[PATCH] D64991: [analyzer][WIP] Implement a primitive reaching definitions analysis

Mon Jul 22 13:19:10 PDT 2019

NoQ added a comment.

In D64991#1595853 <https://reviews.llvm.org/D64991#1595853>, @Szelethus wrote:

> `CFGElementRef`

Wait, is it a thing already?? Did i miss anything???

> This analysis isn't conservative enough yet, I really should include function calls with non-const references into the gen set, but hey, at this point I'm optimistic that it can be done.

"Conservative" is not a property of the algorithm; it depends on how is the algorithm used. For some purposes you'll want to add an extra definition when you're not sure if it'll be a definition, for other purposes you'll want to only have definitions in which you are 100% certain in the set.

================
Comment at: clang/lib/Analysis/ReachingDefinitions.cpp:68
+    auto Assignments =
+        match(stmt(forEachDescendant(AssignmentM)), *S, *Context);
+    if (!Assignments.empty()) {
----------------
I don't think you really need this. Every assignment would anyway appear in the CFG as a separate element. And also `forEachDescendant()` may accidentally scan items in different CFG blocks (i.e., if the whole statement is a `?:` or a logical op).

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D64991/new/

https://reviews.llvm.org/D64991