verify_output: Verify output

Description

This function is superseded in favour of expect_snapshot() and friends.

This is a regression test that records interwoven code and output into a file, in a similar way to knitting an .Rmd file (but see caveats below).

verify_output() is designed particularly for testing print methods and error messages, where the primary goal is to ensure that the output is helpful to a human. Obviously, you can't test that with code, so the best you can do is make the results explicit by saving them to a text file. This makes the output easy to verify in code reviews, and ensures that you don't change the output by accident.

verify_output() is designed to be used with git: to see what has changed from the previous run, you'll need to use git diff or similar.

Usage

verify_output(
  path,
  code,
  width = 80,
  crayon = FALSE,
  unicode = FALSE,
  env = caller_env()
)

Arguments

path

Path to record results.

This should usually be a call to test_path() in order to ensure that the same path is used when run interactively (when the working directory is typically the project root), and when run as an automated test (when the working directory will be tests/testthat).

code

Code to execute. This will usually be a multiline expression contained within {} (similarly to test_that() calls).

width

Width of console output

crayon

Enable cli/crayon package colouring?

unicode

Enable cli package UTF-8 symbols? If you set this to TRUE, call skip_if(!cli::is_utf8_output()) to disable the test on your CI platforms that don't support UTF-8 (e.g. Windows).

env

The environment to evaluate code in.

Syntax

verify_output() can only capture the abstract syntax tree, losing all whitespace and comments. To mildly offset this limitation:

Strings are converted to R comments in the output.
Strings starting with # are converted to headers in the output.

CRAN

On CRAN, verify_output() will never fail, even if the output changes. This avoids false positives because tests of print methods and error messages are often fragile due to implicit dependencies on other packages, and failure does not imply incorrect computation, just a change in presentation.