[PATCH] D28771: [Analyzer] Various fixes for the IteratorPastEnd checker

[Balogh, Ádám via Phabricator via cfe-commits]

baloghadamsoftware added inline comments.


================
Comment at: lib/StaticAnalyzer/Checkers/IteratorPastEndChecker.cpp:530
+  auto value = RVal;
+  if (auto loc = value.getAs<Loc>()) {
+    value = State->getRawSVal(*loc);
----------------
NoQ wrote:
> Is there a test case for this hack?
> 
> I'd also consider inspecting the AST (probably before passing the values to `handleRandomIncrOrDecr()`) and making the decision based on that. Because even though this pattern ("if a value is a loc and we expect a nonloc, do an extra dereference") is present in many places in the analyzer, in most of these places it doesn't work correctly (what if we try to discriminate between `int*` and `int*&`?).
I just want to get the sign of the integer value (if it is available). It turned out that I cannot do comparison between loc and nonloc. (Strange, because I can do anything else). After I created this hack, the Analyzer did not crash anymore on the llvm/clang code.

I do not fully understand what I should fix here and how? In this particular place we expect some integer, thus no int* or int*&.


================
Comment at: lib/StaticAnalyzer/Checkers/IteratorPastEndChecker.cpp:553
+
+  // When increasing by positive or decreasing by negative an iterator past its
+  // end, then it is a bug. We check for bugs before the operator call.
----------------
NoQ wrote:
> I think incrementing `end()` by `0` is not a bug (?)
I think it is not a bug, but how to solve it properly? If I chose just greaterThanZero, then we have the same problem for decrementing end() by 0. Is it worth to create three states here?


https://reviews.llvm.org/D28771