Warning not catched

Warning not catched

[Angelo Graziosi]

Consider this simple test case:

program test_warn
  implicit none

  integer, parameter :: DP = selected_real_kind(12,60)
  real(DP), parameter :: Z0 = 0

  integer :: b(5)

  b(1:5) = Z0
end program test_warn

and compile it with:

gfortran -Wall -O2 -c test_warn.f90

Now, only with gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.4) it prints the warning:

   test_warn.f90:9:11:
   b(1:5) = Z0
           1
   Warning: Possible change of value in conversion from REAL(8) to INTEGER(4) at (1) [-Wconversion]

Instead the following more recent versions of gfortran do NOT emit any warning:

1) (on MSYS2, msys2 shell)
Target: x86_64-pc-msys
gcc version 6.3.0 (GCC)

2) (on MSYS2, mingw32 shell)
Target: i686-w64-mingw32
gcc version 7.2.0 (Rev1, Built by MSYS2 project)

3) (on MSYS2, mingw64 shell)
Target: x86_64-w64-mingw32
gcc version 7.2.0 (Rev1, Built by MSYS2 project)

4) (Win10 - Windows Subsystem for Linux)
Target: x86_64-pc-linux-gnu
gcc version 7.1.0 (GCC)

5) (on macOS 10.12 + MacPorts)
Target: x86_64-apple-darwin16
gcc version 6.4.0 (MacPorts gcc6 6.4.0_0)

(all Thread model: posix).

Is this to be expected? (My mind says no..)

Thanks,
 Angelo.

Re: Warning not catched

[Janus Weil]

Hi Angelo,

> Consider this simple test case:
>
> program test_warn
>   implicit none
>
>   integer, parameter :: DP = selected_real_kind(12,60)
>   real(DP), parameter :: Z0 = 0
>
>   integer :: b(5)
>
>   b(1:5) = Z0
> end program test_warn
>
> and compile it with:
>
> gfortran -Wall -O2 -c test_warn.f90
>
> Now, only with gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.4) it prints the warning:
>
>    test_warn.f90:9:11:
>    b(1:5) = Z0
>            1
>    Warning: Possible change of value in conversion from REAL(8) to INTEGER(4) at (1) [-Wconversion]
>
> Instead the following more recent versions of gfortran do NOT emit any warning:
>
> [...]
>
> Is this to be expected? (My mind says no..)

This is PR 82018, for which I submitted a patch just yesterday (see
https://gcc.gnu.org/ml/fortran/2017-09/msg00046.html).

Removing this warning from -Wall was a conscious decision apparently.
My patch adds it to -Wextra at least.

Since there was quite a bit of discussion about this, maybe you can
give us your opinion as a gfortran user: Is this warning useful to
you? Do you think it's suitable for -Wall? Is it sufficient to have it
in -Wextra? (If other users reading this list have an opinion on this
matter, I'd like to hear that as well).

Cheers,
Janus

Re: Warning not catched

[Angelo Graziosi]

Ciao Janus,

> Il 17 settembre 2017 alle 12.37 Janus Weil <janus@gcc.gnu.org> ha scritto:
> 
> [...]
> Since there was quite a bit of discussion about this, maybe you can
> give us your opinion as a gfortran user: Is this warning useful to
> you? Do you think it's suitable for -Wall? Is it sufficient to have it
> in -Wextra? (If other users reading this list have an opinion on this
> matter, I'd like to hear that as well).

I think it is useful in -Wall. In any case it should be at least in -Wextra. Thanks.

Ciao,
 Angelo

Re: Warning not catched

[Thomas Koenig]

Hi Angelo,

> program test_warn
>    implicit none
> 
>    integer, parameter :: DP = selected_real_kind(12,60)
>    real(DP), parameter :: Z0 = 0
> 
>    integer :: b(5)
> 
>    b(1:5) = Z0
> end program test_warn
> 
> and compile it with:
> 
> gfortran -Wall -O2 -c test_warn.f90
> 
> Now, only with gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.4) it prints the warning:
> 
>     test_warn.f90:9:11:
>     b(1:5) = Z0
>             1
>     Warning: Possible change of value in conversion from REAL(8) to INTEGER(4) at (1) [-Wconversion]
> 
> Instead the following more recent versions of gfortran do NOT emit any warning:

There is no change of value in conversion; the compiler has gotten
smart enough to detect this in the meantime.

Consider this, by comparison:

program test_warn
   implicit none
   integer, parameter :: DP = selected_real_kind(12,60)
   real(DP), parameter :: Z0 = 1.2_DP
   integer :: b(5)
   b(1:5) = Z0
end program test_warn
$ !gf
gfortran -Wall a.f90
a.f90:6:11:

    b(1:5) = Z0
            1
Warning: Change of value in conversion from 'REAL(8)' to 'INTEGER(4)' at 
(1) [-Wconversion]

The reason was to remove clutter from -Wall.

Re: Warning not catched

[Angelo Graziosi]

Ciao Thomas
> Il 17 settembre 2017 alle 16.24 Thomas Koenig <tkoenig@netcologne.de> ha scritto:
> There is no change of value in conversion; the compiler has gotten
> smart enough to detect this in the meantime.
> 
> Consider this, by comparison:
> 
> program test_warn
>    implicit none
>    integer, parameter :: DP = selected_real_kind(12,60)
>    real(DP), parameter :: Z0 = 1.2_DP
>    integer :: b(5)
>    b(1:5) = Z0
> end program test_warn
> $ !gf
> gfortran -Wall a.f90
> a.f90:6:11:
> 
>     b(1:5) = Z0
>             1
> Warning: Change of value in conversion from 'REAL(8)' to 'INTEGER(4)' at 
> (1) [-Wconversion]
>

I verified that playing with the test case.. Maybe my preference for that is only matter of style, a sort of coherence using the same type to initialize a variable... Maybe -Wextra is enough. Thanks for clarification.

 Angelo