Have a better code coverage

[dawagner]

The current code coverage is . We can make it better by:

• removing dead code
• refactor
• more tests

Let's analyse the code coverage report to identify dead code and untested parts:

libparameter

• Unused code related to the Element tree not being strongly-typed. E.g. some classes
• logValue (in several classes): only used by the remote interface; maybe it shouldn't be a method.
• lots of to/from blackboard methods (both to/from user and to/from hardware)

ParameterMgr

• lots of remote command handlers code; has implications on other classes, e.g. ConfigurableDomains

Subsystem

• showMapping on formatted subsytem object (here, here and here).

Element Handle

• to/from user as different types (here; it also impacts other classes)

Fixed Point

• convert fixed point from decimal
• out-of-range test in raw space (here and here)
• to/from user value as double (element handle interface not tested for FixedPoint)

Integer

• to/from user value as double
• toPlainInteger (only used in the alsa plugin)

ParameterType

• setUnit (nobody uses it)
• "Unit" attributes in XML files (here and here)

ArrayParameter

• access as several types (see the general comment on libparameter)

BitParameter / BitwiseAreaConfiguration

• bit parameter as area configuration (here and here)
• impacts AreaConfiguration as well.

FormattedSubsystemObject

• not tested at all

SubsystemObject

• some stuff only used by plugin X or Y (e.g. here and here)

DomainConfiguration

• validation: configurations never seem to be invalid (here and here)
• merging

Floats

• serialization
• value space de/serialization

CriterionRule

• parse errors (e.g. here and here)

Linear and Log Adaptations

not tested at all:

• base class
• linear
• log

Element god class

• "Description" attribute
• non-const findChildOfKind

Criterion

• methods used by the client:
  • hasBeenModified
  • getCriterionName

Mapping

• methods used by plugins (getItem, getItemAsInteger)

Simulated BackSynchronizer

not tested at all

BitParameterBlockType

• serialization

Default builder

• mechanism enabled but never actually tested (could it be dead code?)

SystemClass

• non-empty plugin location
• needResync never tested to return "true"

Subsystem Object Creator

• virtual dtor never called? Should be called by Subsystem's dtor, though.

libxmlserializer

• XmlUtil.cpp not analyzed...
• The rest of misses are only error cases

libremote-processor

The coverage misses only involve error cases except a function only used by a non-tested remote command.

utility

• convertToVia (conversion to int8/uint8)

Summary

In subjective, ascending order of criticality.

• lots of dead code due to the weak typing of the Element tree (e.g. virtual methods that have to be implemented but are never actually used, such as getKind());
• a lot of errors cases are not covered (probably the largest source of coverage misses);
• most of remote commands aren't tested (ParameterMgr.cpp);
• some methods are only used by plugins and aren't tested; likewise, some client methods are not tested;
• some features are not tested at all (e.g. simulation mode, parameter adaptation) or poorly tested (e.g. showMapping);
• a large amount of parameter getters/setters (on one hand, between the user and the blackboard; on the other hand, between the blackboard and the subsystems) are not tested.

[krocard]

Maybe we should add -fno-inline -fno-inline-small-functions -fno-default-inline to improve method coverage accuracy. I am not sure they are needed though, as we are compiling in -O0

---

FormattedSubsystemObject

FormatedSubsystemObject is used by nearly all plugins. Maybe we should modify the test subsystem to use it.

---

rogue parameters/parameter handle in parameter/ParameterHandle.cpp

get and set of rogue parameters should be made template. The set and get code is duplicated everywhere.

---

Anyway, templating can reduce line count

True method that are not called are not instantiated.