NoQ added a comment.
I think this looks good. There's a problem with missing construction contexts,
but i guess that's not the checker's fault, so let's add a FIXME and commit.
================
Comment at: lib/StaticAnalyzer/Checkers/IteratorChecker.cpp:454-455
+ return;
+ const auto OldState = C.getPredecessor()->getFirstPred()->getState();
+ const auto OldThis = OldState->getSVal(ThisExpr, LCtx);
+ const auto *Pos = getIteratorPosition(OldState, OldThis);
----------------
This deserves a FIXME because that's definitely unreliable (i.e. if another
checker subscribes to the operator call and adds a transition before you,
you'll break because you'd have to ascend two nodes above, not one).
The proper fix is to make the CFG provide a `ConstructionContext` for the
`CXXOperatorCallExpr`, which would turn the corresponding `CFGStmt` element
into a `CFGCXXRecordTypedCall` element, which will allow `ExprEngine` to
foresee that the `begin()`/`end()` call constructs the object directly in the
temporary region that `CXXOperatorCallExpr` takes as its implicit object
argument.
The proper fix is not hard, but there are still a lot of simpler and more
common cases that we don't handle.
================
Comment at: lib/StaticAnalyzer/Checkers/IteratorChecker.cpp:476-502
+void IteratorChecker::checkLiveSymbols(ProgramStateRef State,
+ SymbolReaper &SR) const {
+ // Keep symbolic expressions of iterator positions, container begins and ends
+ // alive
+ auto RegionMap = State->get<IteratorRegionMap>();
+ for (const auto Reg : RegionMap) {
+ const auto Pos = Reg.second;
----------------
I guess we'll have this sorted out in another patch.
https://reviews.llvm.org/D32642
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
