https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83692
--- Comment #6 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Dang it, hit Save Changes accidentally.
That evaluates the AGGR_INIT_EXPR inside by evaluating the constructor of
outer:
if (m_x.value() != 0) // 1
throw 0;
m_x.m_value = integer{1}.value(); // 2
if (m_x.value() != 1) // 3
throw 0;
1: we evaluate m_x.value() and save it to constexpr_call_table with result 0,
because were in strict mode
2: this store expression is evaluated, it changes m_value to 1
3: we try to evaluate m_x.value() again but this time we use the cached result
because we're still in strict mode. But the result is not 0 anymore, so Bad
Stuff ensues.
So either we should prune the constexpr call table (but when?), or not cache
the result in the first place.
