Build Error: Arguments of iand have different kind type parameters

[isce-framework]

Build Error: Arguments of iand have different kind type parameters

[piyushrpt]

The solution to this would be something like this:

!c Do this early in the function
integer*2 b1, b2, b3, b4  

...
!c Whenever you want to interpret 4 bytes as big endian int32
b1 = indata(40)
b2 = indata(39)
b3 = indata(38)
b4 = indata(37)

and then use these integer*2 variables in the iand statements.

This error is due to gfortran in gcc9 complying with the standards unlike earlier versions which were lax.

[S0sh0rt]

FYI: I just noticed that the multiplication signs are mostly missing. Looks like the blog is interpreting them as formatting characters. There should be a multiplication "asterisk" before each of the 256s.
Thank you. The lines I removed and added are below.
integer*4 b1, b2, b3, b4 !iyear=iand(indata(40),255)256256*256+iand(indata(39),255)256256+iand(indata(38),255)256+iand(indata(37),255) !idoy=iand(indata(44),255)256256256+iand(indata(43),255)256256+iand(indata(42),255)256+iand(indata(41),255)
!ims=iand(indata(48),255)256256256+iand(indata(47),255)256256+iand(indata(46),255)256+iand(indata(45),255)
b1 = indata(40)
b2 = indata(39)
b3 = indata(38)
b4 = indata(37)
iyear=iand(b1,255)256256256+iand(b2,255)256256+iand(b3,255)256+iand(b4,255)
b1 = indata(44)
b2 = indata(43)
b3 = indata(42)
b4 = indata(41)
idoy=iand(b1,255)256256256+iand(b2,255)256256+iand(b3,255)256+iand(b4,255)
b1 = indata(48)
b2 = indata(47)
b3 = indata(46)
b4 = indata(45)
ims=iand(b1,255)256256256+iand(b2,255)256256+iand(b3,255)*256+iand(b4,255)

[piyushrpt]

Could you issue a Pull Request with these changes?

[S0sh0rt]

I am not a developer and have not worked with Fortran for 30 years. It is probably best for someone else to make the adjustment. Chris

[rtburns-jpl]

I'm far from a fortran expert but would it be more correct to use a _1 suffix to specify that the literals are integer*1s? Changing each 255 to 255_1 fixes the issue for me (using macports gfortran-mp-9)

[S0sh0rt]

rtburns has the best suggestion. It is cleaner and has the smallest change.

[rtburns-jpl]

Great, I'll submit a PR then

[piyushrpt]

test.f90:9:40:

         print *, check, iand(check,255_1)
                                        1
Error: Integer too big for its kind at (1). This check can be disabled with the option -fno-range-check

Would be a little cautious. Here is a test program to confirm:

program testbits
    integer*1 check
    integer i
    integer*1 allbits
    integer*2 shortcopy

    allbits = -1
    do i=-128, 127
        check = i
        shortcopy = i
        print *, check, iand(check,allbits),iand(shortcopy,255)
    end do

end program

Changing allbits to 255_1 still results in negative numbers.

[rtburns-jpl]

Hmm. And unsigned ints don't appear to be standard in fortran.

What do the iands achieve here? If indata is already type integer*1 are they even needed?

[S0sh0rt]

My question also: The iands mask off the sign bit extended from the 8 bit value. since it needs to be treated as unsigned? I do recall that unsigned bytes were always a difficulty with Fortran.
Note that I had to change I*2 to I*4 in the initial fix I posted.

[piyushrpt]

the problem is that fortran doesnt support unsigned ints. The information is read in as a byte stream (integer1 which is signed) and then converted to 4byte unsigned ints representing time tags. The iand converts values to unsigned equivalents. 255 cannot be stored in an integer1. An integer*2 can store the result of an iand operation.

[rtburns-jpl]

How about just making the byte mask -1_1?

Edit: nevermind, looks like you can't assume twos-complement in fortran