Alexander_Droste added inline comments.
================
Comment at: lib/StaticAnalyzer/Core/MemRegion.cpp:653
@@ +652,3 @@
+ // name by calling 'getDescriptiveName' recursively.
+ else {
+ std::string Idx = ER->getDescriptiveName(false);
----------------
Alexander_Droste wrote:
> zaks.anna wrote:
> > xazax.hun wrote:
> > > Alexander_Droste wrote:
> > > > I wasn't able to build a test case yet for which the analyzer could not
> > > > determine the constant value. Is there a way to trick the analyzer so
> > > > that the else case is used ? Then I could test for something like
> > > > `'sendReq1[a][7][b]'`.
> > > You can try use a value returned from a function that has an unknown
> > > body. E.g.:
> > >
> > > int getUnknown();
> > >
> > > void f() {
> > > int a = getUnKnown();
> > > }
> > What happens when you try 'sendReq1[a][7][b]'? Does it know the values for
> > "a" and "b" for some reason? If 'a' would be an input parameter and the
> > analyzer did not see a call site, it won't know the value of 'a'.
> If the return value of a function is used for which the body is not known
> Clang crashes.
>
> ```
> int getUnknown(void);
> int idxA = getUnknown();
> MPI_Request sendReq1[10][10][10];
> MPI_Wait(&sendReq1[1][idxA][9], MPI_STATUS_IGNORE); //
> expected-warning{{Request 'sendReq1[1][7][9]' has no matching nonblocking
> call.}}
> ```
> Clang also crashes if the index-variable is not initialized.
>
> ```
> int idxA;
> MPI_Request sendReq1[10][10][10];
> MPI_Wait(&sendReq1[1][idxA][9], MPI_STATUS_IGNORE); //
> expected-warning{{Request 'sendReq1[1][7][9]' has no matching nonblocking
> call.}}
> ```
> In case the variable is initialized with a constant, the `ConcreteInt` is
> determined.
..MPI-Checker doesn't emit diagnostics which include variable names as indices
like `sendReq1[a][7][b]`. The problem is that the checker has to reason about
single regions. If the specific index is not known, `getDescriptiveName()`
tries to obtain the variable name. But those cases are skipped by the checker,
as it does not know which specific region is used.
http://reviews.llvm.org/D16044
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
