[Bug c/99211] New: gcc fails on program which overrides __builtin_clzll

[Bug c/99211] New: gcc fails on program which overrides __builtin_clzll

[zhan3299 at purdue dot edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99211

            Bug ID: 99211
           Summary: gcc fails on program which overrides __builtin_clzll
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: zhan3299 at purdue dot edu
  Target Milestone: ---

GCC fails to compile the attached code: https://godbolt.org/z/sdqq3M

One interesting thing is that, only when we override __builtin_clzll this issue
will occur. If we replace the __builtin_clzll as any other builtin function
(e.g., __builtin_expect), GCC will accept the program with a conflicting type
warning: https://godbolt.org/z/5c45T6

For other compilers,

Clang rejects such conflicts with built-in functions:
https://godbolt.org/z/9WzTKP 

Icc accepts and successfully compiles it (with warnings):
https://godbolt.org/z/W4zK89


$ cat test.c
struct S {
    int a;
};

struct S __builtin_clzll(long x) {
    if (x == 1) {
        struct S y = { .a = 1 };
        return y;
    } else {
        struct S y = { .a = __builtin_clzll(1).a + 1 };
        return y;
    }
}

int main() {
    return __builtin_clzll(10).a;
}

$ gcc test.c
test.c: In function '__builtin_clzll':
test.c:10:16: internal compiler error: in wide_int_to_tree_1, at tree.c:1644
   10 |         struct S y = { .a = __builtin_clzll(1).a + 1 };
      |                ^
0x1a31d99 internal_error(char const*, ...)
        ???:0
0x6877ad fancy_abort(char const*, int, char const*)
        ???:0
0x9bed4a fold_build_call_array_loc(unsigned int, tree_node*, tree_node*, int,
tree_node**)
        ???:0
0x6e1010 build_function_call_vec(unsigned int, vec<unsigned int, va_heap,
vl_ptr>, tree_node*, vec<tree_node*, va_gc, vl_embed>*, vec<tree_node*, va_gc,
vl_embed>*, tree_node*)
        ???:0
0x7322e9 c_parse_file()
        ???:0
0x7a02c2 c_common_parse_file()
        ???:0
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

[Bug c/99211] gcc fails on program which overrides __builtin_clzll

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99211

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |accepts-invalid,
                   |                            |ice-on-invalid-code
                 CC|                            |jsm28 at gcc dot gnu.org

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
I wonder if we can reject definitions/declarations of identifiers in GCCs
implementation namespace (__builtin_*), even if correctly typed?
The C family frontends have been prone to accept random crap as builtin
functions.

[Bug c/99211] gcc fails on program which overrides __builtin_clzll

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99211

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I think we have other PRs like this and it has been discussed.
Problem with disabling altogether __builtin_* function declarations/definitions
is that code in the wild uses
typeof (whatever) __builtin_whatever __attribute__((...));
style declarations.

[Bug c/99211] gcc fails on program which overrides __builtin_clzll

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99211

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #2)
> Problem with disabling altogether __builtin_* function
> declarations/definitions
> is that code in the wild uses
> typeof (whatever) __builtin_whatever __attribute__((...));
> style declarations.

It deserves to be broken.

[Bug c/99211] gcc fails on program which overrides __builtin_clzll

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99211

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
We have it even in gcc sources, though ifdefed so it won't trigger with
recent-ish gcc:
libstdc++-v3/include/ext/throw_allocator.h:      __typeof__(&std::sprintf)
__builtin_sprintf = &std::sprintf;

I think this was PR32455, I believe the patch has never been applied, but I
don't really remember all the details about it.