consteval calls in default args

[urnathan]

Consteval calls in default args do not work

consteval int here() { return 5; }
void error(int location = here()) { }
void bob () { error(); }

The problem here is we markused the consteval fn, which causes us to emit an error about taking the address of the consteval fn. I patched up EvaluatedExprMarker, called from Sema::CheckCXXDefaultArgExpr to skip such consteval calls, but we then ICE in code emission, trying to emit a consteval fn body.

https://bugs.llvm.org/show_bug.cgi?id=5810 seems relevant, as that's mentioned in the comment in CheckCXXDefaultArgExpr:
// FIXME: We should really be rebuilding the default argument with new
// bound temporaries; see the comment in PR5810.
// We don't need to do that with block decls, though, because
// blocks in default argument expression can never capture anything.

5810 includes the ominous note:
One could imagine rebuilding the AST nodes with new CXXTemporary pointers, although that's going to require a lot of code :(

Notice that because of __source_location(), which is different on every evaluation, we can't just morph the original defaultArgExpr as is held in the ParamDecl. We have to clone and evaluate at each use, I think.

(this default arg pattern is used in the source_location header to get the location corresponding to the caller of the source_location ctor, not the location of the ctor itself)

[llvmbot]

@llvm/issue-subscribers-clang-frontend

[tbaederr]

This is a duplicate of #48230 afaics.

[urnathan]

It appears to be the same -- the problem doesn't just effect consteval ctors. The recent addition of builtin_source_location (https://reviews.llvm.org/rGd61487490022) seems to have exposed this issue -- suddenly the (gcc) libstdc++ source_location header bursts into light, and then goes down in flames :(

[ecatmur]

#48230 is fixed by a4f8590, should this be closed?

[usx95]

This is fixed at head.
https://godbolt.org/z/nvjb7hEoh