https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104171
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jsm28 at gcc dot gnu.org,
| |mpolacek at gcc dot gnu.org
Status|ASSIGNED |NEW
Assignee|rguenth at gcc dot gnu.org |unassigned at gcc dot
gnu.org
Component|middle-end |c
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue is that the GENERIC folding routine does not verify we have a
properly
typed call (and the diagnostics hint at the mismatch), but we still identify
the call as a call to the known cimag function. Rather than using its
known prototye we use the unprototyped cimagl with unpromoted argument and thus
run into
/* Fold a call to built-in function FNDECL with 1 argument, ARG0.
This function returns NULL_TREE if no simplification was possible. */
static tree
fold_builtin_1 (location_t loc, tree expr, tree fndecl, tree arg0)
{
...
CASE_FLT_FN (BUILT_IN_CIMAG):
if (validate_arg (arg0, COMPLEX_TYPE)
&& TREE_CODE (TREE_TYPE (TREE_TYPE (arg0))) == REAL_TYPE)
return non_lvalue_loc (loc, fold_build1_loc (loc, IMAGPART_EXPR, type,
arg0));
we ultimatively arrive here from fold_call_expr (and c_fully_fold_internal)
where we could do more extensive validation. In particular the CALL_EXPR
has &cimagl of unprototyped type and we could also employ checks like those
done in gimple_builtin_call_types_compatible_p.
Interestingly for the following testcase we end up _not_ setting
DECL_FUNCTION_CODE on cimagl and thus we do not ICE. IMHO a frontend problem.
ong double cimagl(_Complex);
test_cimagl(_Complex z) { return cimagl(z); }