clang: /root/llvm-project/clang/lib/AST/Expr.cpp:3289: bool clang::Expr::isConstantInitializer(clang::ASTContext&, bool, const clang::Expr**) const: Assertion `!isValueDependent() && "Expression evaluator can't be called on a dependent expression."' failed.

[danix800]

Clang trunk assertion failed on the following invalid initializer:

struct S {
	int v;
};

struct T;

void test(struct T *t)
{
	struct S s = { .v = t->y }; // t->y is invalid
}

https://godbolt.org/z/GnT41GbT5

Possibly similar to https://reviews.llvm.org/D84145.

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Ding Fei (danix800)

Clang trunk assertion failed on the following invalid initializer:

struct S {
	int v;
};

struct T;

void test(struct T *t)
{
	struct S s = { .v = t->y }; // t->y is invalid
}

https://godbolt.org/z/GnT41GbT5

Possibly similar to https://reviews.llvm.org/D84145.

[shafik]

Looks like we had this since clang-12: https://godbolt.org/z/ohd4bMTEh

Perhaps here:

llvm-project/clang/lib/Parse/ParseInit.cpp

Lines 509 to 513 in 896b5e5

	// If we couldn't parse the subelement, bail out.
	if (SubElt.isUsable()) {
	InitExprs.push_back(SubElt.get());
	} else {
	InitExprsOk = false;

we need if (SubElt.isUsable() && !SubElt.get()->containsErrors())

or maybe here:

llvm-project/clang/lib/Parse/ParseDecl.cpp

Line 2713 in 896b5e5

if (Init.isInvalid()) {

We need if (Init.isInvalid() || Init.get()->containsErrors())

These are just some ideas from debugging the crash.

[danix800]

Looks like we had this since clang-12: godbolt.org/z/ohd4bMTEh

Perhaps here:

llvm-project/clang/lib/Parse/ParseInit.cpp

Lines 509 to 513 in 896b5e5

	// If we couldn't parse the subelement, bail out.
	if (SubElt.isUsable()) {
	InitExprs.push_back(SubElt.get());
	} else {
	InitExprsOk = false;

we need if (SubElt.isUsable() && !SubElt.get()->containsErrors())

or maybe here:

llvm-project/clang/lib/Parse/ParseDecl.cpp

Line 2713 in 896b5e5

if (Init.isInvalid()) {

We need if (Init.isInvalid() || Init.get()->containsErrors())

These are just some ideas from debugging the crash.

Thanks for pointing this out. Could you please also take a look at PR #88014 . It might be more consistent as a fix.

[shafik]

Thanks for pointing this out. Could you please also take a look at PR #88014 . It might be more consistent as a fix.

Apologies, for the late reply. That looks like the right fix.