CWL Examples

CWL and job files are written in the YAML format. All indentations are made using double spaces.

Wrapping a console tool

Without docker

A command line tool requires a baseCommand which is used to call the tool. This can be the name of the tool, if it is added to the PATH, or a relative/absolute path to the executable (.exe). If the basecommand consists of multiple words, it can be written as a list: [my, Tool]. The next part is the inputs section. It consists of a variable number of arguments required for the tool to run. For each input you can specify the type, position and prefix of the input. The last part is the outputs section. Multiple outputs can be specified by name, type and their location.

cwlVersion: v1.2
class: CommandLineTool
baseCommand: myTool
inputs:
  input1:
    type: File
    inputBinding:
      position: 1
  input2:
    type: Directory
    inputBinding:
      position: 2
      # prefix is optional
      prefix: -i
outputs:
  myFileOutput:
  type: File
  outputBinding:
      # this returns a specific file
    glob: myOutput.txt
  myFileArrayOutput:
    type: File[]
    outputBinding:
    # this returns all files with the extension .txt
    glob: $(runtime.outdir)/*.txt
  myDirectoryOutput:
    type: Directory
    outputBinding:
      # this returns a specific directory
      glob: $(runtime.outdir)/myDirectory

There are several possibilities to retrieve the output of a tool. Common options would be File, File[], or Directory. Depending on your tool or script, the output varies. If your tool returns a fixed number of known files, you should specify them as File. If it has a variable number of files with a known extension, you should specify them as File[]. If it has a variable output structure, you should specify it as Directory. For usage in workflows and provenance tracking, if your tool or script allows it, it is recommended to use File as the output type. This way, several output files can be specified and used in the following workflow steps.

With a docker container

To ensure reproducibility of the tool’s execution on any system, it is recommended to pack it in a docker container. The docker container is automatically pulled with a reference to the container under either requirements or hints. Additional requirements, such as network access, can be specified as well.

cwlVersion: v1.2
class: CommandLineTool
hints:
  DockerRequirement:
    dockerPull: address/to/my/docker
requirements:
  - class: NetworkAccess
    networkAccess: true
baseCommand: myTool
inputs:
  input1:
    type: File
    inputBinding:
      position: 1
  input2:
    type: Directory
    inputBinding:
      position: 2
      # prefix is optional
      prefix: -i
outputs:
  myFileOutput:
  type: File
  outputBinding:
      # this returns a specific file
    glob: myOutput.txt
  myFileArrayOutput:
    type: File[]
    outputBinding:
    # this returns all files with the extension .txt
    glob: $(runtime.outdir)/*.txt
  myDirectoryOutput:
    type: Directory
    outputBinding:
      # this returns a specific directory
      glob: $(runtime.outdir)/myDirectory

With a fixed script file

Oftentimes, individual analysis is done within scripts for the flexibility, instead of tools with fixed tasks. In this case, it is recommended to encode the script as a fixed part of the CWL description. The script is then accessible in the CWL description and can be called as part of the baseCommand. It can be a script that functions as a command line and still expects inputs, or as a self contained analysis without further input requirements.

cwlVersion: v1.2
class: CommandLineTool
hints:
  DockerRequirement:
    dockerPull: address/to/my/docker
requirements:
  - class: InitialWorkDirRequirement
    listing:
      - entryname: myAnalysis.script
        entry:
          $include: myAnalysis.script
  - class: NetworkAccess
    networkAccess: true
baseCommand: [run, myAnalysis.script]
inputs:
  input1:
    type: File
    inputBinding:
      position: 1
  input2:
    type: Directory
    inputBinding:
      position: 2
      # prefix is optional
      prefix: -i
outputs:
  myOutput:
    type: File
    outputBinding:
      glob: "result.csv"

Example

With a fixed script in a mounted arc

When scripting, it is convenient to work within the environment of the ARC (e.g. location of files, writing results in the runs folder and so on). When the ARC is mounted within the CWL process, the script can be ran in that environment and the corresponding runs folder is then returned as an output.

cwlVersion: v1.2
class: CommandLineTool
hints:
  DockerRequirement:
    dockerPull: address/to/my/docker
requirements:
  - class: InitialWorkDirRequirement
    listing:
      # this specifies the name of the root folder
      - entryname: arc
        entry: $(inputs.arcDirectory)
        writable: true
  - class: NetworkAccess
    networkAccess: true
baseCommand: [run, ./arc/workflows/myWorkflow/myAnalysis.script]
inputs:
  # the arc root directory is given as an input, but not called by the process
  arcDirectory:
    type: Directory
  input1:
    type: File
    inputBinding:
      position: 1
  input2:
    type: Directory
    inputBinding:
      position: 2
      # prefix is optional
      prefix: -i
outputs:
  myOutput:
    type: File
    outputBinding:
      glob: "./arc/runs/myRun/result.csv"

Example

Within an ARC with a devcontainer

Within the context of an ARC, researches often work within devcontainers or the ARC environment. CWL is able to replicate this workflow under the premise, that in the end everything can be executed in one go by including the Dockerfile of the devcontainer. The entire arc directory can be mounted into the working directory of the CWL process as well, making the script for the devcontainer and CWL process identical. This enables explorative work in scripts which can then be executed with CWL after completion without much overhead.

cwlVersion: v1.2
class: CommandLineTool
hints:
    dockerImageId: "devcontainer"
    dockerFile: {$include: "devcontainer/Dockerfile"}
requirements:
  - class: InitialWorkDirRequirement
    listing:
      # this specifies the name of the root folder
      - entryname: arc
        entry: $(inputs.arcDirectory)
        writable: true
  - class: NetworkAccess
    networkAccess: true
baseCommand: [run, ./arc/workflows/myWorkflow/myAnalysis.script]
inputs:
  # the arc root directory is given as an input, but not called by the process
  arcDirectory:
    type: Directory
  input1:
    type: File
    inputBinding:
      position: 1
  input2:
    type: Directory
    inputBinding:
      position: 2
      # prefix is optional
      prefix: -i
outputs:
  myOutput:
    type: File
    outputBinding:
      glob: "./arc/runs/myRun/result.csv"

The Dockerfile should only include operations that reference resources that are available online or within the baseimage. COPY operations that point to local files for example won’t work in the context of CWL. If they are necessary for the execution in the devcontainer context (e.g. configuration for editors), but not the execution of the script, they can be prefixed with a * to make the execution of the operation optional.

Example

Workflows

Workflows can connect multiple command line tools, for example. It is possible to use the output of a tool as an input for the following tool and return them as intermediate results as well.

cwlVersion: v1.2
class: Workflow

requirements:
  MultipleInputFeatureRequirement: {}

inputs:
  inputTool1_1: File
  inputTool1_2: Directory
  inputTool2_1: File

steps:
  myTool1:
    run: path/to/myTool1.cwl
    in:
      input1: inputTool1_1
      input2: inputTool1_2
    out: [myOutput1]
  myTool2:
    run: path/to/myTool2.cwl
    in:
      input1: inputTool2_1
      # direct reference to the output of myTool1. This creates a dependency.
      input2: myTool1/myOutput1
    out: [myOutput2]
outputs:
  outputTool1:
    type: File
    outputSource: myTool1/myOutput1
  outputTool2:
    type: Directory
    outputSource: myTool2/myOutput2