How to control/prevent cast between from complex to real variables
in [Fortran]
|
Prev: How to pass function names as arguments when function returns an array?
Next: How to control/prevent cast between from complex to real variables
From: Berfux on 15 Feb 2010 16:09 Hello, consider the following program : program diag implicit none complex :: z real :: x integer :: i z = cmplx(1.5,2.5) print*, "z=",z x = z print*, "x=",x z = x print*,"z=",z i = x print*,"i=",i end program diag In g95 (with option -Wall) I got two warnings : x = z 1 Warning (140): Implicit conversion at (1) may cause precision loss Iin the following compilers (Sun, Intel, gfortran) I could not find such an option. Does it exist ? If not, do you think there is a good reason ? I ask the question because I have to modify a F90 software so that some arguments can be complex instead of real. Such an option in the compiler would help me check I my modifications. Best regards
From: Harald Anlauf on 15 Feb 2010 16:27 On Feb 15, 10:09 pm, Berfux <ber...(a)gmail.com> wrote: > In g95 (with option -Wall) I got two warnings : > > x = z > 1 > Warning (140): Implicit conversion at (1) may cause precision loss The warning are correct. > Iin the following compilers (Sun, Intel, gfortran) I could not find > such an option. Does it exist ? If not, do you think there is a good > reason ? With gfortran, try -Wconversion to see implicit conversions in expressions and assignments. > I ask the question because I have to modify a F90 software so that > some arguments can be complex instead of real. Such an option in the > compiler would help me check I my modifications. Can you be more specific? It looks like you're after something different. The implicit conversions above do not apply to arguments passed to procedures.
From: dpb on 15 Feb 2010 17:34 Berfux wrote: > Hello, > > consider the following program : > > program diag > > implicit none > > complex :: z > real :: x > integer :: i > > z = cmplx(1.5,2.5) > print*, "z=",z > > x = z > print*, "x=",x > > z = x > print*,"z=",z > > i = x > print*,"i=",i > > end program diag > > In g95 (with option -Wall) I got two warnings : > > x = z > 1 > Warning (140): Implicit conversion at (1) may cause precision loss > > Iin the following compilers (Sun, Intel, gfortran) I could not find > such an option. Does it exist ? If not, do you think there is a good > reason ? > > I ask the question because I have to modify a F90 software so that > some arguments can be complex instead of real. Such an option in the > compiler would help me check I my modifications. While I can could see a reason why you'd like to be able to find instances of implicit type conversion as you've shown, I don't see that it necessarily would warrant a warning as it can be intended. Hence I'm not at all surprised many compilers don't include it in their warnings. You might look at Understand for Fortran from Scitools and see if it's clever enough to find such. If not, in my previous usage/evaluation w/ them I found them very amenable to adding features if they weren't very difficult to do. <www.scitools.com> --
From: feenberg on 16 Feb 2010 08:00 On Feb 15, 4:09 pm, Berfux <ber...(a)gmail.com> wrote: > Hello, > > consider the following program : > > program diag > > implicit none > > complex :: z > real :: x > integer :: i > > z = cmplx(1.5,2.5) > print*, "z=",z > > x = z > print*, "x=",x > > z = x > print*,"z=",z > > i = x > print*,"i=",i > > end program diag > > In g95 (with option -Wall) I got two warnings : > > x = z > 1 > Warning (140): Implicit conversion at (1) may cause precision loss > > Iin the following compilers (Sun, Intel, gfortran) I could not find > such an option. Does it exist ? If not, do you think there is a good > reason ? > > I ask the question because I have to modify a F90 software so that > some arguments can be complex instead of real. Such an option in the > compiler would help me check I my modifications. > > Best regards Can you suppress the message by making the conversion explicit? http://gcc.gnu.org/onlinedocs/gcc-3.3.6/g77/RealPart-Intrinsic.html Daniel Feenberg
From: JB on 16 Feb 2010 16:46
On 2010-02-16, feenberg <feenberg(a)gmail.com> wrote: > On Feb 15, 4:09�pm, Berfux <ber...(a)gmail.com> wrote: >> I ask the question because I have to modify a F90 software so that >> some arguments can be complex instead of real. Such an option in the >> compiler would help me check I my modifications. >> >> Best regards > > Can you suppress the message by making the conversion explicit? > > http://gcc.gnu.org/onlinedocs/gcc-3.3.6/g77/RealPart-Intrinsic.html Considering that the OP explicitly stated that he/she is working on a F90 program, perhaps a more appropriate reference would be to the documentation of a compiler that is actually a F90+ compiler, such as g77's successor gfortran: http://gcc.gnu.org/onlinedocs/gcc-4.4.3/gfortran/REAL.html While gfortran does support the g77 REALPART intrinsic, the documentation quite properly discourages it's use instead of the standard REAL intrinsic. While the comment in the g77 manual about using REALPART instead of REAL due to inconsistent behavior of REAL might have made sense at some point in the past, I find it very difficult to believe that any of the widely available production quality F90 compilers today would get this wrong anymore. -- JB |