https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118120
--- Comment #6 from kargls at comcast dot net --- (In reply to Slava Zakharin from comment #5) > Right, this test deliberately introduces the aliasing to demonstrate that > gfortran has too optimistic assumptions about aliasing of COMPLEX and REAL > pointers. "Too optimistic assumptions about aliasing of COMPLEX and REAL pointers" is likely an optimistic description of the issue. :-) One can probably set an aliasing issue with pointers without involvement of a complex-part-ref, and run into an issue. complex-part-refs were added in Nov 2018; while gfortran was added to GCC in 2005. As the TODO in trans-array.cc notes, someone needs to do a deeper dive into dependency analysis for 'var = expr' by walking the right-hand expression looking for aliasing when 'var' is an array pointer.
