[flang] incorrect iostat for failed list-separated read into real

[Inker2401]

There appears to be some (possibly) non-standard conforming behaviour concerning the reading of an internal file (string) containing .. into a real variable. Consider the following program:

program flang_read_dots
  implicit none

  character(len=2) :: str
  real :: x
  complex :: z
  integer  :: n, ierr

  str='..'

  ! With status checks
  read(str,*,iostat=ierr) x ! ierr=0 (success) and then x is read as 0.0 (non-standard behaviour)
  print *, 'Read real number ierr, val =', ierr, x

  read(str,*,iostat=ierr) n ! ierr is non-zero as expected as cannot read to integer
  print *, 'Read integer number ierr, val =', ierr, n

  read(str,*,iostat=ierr) z ! ierr is non-zero as expected as cannot read to complex
  print *, 'Read complex number ierr, val=', ierr, z

  ! Without status checks
  read(str,*) x ! read succeeds (incorrectly!)
  print *, 'Read real number (no stat check) val =', x

  read(str,*) n ! read does not succeed, runtime error raised
  print *, 'Read integer number (no stat check) val =', n

  read(str,*) z ! read does not succeed, runtime warning
  print *, 'Read complex number (no stat check) val =', z
end program flang_read_dots

The str containing .. when read into x gives 0.0 with iostat=0 for success (likewise, iostat is omitted, no runtime error is raised). This appears to be possibly non-standard conforming behaviour, as various other compilers such as ifort, Cray Fortran and gfortran all raise a runtime error (and will also return non-zero iostat codes).

I suspect it has to do with the interpretation decimal point as if str is initialised to hi, then a non-zero iostat value is returned as expected. Similarly, if the decimal="comma" argument is passed into the read statement, a non-zero value for iostat is also returned.

Compiler explorer link: https://godbolt.org/z/9doTaGqKT

[llvmbot]

@llvm/issue-subscribers-flang-runtime

Author: Visagan Ravindran (Inker2401)

There appears to be some (possibly) non-standard conforming behaviour concerning the reading of an internal file (string) containing `..` into a real variable. Consider the following program:

program flang_read_dots
  implicit none

  character(len=2) :: str
  real :: x
  complex :: z
  integer  :: n, ierr

  str='..'

  ! With status checks
  read(str,*,iostat=ierr) x ! ierr=0 (success) and then x is read as 0.0 (non-standard behaviour)
  print *, 'Read real number ierr, val =', ierr, x

  read(str,*,iostat=ierr) n ! ierr is non-zero as expected as cannot read to integer
  print *, 'Read integer number ierr, val =', ierr, n

  read(str,*,iostat=ierr) z ! ierr is non-zero as expected as cannot read to complex
  print *, 'Read complex number ierr, val=', ierr, z

  ! Without status checks
  read(str,*) x ! read succeeds (incorrectly!)
  print *, 'Read real number (no stat check) val =', x

  read(str,*) n ! read does not succeed, runtime error raised
  print *, 'Read integer number (no stat check) val =', n

  read(str,*) z ! read does not succeed, runtime warning
  print *, 'Read complex number (no stat check) val =', z
end program flang_read_dots

The str containing .. when read into x gives 0.0 with iostat=0 for success (likewise, iostat is omitted, no runtime error is raised). This appears to be possibly non-standard conforming behaviour, as various other compilers such as ifort, Cray Fortran and gfortran all raise a runtime error (and will also return non-zero iostat codes).

I suspect it has to do with the interpretation decimal point as if str is initialised to hi, then a non-zero iostat value is returned as expected. Similarly, if the decimal="comma" argument is passed into the read statement, a non-zero value for iostat is also returned.

Compiler explorer link: https://godbolt.org/z/9doTaGqKT

[klausler]

Fortran '77 allows . with no digits as real input, and this usage appears in the FCVS, so we support it.