* clang will complain.

* Cray ftn version 10.0.2 also runs into this problem in debug mode.
  It appears the incorrect preprocessor macro recently fixed only
  went unnoticed because the bug disappears when the optimizer kicks
  in.
* debug mode here referring to -O0 -g in FCFLAGS.

* clang will warn about not explicitly listing all enum values.
* gcc then realizes num_sub_elems might be uninitialized if elem->type
  somehow has a value not part of its type.

* clang complains about pointer conversions if the input type
  has different alignment from the result, unless there is a
  conversion to void * in between.
* Since a void * rhs can be assigned to any pointer object,
  the conversion can simplified accordingly.

* macOS sed reports byte sequences that are not valid UTF-8 characters if
  LC_CTYPE is set to UTF-8.

* mostlyclean is recursive and distclean-local should only depend on
  non-recursive targets.

* On error distcheck can leave a write-protected tree behind.

* Either initialize padding bytes on x86 and x86_64 explicitly or by
  declaring variables in text segment. Otherwise valgrind reports use
  of uninitialized memory in the memcmp test.