[Bug fortran/49755] New: ALLOCATE with STAT= produces invalid code for already allocated vars

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49755

           Summary: ALLOCATE with STAT= produces invalid code for already
                    allocated vars
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Keywords: missed-optimization, wrong-code
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: burnus@gcc.gnu.org
                CC: dcarrera@gcc.gnu.org
            Blocks: 32834


The Fortran standard states that ALLOCATEing an already allocated variable is
an error.

gfortran nicely aborts with a run-time error if there is no STAT= variable.
However, if there is STAT= variable, instead of just setting it and stopping,
gfortran deallocates the variable and then tries the new allocation.


There are several issues with that:

(a) It's invalid according to the Fortran standard, which states in
    "6.7.4 STAT= specifier":
    "* each allocate-object that was not successfully allocated or
       deallocated shall retain its previous allocation status
       or pointer association status."

(b) It adds lots of code, which should be most of the time not used,
    increasing the code size and making optimizations more difficult,
    especially for derived types with allocatable components

(c) With coarrays, there are even more issues - one shall not simply
    free a coarray unilaterally!



Test case (except for the "abort" function, valid Fortran 2008):


integer, allocatable :: A(:, :)
integer :: stat

allocate(A(20,20))
A = 42

! Allocate of already allocated variable
allocate (A(5,5), stat=stat)

! Expected: Error stat and previous allocation status

if (stat == 0) call abort ()
if (any (shape (A) /= [20, 20])) call abort ()
if (any (A /= 42)) call abort ()
end