porting scipy's Fortran code to NAG's nagfor compiler

After the recently merged numpy’s https://github.com/numpy/numpy/pull/17950 nagfor still cannot compile all of scipy’s Fortran code.

Using numpy’s master with the PR just mentioned in, one still gets:

$ python3 setup.py build -j 1 config_fc --fcompiler=nagfor install --prefix $HOME/.local
...
nagfor:f77: scipy/integrate/odepack/srcma.f
Option warning: -c option specified twice, continuing
NAG Fortran Compiler Release 7.0(Yurakucho) Build 7036
Evaluation trial version of NAG Fortran Compiler Release 7.0(Yurakucho) Build 7036
Obsolescent: scipy/integrate/odepack/srcma.f, line 1: Fixed source form
Obsolescent: scipy/integrate/odepack/srcma.f, line 27: DO 10 ends neither with CONTINUE nor ENDDO
Obsolescent: scipy/integrate/odepack/srcma.f, line 29: DO 15 ends neither with CONTINUE nor ENDDO
Obsolescent: scipy/integrate/odepack/srcma.f, line 32: DO 20 ends neither with CONTINUE nor ENDDO
Obsolescent: scipy/integrate/odepack/srcma.f, line 34: DO 25 ends neither with CONTINUE nor ENDDO
Error: scipy/integrate/odepack/srcma.f, line 36: First subscript (49) is greater than upper bound (1) for array ISAV
Error: scipy/integrate/odepack/srcma.f, line 37: First subscript (50) is greater than upper bound (1) for array ISAV
Obsolescent: scipy/integrate/odepack/srcma.f, line 42: DO 110 ends neither with CONTINUE nor ENDDO
Obsolescent: scipy/integrate/odepack/srcma.f, line 44: DO 115 ends neither with CONTINUE nor ENDDO
Obsolescent: scipy/integrate/odepack/srcma.f, line 47: DO 120 ends neither with CONTINUE nor ENDDO
Obsolescent: scipy/integrate/odepack/srcma.f, line 49: DO 125 ends neither with CONTINUE nor ENDDO
Error: scipy/integrate/odepack/srcma.f, line 51: First subscript (49) is greater than upper bound (1) for array ISAV
Error: scipy/integrate/odepack/srcma.f, line 52: First subscript (50) is greater than upper bound (1) for array ISAV
[NAG Fortran Compiler error termination, 4 errors, 9 warnings]

This one is easy to fix (apparently natfor does not grok dimension?):

diff --git a/scipy/integrate/odepack/srcma.f b/scipy/integrate/odepack/srcma.f
index 62c07a571..aa3ba1e22 100644
--- a/scipy/integrate/odepack/srcma.f
+++ b/scipy/integrate/odepack/srcma.f
@@ -11,12 +11,11 @@ c        job  = 1 if common is to be saved (written to rsav/isav)
 c        job  = 2 if common is to be restored (read from rsav/isav)
 c        a call with job = 2 presumes a prior call with job = 1.
 c-----------------------------------------------------------------------
-      integer isav, job
+      integer isav(*), job
       integer ieh, ils, ilsa
       integer i, lenrls, lenils, lenrla, lenila
-      double precision rsav
+      double precision rsav(*)
       double precision rls, rlsa
-      dimension rsav(1), isav(1)
       common /ls0001/ rls(218), ils(39)
       common /lsa001/ rlsa(22), ilsa(9)
       common /eh0001/ ieh(2)

After this fix, the next error, not very clear either:

nagfor:f77: scipy/sparse/linalg/eigen/arpack/ARPACK/SRC/cnaitr.f
Option warning: -c option specified twice, continuing
NAG Fortran Compiler Release 7.0(Yurakucho) Build 7036
Evaluation trial version of NAG Fortran Compiler Release 7.0(Yurakucho) Build 7036
Warning: scipy/sparse/linalg/eigen/arpack/ARPACK/SRC/cnaitr.f, line 850: Unused dummy variable NB
Obsolescent: scipy/sparse/linalg/eigen/arpack/ARPACK/SRC/cnaitr.f, line 208: Fixed source form
Error: scipy/sparse/linalg/eigen/arpack/ARPACK/SRC/cnaitr.f, line 666: Inconsistent structure for arg 3 in call to SVOUT
Error: scipy/sparse/linalg/eigen/arpack/ARPACK/SRC/cnaitr.f, line 737: Inconsistent structure for arg 3 in call to SVOUT
[NAG Fortran Compiler error termination, 2 errors, 2 warnings]
error: Command "/usr/local/bin/nagfor -fixed -O4 -Iscipy/sparse/linalg/eigen/arpack/ARPACK/SRC -I/home/dimpase/.local/lib/python3.7/site-packages/numpy-1.21.0.dev0+137.g91d9bbebc-py3.7-linux-x86_64.egg/numpy/core/include -c -c scipy/sparse/linalg/eigen/arpack/ARPACK/SRC/cnaitr.f -o build/temp.linux-x86_64-3.7/scipy/sparse/linalg/eigen/arpack/ARPACK/SRC/cnaitr.o" failed with exit status 2

Any ideas for this latter?

[rgommers]

And it probably needs to only be done for the NAG compiler, so you need something like this as a guard inside the hook:

if fc.compiler_type == 'nagfor':

[dimpase]

So -dusty is not desired it looks like? Without it, how much sloppy code needs cleaning up? It seems that it’s mostly (only?) sloppy handling of complex data type in scipy/linalg/src/id_dist/src/id_rtrans.f which I mention above. At this point I gave up and switched to using -dusty, so perhaps there is more to it than that.

---

I’d get nagfor installed on Linux, to have more coverage, and also test my branches with gfortran…