Unable to read string attributes larger than some length

[AgilentGCMS]

I have two netcdf files, oco2_l4_regions_patra.nc and vpp_mask_ecoregions.nc. They both have a global attribute REGION_NAMES, which is a string. Only difference is that that string is significantly longer in the second file than in the first. Say I use the following program get_long_attribute.F90 to read that attribute into a fortran variable.

program main
  use netcdf
  implicit none

  integer :: nc_id, nc_stat, total_length
  character(len=:), allocatable :: all_region_names
  !character(len=*), parameter :: input_filename = 'oco2_l4_regions_patra.nc'
  character(len=*), parameter :: input_filename = 'vpp_mask_ecoregions.nc'

  nc_stat = nf90_open(input_filename, NF90_NOWRITE, nc_id)
  if (nc_stat /= NF90_NOERR) then
     write(*,'("Opening file failed")')
     stop
  end if

  nc_stat = nf90_inquire_attribute(nc_id, NF90_GLOBAL, 'REGION_NAMES', len=total_length)
  if (nc_stat /= NF90_NOERR) then
     write(*,'("Getting attribute length failed")')
     stop
  end if
  write(*,'("length of attribute = ", i0)') total_length

  allocate(character(total_length) :: all_region_names)
  nc_stat = nf90_get_att(nc_id, NF90_GLOBAL, 'REGION_NAMES', all_region_names)
  if (nc_stat /= NF90_NOERR) then
     write(*,'("Reading attribute failed")')
     stop
  end if
  write(*,'("all_region_names = ", a)') trim(all_region_names)

end program main

For the file oco2_l4_regions_patra.nc, this correctly reads the 2093 character string. However, for the file vpp_mask_ecoregions.nc, it incorrectly reads the length as 1, and the call to nf90_get_att fails. Is there a limit on the attribute length or size that the fortran API can read? The files were written with the C API, so I assume this is not a fundamental limit of the netcdf format.

• netcdf C version -- 4.9.3
• netcdf Fortran version -- 4.6.2
• compiler version -- Intel oneapi 2024.2.1
• compile line -- ifx $(nf-config --fflags) get_long_attribute.F90 $(nf-config --flibs)
• OS -- Red Hat Enterprise Linux 8.10

[AgilentGCMS]

I realize after some digging that this is basically #181 . Both the files above were made with python-netcdf4, which apparently automatically chooses the attribute type based on... something. So in oco2_l4_regions_patra.nc it wrote a character array, while in vpp_mask_ecoregions.nc it wrote a string, which isn't supported by the Fortran interface because of #181 . Bummer, really.