Skip to main content

Leaving ProtoTest

No lock in was ever the goal. If the foundation stops fitting, you can take it out one test at a time. There is no migration tool. The steps below are the whole runbook.

Remove the host​

Delete the setup class that builds the host and the package references your suite no longer needs. Tests that lose the host lose Proto.Context, the shared lifecycle, and the trace. Remove them in the same commit so nothing references a host that no longer exists.

Swap the attributes and assertions back​

Each test changes the same way:

  1. Swap [ProtoTest] back for the runner attribute ([Test], [Fact], and the rest).
  2. Remove [Application] and the other suite attributes; move the setup they owned back into fixtures or helpers.
  3. Replace Proto.Context clients with plain clients (an HttpClient, the container client, the driver) and the Should assertions with the library you used before.

One test at a time keeps the suite green while you go. The comparison page lists what you give up at each step: the shared lifecycle first, then the trace, then the coverage.

Keep the traces and reports​

What you recorded stays yours. .prototrace files open in the static viewer with no account and no service to shut off. HTML and JSON reports are plain files beside them. Keep a reader from the same era as the traces: the compatibility matrix states which reader opens which version.

What stays coupled​

Two things follow you out:

  • Era locked readers. The library and the CLI read spans 2.x and state 1.x; the viewer also reads state 2.x. A trace outside those versions needs a reader from its own era, so archive the viewer build or the CLI version with traces you must keep.
  • OpenTelemetry gaps. The backend export carries spans and events only. Large values, gate verdicts, run identity and environment, and target resolutions stay archive only (the full list). If dashboards depend on those facts, move them before you switch the export off.

The honest effort​

The cost is linear in test count: each test pays the attribute, client, and assertion swap above, plus whatever shared setup moves back into helpers. A small suite converts in an afternoon. A large one converts incrementally, one fixture at a time, with both styles running side by side until the last host is gone.