[cfe-commits] [PATCH] Fix crash-on-invalid when passing 128-bit integers to attributes that expect ICEs

[Douglas Gregor]

On Jun 15, 2012, at 10:58 AM, Dmitri Gribenko <gribozavr at gmail.com> wrote:

> Now with a patch.  Excuse me for the noise.
> 
> On Fri, Jun 15, 2012 at 10:55 AM, Dmitri Gribenko <gribozavr at gmail.com> wrote:
>> Hello,
>> 
>> The attached patch fixes crash-on-invalid when passing 128-bit
>> integers to attributes that expect ICEs, for example:
>> 
>> void attr6(void *ptr) __attribute__((nonnull(0x10000000000000001i128)));
>> 
>> The root cause of the problem is using getZExtValue() on 128-bit integers.
>> 
>> Please review.
>> 
>> Dmitri

+/// \brief Check if E is a valid integer constant expression.  May output an
+/// error.
 ///
+/// \param AttrArgNum argument number used in diagnostics.
+///
+/// \param E expression to check.
+///
+/// \param [out] Val ICE value.
+static bool checkICEWithOverflow(Sema &S,
+                                 StringRef AttrName, SourceLocation AttrLoc,
+                                 unsigned AttrArgNum, const Expr *E,
+                                 uint64_t &Val) {
+  // The argument must be an integer constant expression.
+  llvm::APSInt Int;
+  if (E->isTypeDependent() || E->isValueDependent() ||
+      !E->isIntegerConstantExpr(Int, S.Context)) {
+    S.Diag(AttrLoc, diag::err_attribute_argument_n_not_int)
+      << AttrName << AttrArgNum << E->getSourceRange();
+    return false;
+  }

Could we get a specialized diagnostic for the case where E is type- or value-dependent?  At least users will know that, in those cases, we can't support arguments that depend on a template parameter. 

Otherwise, looks good!

	- Doug