lakos
is a command line tool and library that can:
- Visualize internal Dart library dependencies in Graphviz
dot
. - Detect dependency cycles.
- Identify orphans.
- Compute useful dependency graph metrics.
Usage: lakos [options] <root-directory>
-f, --format=<FORMAT> Output format.
[dot (default), json]
-o, --output=<FILE> Save output to a file instead of printing it.
(defaults to "STDOUT")
--[no-]tree Show directory structure as subgraphs.
(defaults to on)
-m, --[no-]metrics Compute and show global metrics.
(defaults to --no-metrics)
--[no-]node-metrics Show node metrics. Only works when --metrics is true.
(defaults to --no-node-metrics)
-i, --ignore=<GLOB> Exclude files and directories with a glob pattern.
(defaults to "!**")
--[no-]cycles-allowed With --no-cycles-allowed lakos runs normally
but exits with a non-zero exit code
if a dependency cycle is detected.
Useful for CI builds.
(defaults to --no-cycles-allowed)
Print dot graph for the current directory (not very useful):
lakos .
Pass output directly to Graphviz dot
in one line (much more useful):
lakos . | dot -Tpng -Gdpi=200 -o example.png
Save output to a dot file first and then generate an SVG using Graphviz dot
:
lakos -o example.dot .
dot -Tsvg example.dot -o example.svg
- Nodes are internal Dart libraries. Edges are "uses" or "depends on" relationships.
- Exports are drawn with a dashed edge.
- Orphan nodes are octagons (with
--metrics
). - Only
import
andexport
directives are supported;library
andpart
are not. - Only libraries under the root-directory are shown; Dart core and external packages are not.
lakos
run on itself with metrics, ignoring tests.
lakos -o dot_images/lakos.metrics_no_test.dot -m -i test/** .
Show node metrics.
lakos -o dot_images/args.no_test_node_metrics.dot -m -i test/** --node-metrics /root/.pub-cache/hosted/pub.dartlang.org/args-1.6.0
No directory tree.
lakos --no-tree -o dot_images/string_scanner.no_test_no_tree.dot -i test/** /root/.pub-cache/hosted/pub.dartlang.org/string_scanner-1.0.5
Example JSON output:
lakos -f json -o dot_images/pub_cache.metrics_no_test.json -m -i test/** /root/.pub-cache/hosted/pub.dartlang.org/pub_cache-0.2.3
Click to show the JSON output.
{
"rootDir": "/root/.pub-cache/hosted/pub.dartlang.org/pub_cache-0.2.3",
"nodes": {
"/example/list.dart": {
"id": "/example/list.dart",
"label": "list",
"cd": 3,
"inDegree": 0,
"outDegree": 1,
"instability": 1.0,
"sloc": 14
},
"/lib/pub_cache.dart": {
"id": "/lib/pub_cache.dart",
"label": "pub_cache",
"cd": 2,
"inDegree": 2,
"outDegree": 1,
"instability": 0.33,
"sloc": 162
},
"/lib/src/impl.dart": {
"id": "/lib/src/impl.dart",
"label": "impl",
"cd": 2,
"inDegree": 1,
"outDegree": 1,
"instability": 0.5,
"sloc": 95
}
},
"subgraphs": [
{
"id": "",
"label": "pub_cache-0.2.3",
"nodes": [],
"subgraphs": [
{
"id": "/example",
"label": "example",
"nodes": ["/example/list.dart"],
"subgraphs": []
},
{
"id": "/lib",
"label": "lib",
"nodes": ["/lib/pub_cache.dart"],
"subgraphs": [
{
"id": "/lib/src",
"label": "src",
"nodes": ["/lib/src/impl.dart"],
"subgraphs": []
}
]
}
]
}
],
"edges": [
{
"from": "/example/list.dart",
"to": "/lib/pub_cache.dart",
"directive": "import"
},
{
"from": "/lib/pub_cache.dart",
"to": "/lib/src/impl.dart",
"directive": "import"
},
{
"from": "/lib/src/impl.dart",
"to": "/lib/pub_cache.dart",
"directive": "import"
}
],
"metrics": {
"isAcyclic": false,
"firstCycle": ["/lib/pub_cache.dart", "/lib/src/impl.dart", "/lib/pub_cache.dart"],
"numNodes": 3,
"numEdges": 3,
"avgDegree": 1.0,
"orphans": [],
"ccd": 7,
"acd": 2.33,
"nccd": 1.4,
"totalSloc": 271,
"avgSloc": 90.33
}
}
Use curly brackets in the glob pattern to ignore multiple folders:
lakos -i "{lib/extensions/**,test/**}" .
You may override the default style attributes directly through the Graphviz dot
command line arguments. In Graphviz, the style attributes are divided into 3 sections: graph
, node
, and edge
. To override a specific attribute, use the -G
, -N
, and -E
prefix, respectively. For example, -Nfillcolor=white
will set the node fill color to white. All the Graphviz style attributes can be found here.
Example of left to right layout.
dot -Tpng dot_images/test.lr.dot -Grankdir=LR -Gdpi=200 -o dot_images/test.lr.png
Gradient node color.
dot -Tpng dot_images/string_scanner.gradient.dot -Gdpi=200 -Nfillcolor=steelblue2:steelblue4 -Nfontcolor=white -Ngradientangle=270 -o dot_images/string_scanner.gradient.png
Still can't get enough graph visualization goodness? Try this!
lakos -i test/** /root/.pub-cache/hosted/pub.dartlang.org/test-1.15.3 | gv2gml -o dot_images/test.gml
gv2gml
converts dot format into GML format, which you can open in yEd, Gephi, Cytoscape, etc.
Use --metrics
to compute and show these.
isAcyclic: True if the library dependencies form a directed acyclic graph (DAG). False if the graph contains dependency cycles. True is better.
firstCycle: The first dependency cycle found if the graph is not acyclic. Available in the JSON output.
numNodes: Number of Dart libraries (nodes) in the graph.
numEdges: Number of edges (dependencies) in the graph.
avgDegree: The average number of incoming/outgoing edges per node. Average degree (directed graph) = numEdges / numNodes. This metric can compare graphs of different sizes. Similar to the ACD, the average degree can be interpreted as the average number of nodes that will need to change when one node changes. Lower is better.
numOrphans: Number of Dart libraries that have no imports and are imported nowhere. (inDegree and outDegree are 0.)
ccd: The Cumulative Component Dependency (CCD) is the sum of all Component Dependencies. The CCD can be interpreted as the total "coupling" of the graph. Lower is better.
acd: The Average Component Dependency (ACD). ACD = CCD / numNodes. Similar to the avgDegree, the ACD can be interpreted as the average number of nodes that may need to change when one node changes. Lower is better.
nccd: The Normalized Cumulative Component Dependency. This is the CCD divided by a CCD of a binary tree of the same size. This metric can compare graphs of different sizes. If the NCCD is below 1.0, the graph is "horizontal". If the NCCD is above 1.0, the graph is "vertical". If the NCCD is above 2.0, the graph probably contains cycles. Lower is better.
totalSloc: Total Source Lines of Code for all nodes.
avgSloc: The average SLOC per node. The average SLOC should arguably be kept within some Goldilocks range: not too big and not too small.
Use --metrics --node-metrics
to show these.
cd: The Component Dependency (CD) is the number of nodes a particular node depends on directly or transitively, including itself.
inDegree: The number of nodes that depend on this node. (Number of incoming edges.)
outDegree: The number of nodes this node depends on. (Number of outgoing edges.)
instability: A node metric by Robert C. Martin related to the Stable-Dependencies Principle: Depend in the direction of stability. Instability = outDegree / (inDegree + outDegree). This yields a value from 0 to 1, where 0 means 100% stable and 1 means 100% unstable. In general, node instability should decrease in the direction of dependency. In other words, lower level nodes should be more stable and more reusable than higher level nodes.
sloc: Source Lines of Code is the number of lines of code ignoring comments and blank lines. Note: my SLOC counter function doesn't deal correctly with comments inside multi-line strings or multi-line comments in the middle of code.
Use these in your CI pipeline, especially DependencyCycleDetected.
- Ok
- InvalidOption
- NoRootDirectorySpecified
- BuildModelFailed
- WriteToFileFailed
- DependencyCycleDetected
In addition to the command line tool, lakos
can also be used as a library.
Use buildModel
to construct a Model
object.
Use Model.getOutput
to print the model in dot or JSON format.
Use Model.toDirectedGraph
for further analysis with the directed_graph
library.
See example/example.dart.
lakos
is named after John Lakos, author of Large-Scale C++ Software Design. This book presents:
- A graph-theoretic argument for keeping module dependencies acyclic.
- A set of "levelization techniques", such as Escalation and Demotion, to help avoid cyclic dependencies.
- A set of coupling metrics, such as the CCD, ACD, and the NCCD.
- An emphasis on physical design (files, folders) working in harmony with logical design (functions, classes).
- Dart: pubviz (for packages)
- C/C++: cinclude2dot
- JavaScript/TypeScript: madge, dependency-cruiser
- C#: NDepend
- Java: Sonargraph, JArchitect
- Python: pydeps
- Haskell: graphmod
- Swift: SwiftAlyzer