ClangSA

How to run on julia

First, create a compilation database. The easiest way to do so is to use the intercept-build that comes with clang (in clang/tools/scan-build-py/bin):

cd julia
make clean
intercept-build make -j4

This should have created a compile_commands.json file. Then, from julia, run:

srcdir = joinpath(JULIA_HOME, "../../src"))
paths = filter(path->(endswith(path,".c") || endswith(path,".cpp")),
  readdir(srcdir)
database = ClangSA.loadDatabase(srcdir)
ClangSA.runTool(database, map(p->joinpath(srcdir, p), paths), ClangSA.getAnalysisAction())

Building as part of an LLVM build

Simply symlink the GCChecker.cpp file from this package's src/ directory into clang/lib/StaticAnalyzer/Checkers and add it to the CMakeLists.txt file in the same directory. You can then run clang-check -extra-arg="-Xanalyzer" -extra-arg="-analyzer-checker=julia.GCChecker" -extra-arg -Xclang -extra-arg -analyzer-output=text -analyze src/*{.c,.cpp}

You will also need to add the following at the end of include/clang/StaticAnalyzer/Checkers/Checkers.td:

 //===----------------------------------------------------------------------===//
 // Julia Checkers
 //===----------------------------------------------------------------------===//
 def Julia : Package<"julia">;
 let ParentPackage = Julia in {

 def GCChecker : Checker<"GCChecker">,
    HelpText<"Checks invariants of the Julia GC API.">,
    DescFile<"GCChecker.cpp">;
 } // end "julia"

Development tips

This package is designed for rapid iteration. The package can be reloaded, and rerun by repeating the last line above (though you'll have to re-run getAnalysisAction to make sure to pick up the changed code);