[LLVMdev] Proposal for safe-to-execute meta-data for heap accesses

[Dean Sutherland]

FYI, we (DeLesley Hutchins at Google and my group at CERT) have a new static analysis under development that will be able to express and enforce this kind of property (once the analysis is up and working). When we get to the point where it can handle the LLVM & Clang sources, implementer will be able to count on having code that actually follows the rules, too.

Dean Sutherland
dsutherland at cert.org

On Nov 9, 2013, at 12:36 AM, Chandler Carruth <chandlerc at google.com> wrote:
[SNIP]

> I do *not* want parallelizing LLVM to require careful locking protocols to be followed. Instead, I want the design to naturally arrange for different threads to operate on different constructs and for the interconnecting interfaces to be thread safe. The best system we have yet devised for this is based around function passes not digging into tho bodies of other functions. Instead we rely on attributes to propagate information about the body of another function to a caller.