[Bug fortran/78867] [7 Regression] GFortran function returning string ICE with -flto

["rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78867

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue is that at read-in cfun is unexpectedly NULL.  We're here:

static void
input_function (tree fn_decl, struct data_in *data_in,
                struct lto_input_block *ib, struct lto_input_block *ib_cfg)
{
...
  /* Read the tree of lexical scopes for the function.  */
  DECL_INITIAL (fn_decl) = stream_read_tree (ib, data_in);
  unsigned block_leaf_count = streamer_read_uhwi (ib);
  while (block_leaf_count--)
    stream_read_tree (ib, data_in);
^^^^

and only later we do:


  if (!streamer_read_uhwi (ib))
    return;

  push_struct_function (fn_decl);
  fn = DECL_STRUCT_FUNCTION (fn_decl);
  init_tree_ssa (fn);

this means that the BLOCK tree (and its VAR_DECLs) have references to SSA
names.

Which I think boils down to another instance of a missing DECL_EXPR and thus
not properly gimplified sizes:

test ()
{
  static void bar (character(kind=1)[1:] &, integer(kind=4), integer(kind=8) &
restrict);

  {
    static integer(kind=8) C.3454 = 2;
    integer(kind=8) * D.3455;
    integer(kind=8) D.3456;
    character(kind=1)[1:MAX_EXPR <(integer(kind=4)) D.3456, 0>] * pstr.0;
^^^

    void * restrict D.3458;
    integer(kind=4) D.3459;

    D.3455 = &C.3454;
    D.3456 = *D.3455;
    D.3458 = (void * restrict) __builtin_malloc (MAX_EXPR <(unsigned long)
MAX_EXPR <(integer(kind=4)) D.3456, 0>, 1>);
    pstr.0 = (character(kind=1)[1:MAX_EXPR <(integer(kind=4)) D.3456, 0>] *)
D.3458;
    bar (pstr.0, MAX_EXPR <(integer(kind=4)) D.3456, 0>, D.3455);
    D.3459 = MAX_EXPR <(integer(kind=4)) D.3456, 0>;