, if a model contains no lookup tables, you may wish to not use lookup table coverage when working with the model in Reactis. Or, if accomplishing 100% MCC coverage requires an excessive number of tests, you may wish to not track that metric. To help you focus on the set of metrics of highest interest to you, Reactis lets you disable a whole coverage metric.
To disable a coverage metric, do the following:
Disabling a coverage metric will cause it to not be shown in any of the various places where Reactis lists coverage metrics, including the Tester launch and run dialogs, the Coverage Summary dialog, the Coverage Report Browser, and exported coverage reports. Also, Reactis Tester will not target disabled metrics when generating tests.
6.5.2 Excluding a Subsystem
Reactis lets you disable coverage tracking for all targets in a given subsystem of a model. For example, it may not be desirable to track coverage for a referenced model or library system if that portion of your model is being tested by other unit tests.
When Simulator is disabled, you can enable or disable coverage tracking for a subsystem by right-clicking on the subsystem in the Reactis hierarchy panel and selecting the Coverage Tracking entry which has the following sub-menus:
By default, coverage tracking is enabled for the top-level of the model and set to inherit for all other subsystems. As shown in Figure 6.16, visual cues in the hierarchy panel indicate the current coverage tracking setting for each subsystem. An X mark on the icon to the left of a subsystem name indicates that coverage tracking is disabled, whereas a check mark indicates that coverage tracking is enabled. If the icon has neither a check nor an X, then the subsystem is set to inherit its coverage tracking setting. If a subsystem currently has coverage tracking disabled or it inherits a disabled setting, then the subsystem name is grayed out (drawn in a lighter color).
Turning off coverage tracking for a subsystem has the following effects:
6.5.3 Excluding Individual Targets
Excluding individual targets from coverage tracking allows fine-grained control if a model includes a few specific targets that cannot be exercised. Excluding such targets from coverage tracking (after confirming their unreachable status) can help you achieve the goal of 100% coverage of reachable targets.
Reactis offers two alternative ways to exclude a coverage target. If you simply exclude a target, Reactis will not attempt to exercise the target when generating tests and it will not report it as covered or uncovered. The second way to exclude a target is to exclude and monitor a target. This method lets you assert that a target is unreachable and therefore should not be included in the coverage reports; but, Reactis will also monitor the target and alert you if the target is ever exercised – that is, you will be notified if your claim that the target is unreachable is incorrect. Reactis accomplishes this by automatically creating a Validator assertion for each excluded and monitored target to flag if the target is ever exercised. If such an assertion is violated (either while generating tests in Tester or running the model in Simulator), Reactis will report the violation and Simulator can be employed to investigate how the presumably unreachable target got covered.
All target types tracked by Reactis can be excluded. The method to exclude a target differs slightly by target type.
For Decision, Condition, MC/DC, MCC, CSEPT, Lookup table and input boundary value targets (see Figure 6.17):
For all other targets right-click on the block, Stateflow transition, or C or EML code fragment, select Track Coverage and select the appropriate target from the submenu (see Figure 6.18). A check-mark in the submenu listing coverage targets indicates that coverage is currently being tracked for that target.
Selecting Track Coverage brings up the dialog shown in Figure 6.19. This dialog offers three choices for the coverage exclusion status: