A struct with a zero-width bit-field as first member doesn't work.
The test bug-1981238.c has one, but the the ASSERT in there testing it are commented out. The initialization fails in a rather bad way: No space is allocated for the whole initializer ( at least on stm8), so any subsequent objects would be initialized incorrectly, too. And on OpenBSD on PowerPC G4 we do get sporadic segfaults in the compiler.
Fixed in [r13848].
Related
Commit: [r13848]