[ClangIR][CIRGen] Handle nested union in arrays of struct

[ChuanqiXu9]

Reproducer:

struct nested
{
  union {
    const char *single;
    const char *const *multi;
  } output;
};
static const char * const test[] = {
  "test",
};
const struct nested data[] = 
{
    {
        {
            .multi = test,
        },
    },
    {
        {
            .single = "hello",
        },
    },
};

ClangIR now failed to recognize data as an array since it failed to recognize the initializer for union. This comes from a fundamental difference between CIR and LLVM IR. In LLVM IR, the union is simply a struct with the largest member. So it is fine to have only one init element. But in CIR, the union has the information for all members. So if we only pass a single init element, we may be in trouble. We solve the problem by appending placeholder attribute for the uninitialized fields.

[bcardosolopes]

Just a heads up: I usually don't review patches failing to pass tests, let me know when this is ready

[ChuanqiXu9]

(Triggering rerunning)

[ChuanqiXu9]

@bcardosolopes the CI is green now

[smeenai]

I'm working on adding an undef attr in #993. Do you think we could reuse that here instead of a new placeholder attr? I haven't fully thought through the LLVM semantics implications of that.

[ChuanqiXu9]

I'm working on adding an undef attr in #993. Do you think we could reuse that here instead of a new placeholder attr? I haven't fully thought through the LLVM semantics implications of that.

Yes, this was the reason I called it as a placeholder attr instead of undef. Since undef has a corresponding part to LLVM and placeholder attr won't. My thought to placeholder is, it is literally a place holder for CIR to complete its semantics. We should never lower a place holder. So I think this is the key difference. WDYT?

[ChuanqiXu9]

I think one motivating example may be, if in some transformation we changed the init of a union into a undef, then it looks slightly prettier:

(undef, placeholder, placeholder)

than

(undef, undef, undef)

at least we know the expected initialized field.

[smeenai]

Yeah, that makes sense to me, though I'll wait for Bruno to weigh in too :)

[bcardosolopes]

Thanks for working on this! Comments inline

[bcardosolopes]

Thanks for the updates, one more round of reviews

[bcardosolopes]

looks like new CI failures!

[ChuanqiXu9]

It looks like the windows failure is not relevant.