On 04/17/19 20:31, Andrew Fish wrote:
> I was actually writing a mail to some people that sit on the C/C++
> standards committee that are UB experts to get some clarification when
> you sent this mail. [...]
>
> I'm basically asking if this code pedantic conforms to C99 and C11:
>
> EFI_COMMON_SECTION_HEADER gSec = { { 0x01, 0x02, 0x3 }, 0x10 };
>
> return *(UINT32 *)gSec.Size & 0x00ffffff;
I'm really interested as to how they will respond!
(Because I think (see up-thread) that the code doesn't conform.)
> I ran the clang static analyzer and runtime ubsan on the above code
> and it did not complain (I force strict aliasing via
> -fstrict-aliasing, and I'm using the Sys V ABI since this is just the
> command line compiler on my Mac).
Yesterday I tried something similar with both gcc and clang (not recent
versions of them, unfortunately), and they were all too happy with the
code as well, when I thought they should have flagged it.
The options I used were:
-fsyntax-only -Wall -Wextra -std=c99 -pedantic \
-fstrict-aliasing -Wstrict-aliasing
Beyond that, I checked the documentation of "-fstrict-aliasing" in the
gcc manual at
<https://gcc.gnu.org/onlinedocs/gcc-8.3.0/gcc/Optimize-Options.html>. It
provides several examples; one of them is:
> Similarly, access by taking the address, casting the resulting pointer
> and dereferencing the result has undefined behavior, even if the cast
> uses a union type, e.g.:
>
> union a_union {
> int i;
> double d;
> };
>
> int f() {
> double d = 3.0;
> return ((union a_union *) &d)->i;
> }
I think the access in this example is well defined (I seem able to
deduce it using the effective type rules), and so I consider this a bug
in the gcc docs. I reached out to someone in the toolchain team at Red
Hat to confirm or disprove.
Thanks
Laszlo
-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#39289): https://edk2.groups.io/g/devel/message/39289
Mute This Topic: https://groups.io/mt/31070304/21656
Group Owner: devel+ow...@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
