CFG construction introduces wrong feature annotations on functions

[clhunsen]

In joint work with Stefan from Vienna and Thomas (@bockthom), we discovered that there is a bug in the --dumpcfg routine of TypeChef.

Consider following input file with an empty feature model:

void bar() {
}
void foo() {
#ifdef X
    bar();
#endif
}

As a sidenote, TypeChef (on 083e742, most recent commit on master) is called like this:

./typechef.sh /tmp/testsystem/foo4.c -o /tmp/testsystem/foo4.c --bdd --featureModelFExpr /tmp/testsystem/fm.txt --dumpcfg --lexNoStdout

From the example file, it is clear that the node for the function declaration of foo in the resulting control-flow graph is NOT annotated with a feature. Nevertheless, there is a feature annotation in the resulting graph:

N;21;function;2;bar;1
E;21;21;1
E;21;21;1
N;42;statement;9;bar()::foo;definedEx(X)
E;42;21;definedEx(X)
N;63;function;7;foo;definedEx(X)
E;42;63;1
E;63;42;definedEx(X)
E;63;63;1

As you can see on the bold line, the function foo is annotated with definedEx(X) – instead of 1.
The other nodes and all the edges are fine, however.

---

I guess, the error results from a non-flushed temporary data structure. At least, it sounds like something of this kind.
Hopefully, someone, who is familiar with the code, can help on this problem.

Thanks in advance.

[joliebig]

No, the error is caused by a bug in the dot-writer implementation. The annotation of the Stmt 8: bar() was used when writing node Function 6: foo.

The edge Function 6: foo -> Function 6: foo is wrong, too. It should be ! defined(x). Fixed this one also.

[clhunsen]

Thanks for the fixes. I ran a test with the file, here the results:

N;21;function;1;bar;1
E;21;21;1
E;21;21;1
N;42;statement;8;bar()::foo;definedEx(X)
E;42;21;definedEx(X)
N;63;function;6;foo;1
E;42;63;definedEx(X)
E;63;42;definedEx(X)
E;63;63;!definedEx(X)

The bold parts have changed. Looks good to me now.