On 12/19/2017 01:58 PM, Richard Smith wrote:
On 18 Dec 2017 15:23, "Stephan Bergmann via cfe-commits"
<cfe-commits@lists.llvm.org <mailto:cfe-commits@lists.llvm.org>> wrote:
On 12/18/2017 03:05 PM, Richard Smith wrote:
Can we just strip the noexcept from the function type before
emitting the fsan type info?
As is apparently already done when emitting the (not
noexcept-annotated) __cxxabiv1::__function_type_info referenced from
a noexcept-annotated __cxxabiv1::__pointer_type_info?
That would indeed look like a better way to address this, then. But
wouldn't that also prevent -fsanitize=function from finding
mismatches where a non-noexcept function is called through a
noexcept pointer, as in
void f() {}
void g(void (*p)() noexcept) { p(); }
int main() { g(reinterpret_cast<void (*)() noexcept>(f)); }
That call seems OK to me; if the function does throw, std::terminate()
will be called.
I would have assumed that should be UB. But that arguably could depend
on whether and how that DR about calling noexpect functions through
non-noexcept pointers that both you and I reported will be resolved.
(It's a bit unfortunate that no information about those DR filings is
yet available at
<http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html>.)
One more argument in favour of resolving this issue as per
<https://reviews.llvm.org/D40720#958730> for now. So I'll look into that.
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
