Can you show an example where this happens, that is that
const_wanted is false
and
cur_text_section is NULL
at the same time in cond_expr()?
Sure. It's a side effect of the VLA array patches. On Hurd, there is a header
with something like:
#define FOO_COUNT (cond ? 2 : 42)
type foo_array[FOO_SIZE];
This lead post_type to call gexpr() which will eventually call cond_expr()
without const_wanted. This in turn will try to generate some code but will
segfault due to cur_text_section being NULL (this is how I spotted it, it
caused a segfault in the testsuite on Hurd)