Breadcrumbs adds an additional feature to the go compiler which allows the user to track which inputs the program influence which outputs.
Breadcrumbs extends go version 1.20.5. Mark functions which you want use breadcrumbs on with the //go:dataflow compiler directive to enable dataflow analysis for that function.
All documentation for compiling and developing the standard go compiler applies to breadcrumbs. Run ./make.bash in the src folder to compile the linux version. The binaries will appear in the bin directory.
Dataflow analysis requires sources and sinks to work. Data source can be marked using the runtime.DfMark(variableToMark any) int function. It returns an integer which specifies the index of the source in the dataflow bitmap. Store this to later analyze the sinks.
The dataflow bitmap of a variable can be obtained using the runtime.DfInspect(variable any) int64 function. The return value is the dataflow bitmap.
A maximum of 63 dataflows can be tracked in a single run. To track more, run the same code multiple times with different variables marked.
Most of the breadcrumbs code lives in src/cmd/compile/internal/dataflow/dataflow.go and src/cmd/compile/internal/ssa/dataflow_instrument.go. Additional space for dataflow is allocated in the runtime.newobject function and in src/cmd/compile/internal/types/size.go.
To debug a function dataflow, use GOSSAFUNC=functionname along with the usual go commands to dump the SSA of the function with and without dataflow instrumentation. Use the runtime functions runtime.GetDfArr() []int64 and runtime.GetBlockDfArr() int64 to obtain slices which represent the dataflow array and control-flow based dataflow arrays for a specific function.
Not all language features are supported. The project is under development.