MMory
commented
3 years ago I agree that this is fragile. So far we have not had problems with that, which I don't mean as an excuse not to do it right.
Open blipper opened 3 years ago
MMory
commented
3 years ago I agree that this is fragile. So far we have not had problems with that, which I don't mean as an excuse not to do it right.
blipper
commented
3 years ago One example is the differences than result between stdlibc++ and libcxx
Is your feature request related to a problem? Please describe.
A typical test pattern in unittets is TEST_F(IFDSTaintAnalysisTest, TaintTest_04) { initialize({PathToLlFiles + "dummy_source_sink/taint_04_cpp_dbg.ll"}); IFDSSolver_P TaintSolver(*TaintProblem);
TaintSolver.solve();
map<int, set> GroundTruth;
GroundTruth[19] = set{"18"};
GroundTruth[24] = set{"23"};
compareResults(GroundTruth);
}
Specifying individual IDs is very fragile since it depends on how clang compiles and a single change can break everything.
Describe the solution you'd like A more flexible/less fragile pattern
e.g. specify the instruction pattern (with filters/qulifiers) %kIVSize = alloca i32, align 4, !psr.id !35 | ID: 7
Something like IDLookup.Builder.IRpattern("%kIVSize = alloca i32, align 4").inFunction("main").afterIRpattern("%p = alloca %struct.Params").build().getID();