UnitAutogen

Author(s)

• Munaf Ibrahim Khatri

Tags

SQL SQLServer tSQLt Coverage CI-CD Cobertura JUnit AzureDevOps Testing UnitTest CodeCoverage BranchCoverage DatabaseTesting AutomatedTesting

Functions

Install-UnitAutogenDatabase Invoke-UnitAutogen Export-CoverageCoberturaXml Export-TestResultsJunitXml Export-CoverageHtmlReport

Dependencies

This module has no dependencies.

Release Notes

## v0.9.12 (beta) — 2026-06-04

Two fixes from broad-database validation on WideWorldImporters (both v0.9.11 fixes
held there - 0 failures, no doomed transactions):

- Result-row baseline now quotes JSON keys, so result columns with spaces or special
 characters in their names (e.g. "Quantity On Hand") work. Previously an unquoted
 OPENJSON path raised "JSON path is not properly formatted" (Msg 13607) and the test
 errored. Common in real-world schemas; AdventureWorks' tight names never hit it.
- Functions/procedures declared WITH EXECUTE AS OWNER (or CALLER/SELF/'user') now
 build their coverage shadow correctly. The header splitter was matching the AS
 inside "EXECUTE AS OWNER" and failing with "Incorrect syntax near 'OWNER'". A
 self-contained EXECUTE AS OWNER function now reaches 100% line / 100% branch.

## v0.9.11 (beta) — 2026-06-04

Fix: the v0.9.10 schema-bound cleanup could doom the test transaction, turning
whole groups of procedures into all-errors (0% coverage) on real databases.

- Cause: the schema-bound dependent walk in SafeFakeTable used a recursive query
 with no cycle guard. A table with a persisted computed column built on a
 WITH SCHEMABINDING function references ITSELF as schema-bound (e.g. AdventureWorks
 Sales.Customer.AccountNumber), so the walk hit the recursion limit and errored.
 tSQLt runs tests with XACT_ABORT ON, under which that error dooms the transaction,
 so every later table-fake then failed with "the current transaction cannot be
 committed".
- Fix: the cleanup now runs under SET XACT_ABORT OFF (auto-restored on exit) so it
 can never doom the test, and the walk ignores the self-edge and guards against
 cycles. Validated on a clean AdventureWorks2025: 0 errors across the testable
 procedures and functions.

Also: result-set shape characterization no longer compares IsNullable. SQL Server's
nullability inference for literal/computed result columns is unstable (flips across
recompiles/builds), which produced false-positive "shape drift" failures. Column
count, order, names, types and sizes are still asserted in full.

## v0.9.10 (beta) — 2026-06-03

Fix (reported by the community): "Object cannot be renamed because the object
participates in enforced dependencies" (Msg 15336) aborted generation/testing.

- Cause: tSQLt.FakeTable renames the table to isolate it, and SQL Server blocks the
 rename when the table is referenced by a SCHEMA-BOUND object - an indexed view, a
 WITH SCHEMABINDING view/function, or a schema-bound computed column. (Not plain
 foreign keys - tSQLt handles those.)
- Fix: TestGen.SafeFakeTable now drops those schema-bound dependents before faking;
 tSQLt's per-test transaction rolls back and restores them automatically (no
 permanent change). Predicate-branch tests now route through SafeFakeTable too.
- Limitation: if the procedure itself uses a schema-bound object over a faked table,
 that proc is still not auto-testable (clear-failure rather than a cryptic abort).

## v0.9.9 (beta) — 2026-06-03  (ships the "v0.13" in-database parser + fixes; first auto-published release)

Fixes on top of the SSMS-native parser work below:

- De-dup branch tests: a procedure with data-shape gates no longer emits the
 redundant legacy "executes <branch> path" smoke-SKIP tests alongside the real
 seeded predicate-branch tests. When the in-database parser is active, those legacy
 smoke-only fallbacks are rolled back (the predicate-branch tests already cover every
 gate). AssessCustomer: 15 tests (3 skipped) -> 12 tests, 12 pass, 0 skip, still
 100% line + 100% branch.
- Hardened connection-recovery resync in GenerateAndCoverDatabase. Running the
 in-session SQLCLR parser right before a sweep can trip SQL Server's idle-connection
 recovery ("the connection was recovered ... valid rowcount") on the first object's
 generation; the resync probe is now a swallowed real round-trip that reliably
 absorbs that first-query penalty (was a bare assignment that did not).
- Coverage teardown resilience: RunCoverage now self-heals a proc left stranded by
 an interrupted/killed run (recovers it from its _orig backup) BEFORE doing anything
 else, and its restore is wrapped + idempotent so a soft error can never strand the
 proc as _orig. It never drops the only copy of the real body, and aborts cleanly if
 a proc is genuinely missing rather than cascading. (No more "the proc disappeared.")

SSMS-native predicate parser — ONE parser everywhere (the PowerShell ScriptDom
parser is retired):

- The predicate parser now runs INSIDE SQL Server as a SQLCLR assembly (a C# port
 of the old PowerShell parser, hosting Microsoft's ScriptDom). It is exposed as
 EXEC TestGen.ParseDatabasePredicates and is the single parser used by every entry
 point — SSMS, this module, and CI/CD. No more two-parsers-to-keep-in-sync.
- Install-UnitAutogenDatabase now also registers the parser (bundled
 sql\Install-UnitAutogenClr.SSMS.sql). This needs sysadmin (CONTROL SERVER) ONCE
 and 'clr enabled'=1; it trusts the assemblies by SHA-512 hash via
 sys.sp_add_trusted_assembly (no TRUSTWORTHY required).
- Invoke-UnitAutogen calls EXEC TestGen.ParseDatabasePredicates (was: a PowerShell
 step). -SkipPredicateParse still skips it. No ScriptDom cold start anymore.
- Servers that forbid UNSAFE CLR entirely have no parser; data-shape branches then
 fall back to NOT_TESTABLE / string-gen (tSQLt itself already requires CLR).
- Validated: PredicateZoo parity (CLR == old PowerShell parser on all 28 gates) and
 AssessCustomer 100% line + 100% branch end-to-end with zero PowerShell parsing.

## v0.9.5 (beta) — 2026-06-02

Production-safety release.  Two operational fixes:

1. Recovery from interrupted coverage runs (production-critical):
  When a coverage run is killed mid-flight (test cancelled, agent
  crashed, connection dropped), the procedure being instrumented was
  left in a broken state - the base name became a synonym pointing
  at a stale _cov copy, and any application that called the procedure
  would hit the wrong target.  Two new public procedures:

  - TestGen.CleanupInterruptedRunForProc - single-procedure recovery
  - TestGen.CleanupInterruptedRuns       - database-wide sweep with
                                           optional @SchemaFilter and
                                           @WhatIf preview mode

  GenerateAndCoverDatabase now calls CleanupInterruptedRuns at the
  start as a database-wide sweep, so the framework self-heals
  automatically on the next coverage run.  No user action required
  in the normal path.  Recovery is safe and reversible:  _orig is
  renamed back to the original name via sp_rename; the original
  procedure body is preserved (never DROPped).

2. Compatibility-level pre-flight check:
  Installer now detects databases with compatibility_level < 130 and
  aborts with a clear actionable message (instead of cascading errors
  from STRING_SPLIT).  The required ALTER DATABASE statement is
  included verbatim in the error so the user can copy-paste to fix.

## v0.9.3 (beta) — 2026-06-01

Critical fix (affects ALL prior versions — please upgrade):
- Export functions silently truncated their output at 4000 characters
 in v0.9.0, v0.9.1, and v0.9.2 because Invoke-Sqlcmd's default
 -MaxCharLength was not overridden. Export-CoverageCoberturaXml,
 Export-TestResultsJunitXml, and Export-CoverageHtmlReport all now
 pass -MaxCharLength ([int]::MaxValue), so the full content reaches
 disk. Any user who ran Invoke-UnitAutogen against a non-trivial
 database was affected.

Other improvements:
- Validation on AdventureWorks 2025 now reports 94.9% line coverage
 and 94.4% branch coverage (up from 93.9% line / 50% branch on the
 same database under v0.9.1). Branch detection in scalar functions
 with multi-arm CASE expressions is per-arm, and seeding reaches
 each arm. Three AdventureWorks status-text functions now hit
 100% / 100%.

## v0.9.2 (beta) — 2026-06-01

- TVF shadow teardown: defensively drops the synonym/_cov/_orig objects
 before rebuild, eliminating the "already an object named _covfn" failure
 on rerun when a previous coverage run was interrupted.
- Bundled SQL installer updated with v11 function coverage and seeding
 extensions (30_Function_Support_v1.sql, Patch_v11_SeedExtensions.sql,
 Verify_SeedExtensions.sql, Verify_ShadowTeardown.sql).
- New examples/Demo_Schema.sql for an end-to-end walkthrough.
- PowerShell cmdlets unchanged from v0.9.1.

## v0.9.1 (beta)

Bundled SQL framework updated — user-defined function support + branch coverage:
- Test generation + line/branch coverage for scalar (FN), inline (IF) and
 multi-statement (TF) functions, via a shadow-procedure transform.
- GenerateAndCoverDatabase now reports procedures AND functions in one run.
- Hang-proof coverage probe: every shadow loop is capped, so a runaway loop can
 never stall a coverage run.
- Predicate-inversion branch seeding: value-gated branches (e.g. IF @x = 5) are
 reached on purpose by deriving a satisfying parameter value from the code.
- One-line compound function bodies now instrument correctly.

PowerShell cmdlets are unchanged from v0.9.0.

## v0.9.0 (beta)

New in this release:
- Install-UnitAutogenDatabase: deploy the SQL framework from PowerShell in one command
- Invoke-UnitAutogen: full pipeline — generate tests, measure coverage, export all three output files
- Export-CoverageCoberturaXml: Cobertura XML for Azure DevOps / SonarQube
- Export-TestResultsJunitXml: JUnit XML for all major CI systems
- Export-CoverageHtmlReport: self-contained HTML report

Requirements:
- SQL Server 2017 or later
- tSQLt v1.0.7597.5637 or later installed in the target database