C++: IR generation for repeated initializers

[MathiasVP]

This implements the IR changes necessary after #12712 to support repeated initializers.

Commit-by-commit review recommended.

• The first commit adds @sashabu's failing tests
• The second commit implements proper IR support
• The third commit accepts the test changes (all of which are good)
• The final commit renames a couple of the library predicates to properly reflect that they can have multiple results.

[MathiasVP]

DCA looks fine: Performance looks unchanged, no new results, and some IR inconsistencies have been fixed 🎉

[jketema]

Mostly looks sensible. Few small remarks. Also a question. Consider:

struct Foo {
 int x[2];
 int y;
};

int main() {
  struct Foo f = { .x = {0, 1}, .x[0] = 42, .y = 42 };
  return 0;
}

What happens in this case?

[jketema]

Suggested change

	Expr getFieldExpr(Field field) { result = this.getFieldExpr(field, _) }
	deprecated Expr getFieldExpr(Field field) { result = this.getFieldExpr(field, _) }

[MathiasVP]

I was planning on doing this as a follow-up since these predicates are used a bunch of places in. So we'd need to replace those when we formally deprecate these.

[jketema]

Sounds good to me.

[jketema]

Suggested change

	Expr getElementExpr(int elementIndex) { result = this.getElementExpr(elementIndex, _) }
	deprecated Expr getElementExpr(int elementIndex) { result = this.getElementExpr(elementIndex, _) }

[MathiasVP]

What happens in this case?

Hmmm, good question. I'll add this test and see

[jketema]

Note that clang actually gives a warning on the slightly odd test case:

test.c:7:41: warning: initializer overrides prior initialization of this subobject [-Winitializer-overrides]
  struct Foo f = { .x = {0, 1}, .x[0] = 42, .y = 42 };
                                        ^~
test.c:7:26: note: previous initialization is here
  struct Foo f = { .x = {0, 1}, .x[0] = 42, .y = 42 };
                         ^
1 warning generated.

[MathiasVP]

I've added the case in 4033ed3. The output actually looks sensible, I think 🤞.

[jketema]

I've added the case in 4033ed3. The output actually looks sensible, I think 🤞.

Looks sensible indeed.

[jketema]

LGTM