[Bug c/82186] [7/8 Regression] ICE (segfault), VLA type with inlining

["joseph at codesourcery dot com" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82186

--- Comment #6 from joseph at codesourcery dot com <joseph at codesourcery dot com> ---
For C, what is supposed to happen is that every call to groktypename where 
there might be side effects from the type name passes a non-null EXPR 
argument, and then the caller arranges for evaluation of expression stored 
in *EXPR at an appropriate point, if the language semantics mean it should 
be evaluated.  For example, c_cast_expr creates a C_MAYBE_CONST_EXPR which 
is meant to ensure the expressions required for the type being cast to are 
evaluated before the cast itself; sizeof handling ensures the expression 
is only evaluated if the type is a VLA, not e.g. a pointer to VLA.

The evaluation has the happen at the right point within the containing 
expression, e.g. in

(foo = bar (), (int (*)[foo = baz ()]) something)

the expression foo = baz () in the type name has to be evaluated after foo 
= bar ().

Note that there are cases where the size expression may never be evaluated 
- if the type of a conditional expression involves a VLA size expression 
from an unevaluated half of that conditional expression.  In that case, 
there is undefined behavior at runtime (but as usual the compiler still 
needs to accept that code, since whether the expression is evaluated is a 
property of a particular execution of the program).  See C99 DR#342 and 
C99 DR#340.