Reporting and profiling

How to produce reports and feed data into them. To read a failing report as an agent (the jq failure-diagnosis workflow), see ginkgo:debugging-failures. Docs: https://onsi.github.io/ginkgo/#generating-machine-readable-reports and https://onsi.github.io/ginkgo/#attaching-data-to-reports.

Console output

In parallel (-p), -v/-vv timelines can't stream — the streams would interleave, so Ginkgo emits each spec's full timeline after that spec completes. → CI flags in ginkgo:ci.

Machine-readable reports

ginkgo -r --json-report=report.json --output-dir=./.reports

• --json-report is native and richest — an array of types.Report, each with SpecReports[] of types.SpecReport (godoc); build tooling against the types package.
• --junit-report=x.xml, --teamcity-report=x, --gojson-report=x.go.json exist for external CI but lose Ginkgo metadata — use them to feed a system, not to diagnose. (JUnit maps a Label("owner:XYZ") to the Owner attribute.)
• --output-dir=DIR collects all files in one place; under -r all suites merge into one report file. --keep-separate-reports writes one PACKAGE_report.json per package instead.
• Every machine-readable report embeds the full -vv timeline regardless of console verbosity. Run the suite quiet and mine the file — you don't need -vv on the console to get rich detail in the report.

Programmatic reporting nodes

CurrentSpecReport() works in any run-phase closure — use it for in-spec conditional diagnostics:

AfterEach(func() {
  if CurrentSpecReport().Failed() {
    GinkgoWriter.Println(libraryClient.DebugLogs())   // expensive diagnostics, only on failure
  }
})

For custom report files, don't roll your own in AfterEach — use reporting nodes:

• ReportAfterEach(func(r SpecReport){...}) / ReportBeforeEach — run for every spec including skipped/pending (unlike AfterEach), after all AfterEaches. A failure here fails the spec. In parallel they run on each process concurrently — never write a shared file from them (you'll get interleaved garbage).
• ReportBeforeSuite / ReportAfterSuite("name", func(r Report){...}) — top-level, and run only on process #1. ReportAfterSuite receives the Report aggregated across all parallel processes — this is the safe place for custom suite-level report files:

ReportAfterSuite("custom report", func(report Report) {
  f, _ := os.Create("report.custom")
  for _, sr := range report.SpecReports {
    fmt.Fprintf(f, "%s | %s\n", sr.FullText(), sr.State)
  }
  f.Close()
})

Reporting nodes can be made interruptible by also accepting SpecContext → ginkgo:timeouts-and-async.

Attaching data to a spec: AddReportEntry

It("is reported", func() {
  AddReportEntry("Publish stats", stats, ReportEntryVisibilityFailureOrVerbose)
})

AddReportEntry(name, value, ...args) attaches arbitrary data to the current spec; it lands in SpecReport.ReportEntries, in all machine-readable reports, and (per visibility) on the console. Visibility enum:

• ReportEntryVisibilityAlways (default) — always printed.
• ReportEntryVisibilityFailureOrVerbose — only on failure or -v (like GinkgoWriter).
• ReportEntryVisibilityNever — never on console, but still in the JSON report.

value is JSON-round-tripped into the report (retrieve via entry.GetRawValue() / json.Unmarshal(entry.Value.AsJSON, ...)). A fmt.Stringer/types.ColorableStringer controls console rendering; pass a pointer to capture later mutations. AddReportEntry decorators (Offset, CodeLocation, ReportEntryVisibility) → ginkgo:decorators.

Recap (owned by ginkgo:writing-specs): GinkgoWriter buffers log output and only prints it on failure (or always under -v); By("step") annotates the timeline. Use these for ad-hoc logging; use AddReportEntry for structured data you want in the report file.

Profiling

ginkgo -r --cover --coverpkg=./... --race --cpuprofile=cpu.out --output-dir=./.profiles

• Coverage: --cover, --coverprofile=NAME (filename only — use --output-dir for a path), --coverpkg=./... to cover beyond the suite's own package, --keep-separate-coverprofiles per-package. --race forces --covermode=atomic.
• Profiles: --cpuprofile / --memprofile / --blockprofile / --mutexprofile (Ginkgo keeps the test binary so you can go tool pprof BINARY PROFILE). --output-dir collects and package-namespaces all profile/coverage assets.
• Focused specs break profiling. A suite with Focus/FIt/FDescribe exits early with a non-zero code, which stops go test from writing the profile. Subset with filters (--focus=, --label-filter=), not Focus → ginkgo:filtering.

See also

• Reading a failing report (jq filters, the panicked-vs-failed trap) → ginkgo:debugging-failures
• The full CI flag set and version-matching → ginkgo:ci
• Decorators (Label, ReportEntryVisibility, Offset) → ginkgo:decorators