http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51208
--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-11-18
16:03:04 UTC ---
(In reply to comment #1)
> (In reply to comment #0)
> > allocate(i, stat=i)
> > 1
> > Error: Stat-variable at (1) shall not be ALLOCATEd within
> > the same ALLOCATE statement
>
> The above error is easy to produce, because gfortran
> only has to check if an alloc-object is used as a
> stat-variable.
Well, you can make it also more complicated:
allocate (a(1)%i, stat=a%(2-1)%i)
Or extremely complicated:
integer, pointer :: ptr
allocate (ptr, stat=f())
contains
function f()
integer, pointer :: f
f => ptr
end function
end
(Recall that a pointer-function result is a variable in Fortran 2008.)
> > However, for SOURCE= and MOLD= it does not work:
>
> Doing a check here requires a walk of the mold-expr
> and the source-expr, which is of course much more
> work
Well, it is not. One can restrict one to the common case of expr->expr_type ==
EXPR_VARIABLE and just do the same as for STAT=: Checking whether the variable
is the same.
Will that catch all wrong usage? No, but it will catch the most common mistake
of choosing the wrong variable. That's as illustrated above the same with
STAT=.
I am sure that Intel's compiler does not do anything more advanced - and it
would have found the mistake I made in PR 51207.
