Crate trycmd

source ·
Expand description

§Snapshot testing for a herd of CLI tests

Treat your tests like cattle, instead of pets

trycmd is a test harness that will enumerate test case files and run them to verify the results, taking inspiration from trybuild and cram.

§Which tool is right

  • cram: End-to-end CLI snapshotting agnostic of any programming language
  • trycmd: For running a lot of blunt tests (limited test predicates)
    • Particular attention is given to allow the test data to be pulled into documentation, like with mdbook
  • snapbox: When you want something like trycmd in one off cases or you need to customize trycmds behavior.
  • assert_cmd + assert_fs: Test cases follow a certain pattern but special attention is needed in how to verify the results.
  • Hand-written test cases: for peculiar circumstances

§Getting Started

To create a minimal setup, create a tests/ with

fn cli_tests() {

and write out your test cases in your .toml files along with examples in your

Run this with cargo test like normal. TestCases will enumerate all test case files and run the contained commands, verifying they run as expected.

To temporarily override the results, you can do:

fn cli_tests() {
        // See Issue #314


To generate snapshots, run

$ TRYCMD=dump cargo test --test cli_tests

This will write all of the .stdout and .stderr files in a dump/ directory.

You can then copy over to tests/cmd the cases you want to test

To update snapshots, run

$ TRYCMD=overwrite cargo test --test cli_tests

This will overwrite any existing .stdout and .stderr file in tests/cmd

To filter the tests to those with name1, name2, etc in their file names, you can run:

cargo test --test cli_tests -- cli_tests trycmd=name1 trycmd=name2...

To debug what trycmd is doing, run cargo test -F trycmd/debug.

§File Formats

For tests/cmd/help.trycmd, trycmd will look for:

  • tests/cmd/
  • tests/cmd/help.out/

Say you have tests/cmd/help.toml, trycmd will look for:

  • tests/cmd/help.stdin
  • tests/cmd/help.stdout
  • tests/cmd/help.stderr
  • tests/cmd/
  • tests/cmd/help.out/


*.trycmd / *.md files are literate test cases good for:

  • Markdown-compatible syntax for directly rendering them
  • Terminal-like appearance for extracting subsections into documentation
  • Reducing the proliferation of files
  • Running multiple commands within the same temp dir (if a *.out/ directory is present)

The syntax is:

  • Test cases live inside of ``` fenced code blocks
    • Everything out of them is ignored
    • Blocks with info strings with an unsupported language (not trycmd, console) or the ignore attribute are ignored
  • $ ” line prefix starts a new command
  • > ” line prefix appends to the prior command
  • ? <status>” line indicates the exit code (like echo "? $?") and <status> can be
    • An exit code
    • success (default), failed, interrupted, skipped
  • All following lines are treated as stdout + stderr

The command is then split with shlex, allowing quoted content to allow spaces. The first argument is the program to run which maps to in the .toml file.


With a [[bin]] like:

fn main() {
    println!("Hello world");

You can verify a code block like:

$ my-cmd
Hello world


For a more complete example, see:


As an alternative to .trycmd, the toml are good for:

  • Precise control over current dir, stdin/stdout/stderr (including binary support)
  • 1-to-1 with dumped results
  • TRYCMD=overwrite support

See full schema: Basic parameters:

  • The name of the binary target from Cargo.toml to be used to find the file path
  • args: the arguments (including flags and option) passed to the binary

Data to pass to stdin.

  • If not present, nothing will be written to stdin
  • If binary = false in *.toml (the default), newlines and path separators will be normalized.
§*.stdout and *.stderr

Expected results for stdout or stderr.

  • If not present, we’ll not verify the output
  • If binary = false in *.toml (the default), newlines and path separators will be normalized before comparing

Eliding Content

Sometimes the output either includes:

  • Content that changes from run-to-run (like time)
  • Content out of scope of your tests and you want to exclude it to reduce brittleness

To elide a section of content:

  • ... as its own line: match all lines until the next one. This is equivalent of \n(([^\n]*\n)*)?.
  • [..] as part of a line: match any characters. This is equivalent of [^\n]*?.
  • [EXE] as part of the line: On Windows, matches .exe, ignored otherwise
  • [ROOT] as part of the line: The root directory for where the test is running
  • [CWD] as part of the line: The current working directory within the root
  • [YOUR_NAME_HERE] as part of the line: See TestCases::insert_var

We will preserve these with TRYCMD=dump and will make a best-effort at preserving them with TRYCMD=overwrite.


When present, this will automatically be picked as the CWD for the command.

.keep files will be ignored but their parent directories will be created.

Tests are assumed to not modify files in *.in/ unless an *.out/ is provided or fs.sandbox = true is set in the .toml file.


When present, each file in this directory will be compared to generated or modified files.

See also “Eliding Content” for .stdout

.keep files will be ignored.

Note: This implies fs.sandbox = true.


For testing command line programs.

  • escargot for more control over configuring the crate’s binary.
  • duct for orchestrating multiple processes.
  • assert_cmd for test cases that are individual pets, rather than herd of cattle
  • assert_fs for filesystem fixtures and assertions.
    • or tempfile for scratchpad directories.
  • rexpect for testing interactive programs.
  • dir-diff for testing file side-effects.

For snapshot testing: