BC Observatoryunofficial, made by waldo
community postAL Developmentcommunity - not Microsoftreviewed - checked by Opus

AL Runner v2: What Feedback Told Me v1 Was Missing

AL Runner v2 introduces a fundamental architecture change that enables integration testing against real Microsoft and third-party dependencies instead of mocks, moving from type-renaming and stubbing to using BC's own compiler and patching only low-level database operations. The update emphasizes fail-loudly error handling so tests cannot silently pass with wrong results, and introduces a separate language behavior test suite that validates all tests against real Business Central before trusting them.

Open in Claudehttps-stefanmaron-com-posts-al-runner-v2-emit-through-bc--e1f81e05ab.md

post Stefan Maron

Stefan Maron · 2026-08-10 · 1,538 words · about 7 min

"v2 can run against real dependency code, which means it can do integration testing, not just unit testing against a mock."
Show the post hereRead on Stefan Maron

Key points

  • v1 stubbed all dependencies by renaming types, limiting tests to isolated unit tests; v2 compiles through BC’s real compiler and loads actual dependency DLLs for true integration testing
  • v2 throws exceptions immediately for unsupported APIs with clear messages instead of silently falling back to stubs that produce misleading passing tests
  • A separate BusinessCentral.AL.Language.Tests repository runs the full test suite against real BC Cloud sandboxes on multiple versions to ensure tests are correct before the runner is held to them
  • The migration from regex to syntax tree parser discovered 1,600 mishandled CalcFormula conditions across the Base App that had gone unnoticed
  • Early v2 releases (v2.0.0 through v2.1.1) achieved full test passage and added Windows support, auto-provisioning of test toolkit, and fixes for filter corruption in multi-table queries

Quotes

  • “v2 can run against real dependency code, which means it can do integration testing, not just unit testing against a mock.” (This is the core architectural breakthrough that makes v2 practical for real-world extensions where mocking is insufficient)
  • “If your test exercises something out of the runner’s scope, the test fails. It cannot quietly pass.” (Fail-loud behavior prevents false confidence from tests that pass for the wrong reasons, which is a critical reliability improvement)

Context

  • Features: integration testing against real dependencies, in-process XML report rendering, fail-loud error handling, query execution, filter processing, syntax tree based parsing, auto-provisioning with test toolkit
  • Versions: v2.0.0, v2.0.1, v2.1.0, v2.1.1

Source: Stefan Maron, community blog. Summary, key points and quotes are derived (CONTENT-NOTICE.md); read the original for the full text.