Pretty JSON with discovery's keys in their documented order.

aeson orders an object's keys by its own key map, which is alphabetical — so without this a suite would print discoverCommand before name. The order below is the one list-test-suites.schema.json documents and the shell tool emits, which is also the order that reads best: identity first, then classification, then the commands. Keys not listed keep aeson's order and sort after the listed ones.

Single-line JSON, for piping into jq or another process.

Key order is aeson's here, not the documented one: a JSON object is unordered by definition, every consumer parses it, and encodeJsonPretty is the variant meant to be read.

One suite name per line — the default suites output.

Just the names, with no decoration, so the common uses compose: pbt-cli suites | wc -l, or feeding the list straight back into pbt-cli run. Compatibility is expressed by filtering (--compatible-only) rather than by annotating, so the output stays a clean list of arguments.

A table of every discovered suite, grouped by project file.

The COMPAT column is the answer to "which suites can I stream and discover tests in", which is the question this command exists to answer, so it gets a symbol rather than another word of prose.

The legacy tab-separated format: suite \t packageDir \t mainPath \t entryPoint \t hsSourceDirs.

Kept byte-compatible with list-test-suites.sh --tsv so existing editor integrations and shell pipelines keep working: mainPath is relative to the scanned root (not to the package) and the source dirs are ;-joined so the column count is fixed even for a multi-dir stanza.

The test tree as nested groups, with each test's id.

The id is what --test-id takes, so it is shown first and unpadded: the point of running discovery is usually to pick ids to re-run.

One line per test: id<TAB>fullpathname. Convenient for shell loops.

renderEventWith with no name index: a test_done whose description is empty renders without one.

A one-line human rendering of a streaming event, or Nothing for events with nothing to say on a console.

test_started is dropped rather than printed: it is immediately followed by the test_done line for the same test, and echoing both doubles the output for no added information. test_trace and test_progress are likewise suppressed — they are for tooling, and a trace payload is far too large for a terminal line.

A banner naming the suite whose events follow.

run --stream invokes cabal once per suite, so without this the events of six suites would arrive as one undifferentiated list of PASS lines.

Names for every test in a suite_started event, keyed by id.

Feed this to renderEventWith. Tasty providers differ in whether they set a test_done description — the shimmed Convex.Tasty.HUnit ones do, plain Test.Tasty.HUnit ones send "" — so without the tree a live run would print a column of anonymous PASS lines.

Add a suite field to an event, naming the suite that produced it.

run --json can cover many suites, and the event schema carries no suite identity — a consumer would see several suite_started events with no way to tell them apart. The streaming-events schema does not set additionalProperties: false, so an added field is schema-valid, and a consumer that ignores unknown fields is unaffected.

A non-object event (which the schema does not produce) is returned unchanged.