experimental_shell_command target allows you to run any command during a Pants execution, for the purpose of modifying or creating files to be used by other targets, or its (idempotent: see below) side-effects when accessing services over the network.
experimental_shell_command( command="./my-script.sh download some-archive.tar.gz", tools=["curl", "env", "bash", "mkdir", "tar"], outputs=["files/"], dependencies=[":shell-scripts", ":images"] ) shell_library(name="shell-scripts") files(name="images", sources=["*.png"])
#!/usr/bin/env bash case "$1" in download) echo "Downloading $2..." curl https://my-storage.example.net/blob/$2 -O mkdir files && tar xzf $2 -C files ;; *) echo "Usage: $0 [download|...]" ;; esac
command field is passed to
bash -c <command>, with any files from listed
dependencies in the working directory tree. Any executable tools that might be used must be specified in the
tools field, in order to be available on the
PATH while executing the command.
The command is limited to operating on the specific set of input files provided as dependencies, and only produces output files for other targets to consume. It is not possible to mutate any file in the workspace.
In case there are resulting files that should be captured and passed to any consuming targets, list them in the
outputs field. To capture directories, simply add the path to the directory, with a trailing slash (as in the example
The shell command may be cancelled or retried any number of times, so it is important that any side effects are idempotent. That is, it should not matter if it is run several times, or only partially.
experimental_run_shell_command target runs directly in your workspace, without sandboxing.
This target type allows you to formalize the Pants dependencies of shell scripts, and track when their impact on your workspace might have changed. But since its outputs cannot be captured, it must be a root target in your build graph (i.e.: it may not be consumed by other targets).
Updated 13 days ago