Automatic vs. allocatable arrays for function results

[Beliavsky]

@themos

A simple rule of thumb says always prefer the newer feature. In this case, allocatable arrays. You get orderly error handling in case of insufficient resources and you can complain loudly to the vendor if the performance is lacking because it is a current feature that is used a lot.

But the older a feature is, the more production codes there are using it, the more complaining there will be if it is broken, and the more confidence you can have that your compiler, even an old version, supports it. For example, the code below is not legal

do 1 i=1,3
   do 1 j=1,2
1      print*,i,j,i+j
end

since as gfortran explains,

continue.f90:2:13:

    2 |    do 1 j=1,2
      |             1
Warning: Fortran 2018 deleted feature: Shared DO termination label 1 at (1)
continue.f90:3:21:

    3 | 1      print*,i,j,i+j
      |                     1
Warning: Fortran 2018 deleted feature: DO termination statement which is not END DO or CONTINUE with label 1 at (1)

but as a practical matter you can rely on Fortran compilers supporting it (outside of strict mode) since so much code uses deleted or obsolescent features. One can help Fortran progress by using new features and reporting bugs, but that is a different issue.