MOVE_ALLOC to grow an array

[rgba]

@ivanpribec

also provided in the gfortran documentation

@FortranFan

image779×665 22.5 KB

Oh sorry, somehow I missed the central info on the page.

@ivanpribec

The deallocation of an allocatable dummy argument with intent out is part of the Fortran semantics. From a practitioner point of view it’s one of the Fortran “gotchas” you need to be aware of. Here’s the description from MRC (the “red book”), p. 111

Perfect! Now I fully understand, thanks a lot for the snippet. Gotta buy the red book, has been on the amazon wishlist for too much already

[RonShepard]

@rgaelzer

An important difference between move_alloc and allocation on assignment exists when the lower bounds are different from one.
With AoA, the (re)allocated array will always have the lower bounds equal to one, regardless the bounds of the original array.
If you want to keep the original lower bound, you will need to employ move_alloc.

I consider this to me an important missing feature in fortran. It should be possible for the programmer to directly manipulate the metadata associated with allocatable and pointer entities. Part of that would be the ability to reset the bounds of existing arrays. This could be implemented in several ways, e.g. during the assignment, or after the assignment in a separate statement.

If you remember in f90, it was not possible to set the lower bounds during pointer assignment either. There was a workaround (do a subroutine call that changed the target bounds, then do the pointer assignment). That oversight was eventually corrected in f2003.