[PATCH] D121120: [clang-tidy] New check for safe usage of `std::optional` and like types.

[Stanislav Gatev via Phabricator via cfe-commits]

sgatev added inline comments.


================
Comment at: clang-tools-extra/clang-tidy/bugprone/UncheckedOptionalAccessCheck.cpp:39-45
+  using dataflow::ControlFlowContext;
+  using dataflow::DataflowAnalysisContext;
+  using dataflow::DataflowAnalysisState;
+  using dataflow::Environment;
+  using dataflow::UncheckedOptionalAccessModel;
+  using dataflow::WatchedLiteralsSolver;
+  using llvm::Expected;
----------------
Do we really need all these using declarations? There seems to be one reference for each of these types. I think we can simply qualify the references.


================
Comment at: clang-tools-extra/clang-tidy/bugprone/UncheckedOptionalAccessCheck.cpp:54
+      std::make_unique<WatchedLiteralsSolver>());
+  Environment Environment(AnalysisContext, FuncDecl);
+  UncheckedOptionalAccessModel Analysis(ASTCtx);
----------------
Call this `Env` to disambiguate from the name of the type.


================
Comment at: clang-tools-extra/clang-tidy/bugprone/UncheckedOptionalAccessCheck.cpp:87
+                     withInitializer(HasOptionalCallDescendant)))))
+          .bind("f"),
+      this);
----------------
Let's add a constant for this so that we can reuse it in `UncheckedOptionalAccessCheck::check`.


================
Comment at: clang-tools-extra/clang-tidy/bugprone/UncheckedOptionalAccessCheck.cpp:84
+  if (!BlockToOutputState ||
+      BlockToOutputState->size() <= Context->getCFG().getExit().getBlockID())
+    return;
----------------
xazax.hun wrote:
> ymandel wrote:
> > xazax.hun wrote:
> > > xazax.hun wrote:
> > > > ymandel wrote:
> > > > > xazax.hun wrote:
> > > > > > xazax.hun wrote:
> > > > > > > Could the size of the vector ever be wrong? Should this be an assert instead?
> > > > > > Whoops, after the update this comment is out of place, now it supposed to be on line 60. 
> > > > > Based on my reading, it is a rare, but possible condition. Basically, we need code where the exit block is unreachable, which I believe can happen in weird cases like:
> > > > > 
> > > > > ```
> > > > > while(true) {...}
> > > > > ```
> > > > > https://godbolt.org/z/rfEnfaWTv -- notice the lack of predecessors for the exit block.
> > > > > 
> > > > > See the code here, which follows the ordering of the blocks and doesn't force blocks to be processed:
> > > > > 
> > > > > https://github.com/llvm/llvm-project/blob/main/clang/lib/Analysis/FlowSensitive/TypeErasedDataflowAnalysis.cpp#L337-L364
> > > > Interesting. Since we already have optionals in the vector, I assumed we will always have matching size. I think we might want to change this so there is only one way for the analysis to not provide a state for a basic block to make this a bit less confusing, 
> > > Actually, in the linked code I see ` BlockStates.resize(CFCtx.getCFG().size(), llvm::None);`. So I would expect the size to be always right with possibly some `None`s for the nodes that were not processed.
> > > Actually, in the linked code I see ` BlockStates.resize(CFCtx.getCFG().size(), llvm::None);`. So I would expect the size to be always right with possibly some `None`s for the nodes that were not processed.
> > Ah, my mistake! I thought `resize` only allocated the space. #TIL
> > 
> > Changed to an assert. Thanks.
> > 
> But this discussion shed light on an interesting detail. If the exit block is unreachable, we will not diagnose the unsafe accesses. I wonder if this worth a FIXME. 
The documentation of `runDataflowAnalysis` already hints that the size of the vector matches the size of the CFG. Do you think we should make this more clear?


================
Comment at: clang-tools-extra/test/clang-tidy/checkers/Inputs/absl/types/optional.h:69
+  void swap(optional &rhs) noexcept;
+};
+} // namespace absl
----------------



================
Comment at: clang-tools-extra/test/clang-tidy/checkers/bugprone-unchecked-optional-access.cpp:24
+
+void f2(const absl::optional<int> &opt) {
+  if (opt.has_value()) {
----------------
Call this `checked_access`?


================
Comment at: clang/include/clang/Analysis/FlowSensitive/Models/UncheckedOptionalAccessModel.h:39
 
+/// A matcher for the optional classes covered by this model.
+ast_matchers::DeclarationMatcher optionalClass();
----------------



================
Comment at: clang/include/clang/Analysis/FlowSensitive/Models/UncheckedOptionalAccessModel.h:40
+/// A matcher for the optional classes covered by this model.
+ast_matchers::DeclarationMatcher optionalClass();
+
----------------
This should be a class member, no? `clang::dataflow::optionalClass` seems underspecified.


================
Comment at: clang/lib/Analysis/FlowSensitive/Models/UncheckedOptionalAccessModel.cpp:237
+  // FIXME: include either the name of the optional (if applicable) or a source
+  // range of the access for easier intepretation of the result.
   State.Lattice.getSourceLocations().insert(ObjectExpr->getBeginLoc());
----------------



Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D121120/new/

https://reviews.llvm.org/D121120