Compiler bug with fixed form wide source and continued lines

While compiling some old source code in fixed form with many continuation lines and source lines wider than 72 columns, I ran into perplexing responses from FTN95 8.51 (32 and 64 bit). I have prepared a small example to demonstrate the bug.

      subroutine Jacobian_deg(coef)

      double precision x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,
     1                 x15,x16,x17,x18,x19,x20,
     2                 y1,y2,y3,y4,y5,y6,y7,y8,y9,y10,y11,y12,y13,y14,
     3                 y15,y16,y17,y18,y19,y20,
     4                 z1,z2,z3,z4,z5,z6,z7,z8,z9,z10,z11,z12,z13,z14,
     5                 z15,z16,z17,z18,z19,z20

      double precision coef(115)
! code to evaluate x1-x20, y1-y20, z1-z20 removed for brevity
      coef(7)=   (x1*y10*z11) - (x1*y11*z10)   - (x10*y1*z11)           + (x10*y11*z1) !only long line in file
     1         - (x11*y10*z1) + (x1*y10*z12)   - (x1*y12*z10)
     2         + (x12*y1*z10) - (x12*y10*z1)   + (x1*y10*z13)
     3         + (x10*y13*z1) + (x13*y1*z10)   - (x13*y10*z1)
     4         - (x10*y1*z14) + (x10*y14*z1)   + (x14*y1*z10)
     5         + (x1*y11*z14) - (x1*y14*z11)   + (x1*y15*z10)
     6         - (x11*y1*z14) + (x11*y14*z1)   + (x14*y1*z11)
     7         + (x15*y10*z1) - (x1*y10*z16)   + (x1*y16*z10)
     8         - (x16*y1*z10) + (x16*y10*z1)   + (x1*y13*z14)
     9         + (x13*y14*z1) + (x14*y1*z13)   - (x14*y13*z1)
     1         - (x12*y1*z16) + (x12*y16*z1)   + (x16*y1*z12)
     2         - (x1*y15*z14) - (x14*y1*z15)   + (x14*y15*z1)
     3         + (x1*y14*z16) - (x1*y16*z14)   - (x14*y1*z16)
     4         - (x16*y14*z1) - (x1*y14*z17)   + (x1*y17*z14)
     5         - (x17*y1*z14) + (x17*y14*z1)   + (x1*y14*z18)
     6         + (x14*y18*z1) + (x18*y1*z14)   - (x18*y14*z1)
     7         - (x1*y17*z16) + (x1*y19*z14)   + (x14*y1*z19)
     8         + (x16*y17*z1) + (x17*y1*z16)   - (x17*y16*z1)
     9         - (x1*y16*z18) + (x1*y18*z16)   + (x16*y1*z18)
     1         + (x18*y16*z1) + (x1*y16*z19)   - (x1*y19*z16)
     2         + (x19*y1*z16) - (x19*y16*z1)   - (x10*y11*z12)
     3         - (x11*y12*z10) - (x12*y10*z11) + (x12*y11*z10)
     4         - (x12*y10*z13) + (x12*y13*z10) + (x13*y10*z12)

      end subroutine Jacobian_deg

When compiled with ftn95 /error_num /64 /ignore 197 298 c7b.f, a warning is issued for the truncated line. If the /no_truncate option is added, we are told:

R:\lang\FTN95>ftn95 /no_trunc /error_num /64 /ignore 197 298 c7b.f
[FTN95/x64 Ver. 8.51.0 Copyright (c) Silverfrost Ltd 1993-2019]
0012)       coef(7)=   (x1*y10*z11) - (x1*y11*z10)   - (x10*y1*z11)           + (x10*y11*z1) !only long line in file
*** Error 1100: Line is longer than 72 characters (because of /NO_TRUNCATE)
    1 ERROR  [<JACOBIAN_DEG> FTN95 v8.51.0]
*** Compilation failed

Gfortran with the -ffixed-line-length-132 option compiles this with no error.

The FTN95 online documentation says that up to 40 continuation lines can be used. I found while working with the larger source code that it accepts many more, but because of the above bug issued many confusing messages if /no_truncate was specified.

I can take a look at this but I suspect that there may be a missunderstanding in what /no_truncate does. Maybe /wide_source is intended.

Thanks, Paul, you are correct: I misunderstood the meaning of the option.

Sorry!