code_metrics

The purpose of this tool is to assess qualities of source code that may predict negative aspects such as entity coupling, cohesion, complexity, error-proneness, and overall maintainability. It is meant to be extensible via the addition of objects implementing new metrics.

This tool provides predicates for computing metrics for source files, entities, libraries, files, and directories. The actual availability of a particular predicate depends on the specific metric. A set of predicates prints, by default, the computed metric values to the standard output. A second set of predicates computes and returns a score (usually a compound term with the computed metric values as arguments).

API documentation

This tool API documentation is available at:

../../apis/library_index.html#code-metrics

Loading

This tool can be loaded using the query:

| ?- logtalk_load(code_metrics(loader)).

Testing

To test this tool, load the tester.lgt file:

| ?- logtalk_load(code_metrics(tester)).

Available metrics

Currently, the following metrics are provided:

  • Number of Clauses (noc_metric)

  • Number of Rules (nor_metric)

  • Unique Predicate Nodes (upn_metric)

  • Cyclomatic Complexity (cc_metric)

  • Depth of Inheritance (dit_metric)

  • Efferent coupling, afferent coupling, instability, and abstractness (coupling_metric)

  • Lack of cohesion of methods (lcom_metric)

  • Weighted Methods per Class (wmc_metric)

  • Response For a Class (rfc_metric)

  • Cognitive complexity (cogc_metric)

  • Documentation (doc_metric)

  • Lines of code, comments, and blanks (lines_metric)

  • Source code size (size_metric)

  • Maintainability index (mi_metric and mi_metric(Stroud))

  • Halstead complexity (halstead_metric and halstead_metric(Stroud))

A helper object, code_metrics, is also provided, allowing running all loaded individual metrics. For code coverage metrics, see the lgtunit tool documentation.

Coupling metrics

  • Efferent coupling (Ce): Number of entities that an entity depends on. These include objects receiving messages from the entity, plus the implemented protocols, imported categories, and extended/instantiated/specialized objects.

  • Afferent coupling (Ca): Number of entities that depend on an entity. For a protocol, the number of protocols that extend it, plus the number of objects and categories that implement it. For a category, the number of objects that import it. For an object, the number of categories and objects that send messages to it plus the number of objects that extend/instantiate/specialize it.

  • Instability: Computed as Ce / (Ce + Ca). Measures an entity resilience to change. Ranging from 0.0 to 1.0, with 0.0 indicating a maximally stable entity and 1.0 indicating a maximally unstable entity. Ideally, an entity is either maximally stable or maximally unstable.

  • Abstractness: Computed as the ratio between the number of static predicates with scope directives without a local definition and the number of static predicates with scope directives. Measures the rigidity of an entity. Ranging from 0.0 to 1.0, with 0.0 indicating a fully concrete entity and 1.0 indicating a fully abstract entity.

  • Distance from main sequence: Computed as abs(A + I - 1). Measures how far an entity is from the idealized line A + I = 1 (the “main sequence”). Ranging from 0.0 to 1.0, with 0.0 indicating the entity sits exactly on the main sequence and 1.0 indicating maximum deviation. Entities in the “zone of pain” (A =~ 0, I =~ 0: concrete and stable) are difficult to change; entities in the “zone of uselessness” (A =~ 1, I =~ 1: abstract and unstable) are not used. Ideally, entities should have a distance close to 0.0.

The dependencies count includes direct entity relations plus predicate calls or dynamic updates to predicates in external objects or categories.

For more information on the interpretation of the coupling metric scores, see, e.g., the original paper by Robert Martin:

@inproceedings{citeulike:1579528,
    author = "Martin, Robert",
    booktitle = "Workshop Pragmatic and Theoretical Directions in Object-Oriented Software Metrics",
    citeulike-article-id = 1579528,
    citeulike-linkout-0 = "http://www.objectmentor.com/resources/articles/oodmetrc.pdf",
    keywords = "diplomarbeit",
    organization = "OOPSLA'94",
    posted-at = "2007-08-21 11:08:44",
    priority = 0,
    title = "OO Design Quality Metrics - An Analysis of Dependencies",
    url = "http://www.objectmentor.com/resources/articles/oodmetrc.pdf",
    year = 1994
}

The coupling metric was also influenced by the metrics rating system in Microsoft Visual Studio and aims to eventually emulate the functionality of a maintainability index score.

Lack of cohesion metric

The lack of cohesion of methods (LCOM4) metric is computed as the number of connected components in the undirected graph whose nodes are the locally defined (non-auxiliary) predicates of an object or category and whose edges represent direct internal calls between those predicates. The metric is reported as the compound term lcom(Components, Predicates) where Components is the number of connected components and Predicates is the total number of locally defined predicates.

An lcom(1, N) score indicates a fully cohesive entity: all predicates are reachable from any other predicate via internal calls. Higher Components values suggest the entity may benefit from being split into separate, more focused entities. Entities with no predicates or a single predicate always score lcom(1, N).

Protocols are not scored by this metric (they never define predicate clauses). Only direct bare-predicate calls (as reported by the Logtalk reflection API) are counted as internal edges; external message-sends (Object::Predicate) are excluded.

For more details on the LCOM4 variant, see:

@inproceedings{Hitz:1995,
    author = "Hitz, M. and Montazeri, B.",
    title = "Measuring Coupling and Cohesion in Object-Oriented Systems",
    booktitle = "Proceedings of the International Symposium on Applied Corporate Computing",
    year = 1995
}

WMC metric

The Weighted Methods per Class (WMC) metric counts the number of locally defined (non-auxiliary) predicates in an object or category. Unit weights are used: each predicate contributes 1 to the score regardless of its complexity. Protocols are not scored as they cannot define predicates.

A cyclomatic-complexity-weighted variant is not provided. The Logtalk reflection API exposes inter-predicate call edges but not intra-clause branching structure (if-then-else, disjunction). A CC-weighted WMC would inherit the same approximation limitations already present in cc_metric and would also be numerically equivalent to the number of user clauses reported by noc_metric.

For more details on the WMC metric, see the original CK paper:

@article{Chidamber:1994,
    author = "Chidamber, S. R. and Kemerer, C. F.",
    title = "A Metrics Suite for Object Oriented Design",
    journal = "IEEE Transactions on Software Engineering",
    volume = 20,
    number = 6,
    pages = "476--493",
    year = 1994
}

RFC metric

The Response For a Class (RFC) metric is computed as the size of the response set for an object or category. The response set RS(E) for an entity E is the union of its locally defined (non-auxiliary) predicates and all distinct predicates directly called or updated by those predicates. RFC = |RS(E)|.

Internal calls (to predicates within the entity) are included in the response set but are not double-counted: they are already part of the locally defined predicates. Only additional external callees increase the score beyond the WMC score of the same entity.

When the receiver of a message is only bound at runtime (e.g. Obj::foo/1 where Obj is a variable), all such calls for the same predicate indicator are counted as one callee entry, regardless of how many distinct call sites exist.

Calls to built-in predicates are never returned by the Logtalk reflection API and therefore do not contribute to the score.

Protocols are not scored as they cannot define predicates.

Higher RFC scores indicate entities with larger potential execution footprints, which increases testing effort.

For more details on the RFC metric, see the original CK paper:

@article{Chidamber:1994,
    author = "Chidamber, S. R. and Kemerer, C. F.",
    title = "A Metrics Suite for Object Oriented Design",
    journal = "IEEE Transactions on Software Engineering",
    volume = 20,
    number = 6,
    pages = "476--493",
    year = 1994
}

Cognitive complexity metric

The cognitive complexity (cogc_metric) metric is an approximation of the Cognitive Complexity metric proposed by Campbell (2018). For each non-auxiliary predicate, the score contribution is the number of extra clauses (number of clauses minus one, representing multi-clause branching choices) plus one if the predicate is directly recursive (i.e. calls itself within the same entity). The entity score is the sum of all predicate contributions.

A score of 0 means all predicates are single-clause and non-recursive. Higher scores indicate entities with more branching and recursion, which typically require more effort to understand and test.

Approximation note: the Logtalk reflection API does not expose the use of control constructs such as cuts, if-then-else, disjunction, or catch/3 in the predicate clause bodies. Thus, these constructs cannot contribute to the score, which should be interpreted as a lower bound for the cognitive complexity of the entity.

Protocols are not scored as they cannot define predicates.

Lines metric

The lines metric (lines_metric) computes the number of code lines, comment lines, and blank lines, represented as the compound term lines(Code, Comments, Blanks).

The implementation calls the external cloc tool and parses its CSV report. For entity scores, it first obtains the entity source line range from the reflection API using the lines(BeginLine, EndLine) property, writes only that range to a temporary file, and runs cloc on that temporary file.

For file, directory, and library scores, the metric computes and sums per-file lines/3 scores.

This metric requires the cloc and sed commands to be available in the operating system path.

Maintainability index metric

The maintainability index metric (mi_metric) computes the original maintainability index formula:

MI = 171 - 5.2 * ln(V) - 0.23 * C - 16.2 * ln(L)

where V is the Halstead volume, C is cyclomatic complexity, and L is the number of code lines. The score is represented as mi(Index).

The metric is available as mi_metric (default Stroud coefficient 18) and as mi_metric(Stroud) for a user-defined Stroud coefficient passed to the Halstead metric.

Protocols are not scored as they cannot define predicates.

This metric requires the lines metric and thus the cloc and sed commands to be available in the operating system path.

Halstead metric

Predicates declared, user-defined, and called are interpreted as operators. Built-in predicates and built-in control constructs are ignored. Predicate arguments are abstracted, assumed distinct, and interpreted as operands. Note that this definition of operands is a significant deviation from the original definition, which used syntactic literals. A computation closer to the original definition of the metric would require switching to use the parser to collect information on syntactic literals, which would imply a much larger computation cost. The number of predicate calls doesn’t include calls to built-in predicates and can underestimate recursive calls.

The computation of this metric is parameterized by the Stroud coefficient for computing the time required to program (default is 18). The following individual measures are computed:

  • Number of distinct predicates (declared, defined, called, or updated; Pn).

  • Number of predicate arguments (assumed distinct; PAn).

  • Number of predicate calls/updates + number of clauses (Cn).

  • Number of predicate call/update arguments + number of clause head arguments (CAn).

  • Entity vocabulary (EV). Computed as EV = Pn + PAn.

  • Entity length (EL). Computed as EL = Cn + CAn.

  • Volume (V). Computed as V = EL * log2(EV).

  • Difficulty (D). Computed as D = (Pn/2) * (CAn/An).

  • Effort (E). Computed as E = D * V.

  • Time required to program (T). Computed as T = E/k seconds (where k is the Stroud number; defaults to 18).

  • Number of delivered bugs (B). Computed as B = V/3000.

UPN metric

The Unique Predicate Nodes (UPN) metric is described in the following paper:

@article{MOORES199845,
    title = "Applying Complexity Measures to Rule-Based Prolog Programs",
    journal = "Journal of Systems and Software",
    volume = "44",
    number = "1",
    pages = "45 - 52",
    year = "1998",
    issn = "0164-1212",
    doi = "https://doi.org/10.1016/S0164-1212(98)10042-0",
    url = "http://www.sciencedirect.com/science/article/pii/S0164121298100420",
    author = "Trevor T Moores"
}

The nodes include called and updated predicates independently of where they are defined. It also includes multifile predicates contributed to other entities.

Cyclomatic complexity metric

The cyclomatic complexity metric evaluates an entity code complexity by measuring the number of linearly independent paths through the code. In its current implementation, all defined predicates that are not called or updated are counted as graph-connected components (the reasoning being that these predicates can be considered entry points). The implementation uses the same predicate abstraction as the UPN metric. The defined predicates include multifile predicate definitions contributed by the entity to other entities.

For more details on this metric, see the original paper by Thomas J. McCabe:

@inproceedings{McCabe:1976:CM:800253.807712,
    author = "McCabe, Thomas J.",
    title = "A Complexity Measure",
    booktitle = "Proceedings of the 2Nd International Conference on Software Engineering",
    series = "ICSE '76",
    year = 1976,
    location = "San Francisco, California, USA",
    pages = "407--",
    url = "http://dl.acm.org/citation.cfm?id=800253.807712",
    acmid = 807712,
    publisher = "IEEE Computer Society Press",
    address = "Los Alamitos, CA, USA",
    keywords = "Basis, Complexity measure, Control flow, Decomposition, Graph theory, Independence, Linear, Modularization, Programming, Reduction, Software, Testing",
}

Usage

All metrics require the source code to be analyzed to be loaded with the source_data flag turned on. For usage examples, see the SCRIPT.txt file in the tool directory.

Be sure to fully understand the metrics individual meanings and any implementation limitations before using them to support any evaluation or decision process.

Excluding code from analysis

A set of options is available to specify code that should be excluded when applying code metrics:

  • exclude_directories(Directories)
    list of directories to exclude (default is []); all sub-directories of the excluded directories are also excluded; directories may be listed by full or relative path
  • exclude_files(Files)
    list of source files to exclude (default is []); files may be listed by full path or basename, with or without extension
  • exclude_libraries(Libraries)
    list of libraries to exclude (default is [startup, scratch_directory])
  • exclude_entities(Entities)
    list of entities to exclude (default is [])

Defining new metrics

New metrics can be implemented by defining an object that imports the code_metric category and implements its score predicates. There is also a code_metrics_utilities category that defines useful predicates for the definition of metrics.

Third-party tools

The following open-source command-line programs can count blank lines, comment lines, and lines of source code in many programming languages, including Logtalk:

Applying metrics to Prolog modules

Some of the metrics can also be applied to Prolog modules that Logtalk is able to compile as objects. For example, if the Prolog module file is named module.pl, try:

| ?- logtalk_load(module, [source_data(on)]).

Due to the lack of standardization of module systems and the abundance of proprietary extensions, this solution is not expected to work for all cases.

Applying metrics to plain Prolog code

Some of the metrics can also be applied to plain Prolog code. For example, if the Prolog file is named code.pl, simply define an object including its code:

:- object(code).
    :- include('code.pl').
:- end_object.

Save the object to an e.g. code.lgt file in the same directory as the Prolog file and then load it in debug mode:

| ?- logtalk_load(code, [source_data(on)]).

In alternative, use the object_wrapper_hook provided by the hook_objects library:

| ?- logtalk_load(hook_objects(loader)).
...

| ?- logtalk_load(code, [hook(object_wrapper_hook), source_data(on)]).

With either wrapping solution, pay special attention to any compilation warnings that may signal issues that could prevent the plain Prolog code from working when wrapped by an object.