Cdo.{rb,py} - Use Ruby/Python to access the power of CDO

Welcome to the scripting interfaces of CDO! This repository contains interfaces for Ruby and Python. If you are not sure whether this is useful or not, please have a look at: Why the .... should I use this???

Currently this package is in a re-design phase. The target is a 2.0 release that will not be compatible with the existing release 1.5.x:

• Write operator chains like methods chains with . as much as possible
• hopefully reduce the number of kwargs keys
• keep the Ruby and Python interface similar
• possibly drop python-2.x support ... I am not sure when to do this best

Releases are distributed via pypi and rubygems:

• Ruby

    gem install cdo (--user-install)

• Python

    pip install cdo (--user)
    conda -c conda-forge install python-cdo

Cdo.{rb,py} requires a working CDO binary and Ruby 2.x or Python 2.7/3.x

PLEASE NOTE: python-2.7 is unmaintained since January 2021 Many dependencies dropped support for 2.7 so I do manual testing with it,only.

Multi-dimensional arrays (numpy for python, narray for ruby) require addtional netcdf-io modules. These are scipy or python-netcdf4 for python and ruby-netcdf for ruby. Because scipy has some difficulties with netcdf, I dropped the support of it with release 1.5.0.

Thx to Alexander Winkler there is also an IO option for XArray.

You can find a lot of examples in the unit tests for both languages. Here are the direct links to the ruby tests and the python tests.

The following describes the basic features for both languages

Befor calling operators, you have to create an object first:

    cdo = Cdo.new   #ruby

    cdo = Cdo()     #python

Please check the documentation for constructor paramaters. I try to have equal interfaces in both languages for all public methods.

By default the cdo-bindings use the 'cdo' binary found in your $PATH variable. To change that, you can

• load a module before calling your script(module command or another package manager like conda or spack)
• use the CDO environment variable to set the path to be used
• use the python/ruby method cdo.setCdo('/path/to/the/CDO/executable/you/want'). By this technique you can create different objects for different CDO versions.

For debugging purpose, both interfaces provide a "debug" attribute. If it is set to a boolian true, the complete commands and the return values will be printed during execution

    cdo.debug = true    #ruby

    cdo.debug = True    #python

The default is false of cause.

    cdo.infov(input: ifile)        #ruby
    cdo.showlevels(input: ifile)

    cdo.infov(input=ifile)         #python
    cdo.showlevels(input=ifile)

    cdo.timmin(input: ifile ,output: ofile)       #ruby

    cdo.timmin(input = ifile,output = ofile)      #python

By default the return value of each call is the name of the output files (no matter if its a temporary file or not)

If the output key is left out, one or more (depending on the operator) temporary files are generated and used as return value(s). In a regular script or a regularly closed interactive session, these files are removed at the end automatically.

    tminFile = cdo.timmin(input: ifile)  #ruby

    tminFile = cdo.timmin(input = ifile) #python

However these tempfiles remain if the session/script is killed with SIGKILL or if the bindings are used via Jupyter notebooks. Those session are usually long lasting and the heavy usage of tempfiles can easily fill the system tempdir - your system will become unusable then. The bindings offer two ways to cope with that

• Set another directory for storing tempfiles with a constructor option and remove anything left in there when you experienced a crash or something like this

   cdo = Cdo(tempdir=tempPath)      #python
   cdo = Cdo.new(tempdir: tempPath) #ruby

• remove all tempfiles created by this or former usage of the cdo-bindings belonging to your current Unix-user with (taking into account user-defined tempdir from above

   cdo.cleanTempDir() #python
   cdo.cleanTempDir   #ruby

Alternatively you can use environment variables to set this. Python's and Ruby's tempfile libraries support the variables 'TMPDIR', 'TEMP' and 'TMP' in their current versions (python-3.8.2, ruby-2.7.0). This feature might be used by administrators to keep users from filling up system directories.

    cdo.remap([gridfile,weightfile],input:   ifile, output: ofile)   #ruby

    cdo.remap([gridfile,weightfile],input => ifile, output => ofile) #python

    cdo = Cdo.new(logging: true, logFile: 'cdo_commands.log') #ruby

    cdo = Cdo(logging=True, logFile='cdo_commands.log')       #python

    cdo.copy(input:  ifile, output:  ofile,options:  "-f nc4")     #ruby

    cdo.copy(input = ifile, output = ofile,options = "-f nc4")     #python

    cdo.splitname(input: ifile.join(' '),
                  output: 'splitTag',
                  env: {'CDO_FILE_SUFFIX' => '.nc'}) #or
    cdo.env = {'CDO_FILE_SUFFIX' => '.nc'}

    cdo.splitname(input = ' '.join(ifiles),
                  output =  'splitTag',
                  env={"CDO_FILE_SUFFIX": ".nc"})   #or
    cdo.env = {'CDO_FILE_SUFFIX': '.nc'}

    t = cdo.fldmin(:input => ifile,:returnArray => true).var('T').get  #rb, version <  1.2.0
    t = cdo.fldmin(:input => ifile,:returnCdf => true).var('T').get    #rb, version >= 1.2.0
    t = cdo.fldmin(:input => ifile,:returnArray => 'T')                #rb, version >= 1.2.0

    t = cdo.fldmin(input = ifile,returnArray = True).variables['T'][:] #py, version <  1.2.0
    t = cdo.fldmin(input = ifile,returnCdf = True).variables['T'][:]   #py, version >= 1.2.0
    t = cdo.fldmin(input = ifile,returnArray = 'T')                    #py, version >= 1.2.0

Other options are so-called masked arrays (use returnMaArray) for ruby and python and XArray/XDataset for python-only: use returnXArray or returnXDataset for that.

*) If you use scipy >= 0.14 as netcdf backend, you have to use following code instead to avoid possible segmentation faults:

    cdf = cdo.fldmin(input = ifile,returnCdf = True)
    temperatures = cdf.variables['T'][:]

More examples can be found in test/cdo-examples.rb and on the homepage

If you do not want to re-compute files, you can set

• the instance attribute 'forceOutput' to false: this will effect all later call of that instance or
• the operator option 'forceOutput' to false: this will only effect this operator call of this instance

For more information, please have a look at the unit tests.

Please use the forum or ticket system of CDOs official web page: http://code.mpimet.mpg.de/projects/cdo

• next 2.0:
  • reduced usage of keywords:
    • many of them just set return type, so they will go to the run() method
    • options only has effect during run of the tool, so this can also go into run()
    • the different input types can be handled in something like input() or
• 1.6.1:
  • replace deprecated pkg_resource
• 1.6.0:
  • merged pull requests regarding threading, signal and tempfile handling
  • added support for xarray/xdataset input AND output at the same time (request by Pauline Millet)
• 1.5.6:
  • slight adoptions for CDO-2.0.0
  • limitted support for python-2.7: many other libs dropped support for it so I can only do limitted testing
  • new API: cdo.config holds a dictionary/hash with built-in CDO features (availble sind CDO-1.9.x), empty otherwise
  • removed cdo.hasLib() and cdo.libsVersion(): relied on unstable output of cdo -V. use cdo.config instead infiles(). This should clean up the lengthy code, which does this currently
• 1.5.4(python-only):
  • bugfix release for @pgierz regarding #30
• 1.5.1(ruby-only):
  • fix some warnings with latest ruby release 2.7.x
• 1.5.0(ruby)/1.5.3(python) API change :
  • simplify the interface:
    • remove returnCdf from constructor, only use it with operator calls
    • remove methods setReturnArray/unsetReturnArray: I fear it's not used anyway, but 'returnArray' in each call
    • remove the optional dependency to scipy since it offers less functionality than netCDF4 and just blows up the code
    • new attributes: hasNetcdf, hasXArray for checking for the respective support
    • fix for cdo-1.9.6: allow non-zero return code for diff operators
  • the 1.5.2 relase for python is identical to 1.5.0 (was testing a new setup.py and version cannot be used twice in pipy)
  • 1.5.1 had building problems with pip based on python2
• 1.4.0 API change :
  • the operators atribute is no longer a list, but a dict (python) or hash (ruby) holding the number of output streams as value
  • finally fix #16 (missing tempfile generation for more than one output streams)
  • fix #19 (thx @pgierz for the input)
• 1.3.6:
  • bugfix for non-finding the CDO binary on some systems
  • fix hasCdo (py)
  • add hasCdo (rb)
• 1.3.5:
  • read/write support for XArray datasets - thx to @pinplex!
  • drop ruby support for 1.9 and older
  • remove module interface from the ruby version
• 1.3.3:
  • return arrays/lists of output files, which are created by split* operators suggestion from Karl-Hermann Wieners 🌊 NOTE: this is done by simple globbing! Any other files with the appropriate name will be included in the list!
  • use six for python2 and 3 compatibility (thanks to @jvegasbsc)
  • drop full support of CDO version older then 1.5.4: undocumented operators in these version will not be callable
  • new keyword for operators which write to stdout: autoSplit. When set, each line will be split with the given value of the keyword to avoid the need for manual splitting. Nested return arrays of (outer) size 1 are flattened. See #11, thx to @beatorizu
• 1.3.2
  • improvened stdout/stderr handling, thx to jvegasbsc
• 1.3.1
  • fix environment handling per call (ruby version)
• 1.3.0
  • require ruby-2.*
  • support for upcomming CDO release 1.7.1
  • improve loggin for ruby
  • introduce logging for python
  • unicode bugfix - thanks to Sebastian Illing (illing2005) [python-only]
• 1.2.7
  • Added class interface for ruby version 2.x, mainly for thread safety
• 1.2.6
  • bugfix for autocompletion in interactive usage [python-only]
• 1.2.5
  • bugfix for environment handling (Thanks philipp) [python-only]
  • add logging [ruby-only]
• 1.2.4
  • support python3: Thanks to @jhamman
  • bugfix for scipy: Thanks to @martinclaus
  • docu fixes: Thanks to @guziy
  • allow environment setting via call and object construction (see test_env in test_cdo.py)
• 1.2.3
  • bugfix release: adjust library/feature check to latest cdo-1.6.2 release
• 1.2.2
  • allow arrays in additions to strings for input argument
  • add methods for checking the IO libraries of CDO and their versions
  • optionally return None on error (suggestion from Alex Loew, python only)
• 1.2.1
  • new return option: Masked Arrays if the new keyword returnMaArray is given, its value is taken as variable name and a masked array wrt to its FillValues is returned contribution for python by Alex Loew
  • error handling: return stderr in case of non-zero return value + raise exception contribution for python from Estanislao Gonzalez
  • autocompletion and built-in documentation through help() for interactive use contribution from Estanislao Gonzalez [python]
  • Added help operator for displaying help interactively [ruby]
• 1.2.0 API change:
  • Ruby now uses the same keys like the python interface, i.e. :input and :output instead of :in and :out
  • :returnArray will accept a variable name, for which the multidimesional array is returned
• 1.1.0 API change:
  • new option :returnCdf : will return the netcdf file handle, which was formerly done via :returnArray
  • new options :force : if set to true the cdo call will be run even if the given output file is presen, default: false

Cdo.{rb,py} makes use of the BSD-3-clause license