search

correctcomputation / checkedc-clang

This is the primary development repository for 3C, a tool for automatically converting legacy C code to the Checked C extension of C, which aims to enforce spatial memory safety. This repository is a fork of Checked C's.
14 stars 5 forks source link

Should `DeclRewriter::buildItypeDecl` use the actual internal `PVConstraint`? #704

Open mattmccutchen-cci opened 3 years ago

mattmccutchen-cci commented 3 years ago

It looks like DeclRewriter::buildItypeDecl currently uses either the original type or an all-wild type for the unchecked side of the itype. This may be wrong in cases such as double pointers. A simple example:

void foo(int **x) {
  *x = (int *)1;
  int **y;
  x = y;
}

3C output:

void foo(int **x : itype(_Ptr<_Ptr<int>>)) {
  *x = (int *)1;
  _Ptr<int *> y = ((void *)0);
  x = y;  // error: assigning to 'int **' from incompatible type '_Ptr<int *>'
}

Here the solution for the internal PVConstraint of x was _Ptr<int *>, but it was written as int **, causing the indicated compile error.

Should we make DeclRewriter::buildItypeDecl use the actual internal PVConstraint? I prototyped this and it fixed the example above, but that simple-minded approach broke many regression tests and I don't know how much work it would take to fix them (maybe not much, but I don't want to spend any more time on it now).