Testing

ripgrep has two complementary test layers:

1. Unit tests inside each crate, exercising library APIs.
2. Integration tests under tests/, exercising the compiled rg binary as a black box.

The full suite is run via cargo test --all.

Integration tests

Integration tests are wired through tests/tests.rs, which simply pulls in every other test file. The actual test bodies live in:

File	What it covers
tests/feature.rs	The bulk of CLI behaviour: flag combinations, exit codes, output formatting
tests/regression.rs	Tests captured from past bug reports; ~54K of code
tests/json.rs	The --json output format
tests/multiline.rs	The -U/--multiline mode
tests/binary.rs	Binary-file detection and --text/--binary
tests/misc.rs	Anything that doesn't fit elsewhere
tests/util.rs	The harness: builds rg invocations, captures stdout/stderr, asserts
tests/data/	Canned haystacks (e.g., sherlock) used by tests
tests/macros.rs	The rgtest!/sherlock! macros
tests/hay.rs	Constants for the canned haystacks

The rgtest! macro

Most tests look like this:

rgtest!(my_test, |dir: Dir, mut cmd: TestCommand| {
    dir.create("haystack", "foo\nbar\n");
    cmd.arg("foo").assert_eq("haystack:foo\n");
});

tests/util.rs defines:

• Dir: a temporary directory for each test, auto-cleaned at the end.
• TestCommand: a wrapper around std::process::Command that knows how to find the rg binary and assert against its output.

Helpers like assert_eq, assert_err, lines, stdout, and stderr make the assertions readable.

sherlock! macro

Many tests use the canned SHERLOCK haystack (Sherlock Holmes prose, defined in tests/hay.rs). sherlock! wraps rgtest! and pre-populates the directory with it.

Unit tests

Each crate carries its own unit tests inline. Notable suites:

• crates/regex/src/literal.rs — literal extraction, the source of historical correctness bugs.
• crates/searcher/src/lib.rs and crates/searcher/src/searcher/mod.rs — line stepping, mmap vs. buffered, encoding.
• crates/ignore/src/gitignore.rs and crates/ignore/src/walk.rs — gitignore precedence rules.
• crates/globset/src/lib.rs and crates/globset/src/glob.rs — glob compilation and matching.
• crates/printer/src/standard.rs and crates/printer/src/json.rs — printer formatting (extensive).

crates/searcher/src/testutil.rs provides a generic harness (SearcherTester) used by both unit tests and the printer's tests. It runs the same query through every relevant configuration (mmap on/off, multi-line on/off, line numbers on/off) and compares against expected output.

Running a subset

# Single test by name substring
cargo test --test integration feature_glob

# Whole crate
cargo test -p grep-searcher

# A specific module
cargo test -p grep-printer standard

Snapshot data

There are no snapshot tests in the conventional sense. Expected output is inlined in the test source as raw strings. This is consistent across the project; do not introduce a snapshot crate.

Performance regressions

There is no in-tree CI benchmark gate. The external benchsuite/ directory contains a benchmark harness that the maintainer runs by hand against new releases. If your change might affect performance, run benchsuite/ before and after locally and include the numbers in your PR description.