Various fixes for using NAG #637
Conversation
|
Is there a way we can download the NAG compiler and use it in GitHub? Otherwise, how can we know that the NAG build does not break in the future? |
|
I don't know about NAG and github. Beyond that the only solution would be to wait for people who are testing with it to raise PRs. |
|
Seems like NAG is the one compiler that cannot run on GitHub due to weird license issues. I wonder if any of these could be detected as GNU warnings. See #300. |
| @@ -21,6 +21,10 @@ | |||
| !> @author J. Ator @date 2009-07-01 | |||
| program apxdx | |||
|
|
|||
| #ifdef NAGFOR | |||
There was a problem hiding this comment.
We really don't want compiler specific ifdefs in the code. Is there another way we can resolve this?
What is the problem with exit?
There was a problem hiding this comment.
The problem is that exit is not a standard Fortran intrinsic, so compilers don't have to provide it.
There was a problem hiding this comment.
Maybe STOP and/or ERROR STOP would be a standardly correct alternative? They can return error codes to the environment.
The exit statement might be detectably by gnu if there is a warning to detect non-standard intrinsics. Not sure about the declaration order stuff. The intent(out) would need runtime checks I would guess, and then it would depend on how good gfortran's undefined variable checking is. |
I have made some changes to enable the use of NAG. Changes are:
The changes in mastertable.F90 are related to intent(out) arguments being used without being defined. For instance in subroutine sntbde, nmtbd is declared as intent(out) and promptly used in a comparison. Intent(out) arguments become undefined upon subroutine entry so this is undefined.
In missing.F90 there is an attempt to write a real value with a Z edit descriptor. This is not valid according to the standard, only integers can be used with a Z edit descript. I have used a transfer to fix this.
In readwriteval.F90 I have re-ordered some declarations. When declaring something like
integer :: a(b,c)
the b and c have to be declared before a; the rearrangements make sure this is the case.
For the rest the exit function is not standard Fortran and for NAG is accessed via a compiler specific module. I have done this hidden behind a CPP flag.