Domain specific language for querying code property graphs
Domain specific language for querying code property graphs
This is the API reference for the CPG query language, a language to mine code for defects and vulnerabilities both interactively on a code analysis shell (REPL), or using non-interactive scripts.
Queries written in the CPG query language express graph traversals (see https://en.wikipedia.org/wiki/Graph_traversal). Similar to the standard graph traversal language "Gremlin" (see https://en.wikipedia.org/wiki/Gremlin_(programming_language))) these traversals are formulated as sequences of primitive language elements referred to as "steps". You can think of a step as a small program, similar to a unix shell utility, however, instead of processing lines one by one, the step processes nodes of the graph.
All traversals begin by selecting a set of start nodes, e.g.,
cpg.method
will start the traversal at all methods, while
cpg.local
will start at all local variables. The complete list of starting points can be found at
io.shiftleft.codepropertygraph.Cpg
Queries are lazily evaluated, e.g., cpg.method creates a traversal which you can add more
steps to. You can, for example, evaluate the traversal by converting it to a list:
cpg.method.toList
Since toList is such a common operation, we provide the shorthand l, meaning that
cpg.method.l
provides the same result as the former query.
Nodes have "properties", key-value pairs where keys are strings and values are primitive data types such as strings, integers, or Booleans. Properties of nodes can be selected based on their key, e.g.,
cpg.method.name
traverses to all method names. Nodes can also be filtered based on properties, e.g.,
cpg.method.name(".*exec.*")traverse to all methods where name matches the regular expression ".*exec.*". You can see
a complete list of properties by browsing to the API documentation of the corresponding step.
For example, you can find the properties of method nodes at
io.shiftleft.semanticcpg.language.types.structure.MethodTraversal.
Useful if you want to mutate something outside the traversal, or simply debug it:
This prints all typeDecl names as it traverses the graph and increments i for each one.
var i = 0 cpg.typeDecl.sideEffect{typeTemplate => println(typeTemplate.name); i = i + 1}.exec
If you are interested in multiple things along the way of your traversal, you label anything
using the as modulator, and use select at the end. Note that the compiler automatically
derived the correct return type as a tuple of the labelled steps, in this case with two elements.
cpg.method.as("method").definingTypeDecl.as("classDef").select.toList // return type: List[(Method, TypeDecl)]
You can always start a new traversal from a node, e.g.,
val someMethod = cpg.method.head
someMethod.start.parameter.toListYou can use this e.g. in a for comprehension, which is (in this context) essentially an alternative way to select multiple intermediate things. It is more expressive, but more computationally expensive.
val query = for { method <- cpg.method param <- method.start.parameter } yield (method.name, param.name) query.toList
This pass has MethodStubCreator and TypeDeclStubCreator as prerequisite for language frontends which do not provide method stubs and type decl stubs.