Repository navigation
[API tests] Ensure vendor-payment journal number series - #11228
Prangshuman Das (t-prda) wants to merge 5 commits into
Conversation
|
|
Issue #11561 is not valid. Please make sure you link an issue that exists, is open and is approved. |
Good Sense Reviewer - Round 1Recommendation: AcceptWhat this PR doesThis change gives a test-created vendor-payment journal batch a generated number series when its template leaves that value blank. The guard addresses the failure at its source, preserves an existing configured series, and lets API tests generate document numbers as expected. Problem-solution fitFit: Strong The change directly covers the reported vendor-payment failures, and both re-enabled scenarios exercise the updated shared helper. SuggestionsNone. Risk assessment and necessityRisk: Low. The change is limited to test infrastructure. It affects the vendor-payment tests that use this helper, but only fills a blank number series and leaves existing configuration unchanged. Necessity: The change is needed to run the two previously disabled scenarios in environments where the vendor-payment template has no number series.
|
Good Sense Reviewer - Round 2Recommendation: AcceptWhat this PR doesThe commits since round 1 reconcile the stacked parent only. The layer-owned AL code is unchanged, and the net diff still assigns a generated number series only when the test-created vendor-payment batch lacks one, then re-enables the same two scenarios. Status of previous suggestionsRound 1 had no suggestions. New observations (commits since round 1)None — no new layer-owned hunk appears in both the incremental view and the current net PR diff. Risk assessment and necessityRisk: Low. The change remains test-only, preserves configured number series, and changes shared-helper behavior only when the generated batch has no series. Fresh exact-head runtime validation is pending. Necessity: The change remains necessary to run the two vendor-payment scenarios where the template provides no number series.
|
| begin | ||
| JournalName := LibraryUtility.GenerateRandomCode(GenJournalBatch.FieldNo(Name), DATABASE::"Gen. Journal Batch"); | ||
| LibraryAPIGeneralJournal.EnsureGenJnlBatchExists(GraphMgtJournal.GetDefaultVendorPaymentsTemplateName(), JournalName); | ||
| GenJournalBatch.Get(GraphMgtJournal.GetDefaultVendorPaymentsTemplateName(), JournalName); |
There was a problem hiding this comment.
CreateVendorPaymentsJournal calls EnsureGenJnlBatchExists, which already performs a primary-key Get for this template and batch, and then immediately performs the same Get again before checking No. Series. This adds an avoidable database round trip on every helper call; return or pass through the record loaded by the ensure helper, or otherwise combine the existence check and No. Series update into one lookup.
👍 useful · ❤️ especially valuable · 👎 wrong - reply with why · AL review agent v1.48.6
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session:3952f078-a881-4da8-ad96-13b727e48a91
5b55f51 to
8a45509
Compare
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session:3952f078-a881-4da8-ad96-13b727e48a91
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session:3952f078-a881-4da8-ad96-13b727e48a91
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session:3952f078-a881-4da8-ad96-13b727e48a91
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session:3952f078-a881-4da8-ad96-13b727e48a91
|
Issue #11224 is not valid. Please make sure you link an issue that exists, is open and is approved. |
Good Sense Reviewer - Round 3Recommendation: AcceptWhat this PR doesThe commits since round 2 reconcile the stacked parent only. Both scoped file blobs are unchanged, and the net diff still assigns a generated number series only when the test-created vendor-payment batch lacks one, then re-enables the same two scenarios. Status of previous suggestionsRounds 1 and 2 had no suggestions. New observations (commits since round 2)None — the two scoped files are byte-identical to round 2, and no new layer-owned hunk appears in the current net PR diff. Risk assessment and necessityRisk: Low. The change remains test-only, preserves an existing configured number series, and changes shared-helper behavior only when the generated batch has no series. Fresh exact-head runtime validation is still pending. Necessity: The change remains necessary to run the two vendor-payment scenarios where the template provides no number series.
|
…microsoft#12325) ## Scope Related authentication and API enablement: [AB#646383](https://dynamicssmb2.visualstudio.com/Dynamics%20SMB/_workitems/edit/646383) (link only; this PR does not resolve the umbrella item). Fixes [AB#653119](https://dynamicssmb2.visualstudio.com/1fcb79e7-ab07-432a-a3c6-6cf5a88ba4a5/_workitems/edit/653119) Separate test defect: [653119 - Policy snapshot API URL composition and response isolation](https://dynamicssmb2.visualstudio.com/Dynamics%20SMB/_workitems/edit/653119). This tracks the policy-snapshot test corrections, not production access permissions or unrelated runtime failures. Expense-first adoption directly against main after the external squash merge of workflow microsoft#11891; remaining general APIs follow in microsoft#11860. - Migrate the 11 reviewed Expense API/helper paths to shared authentication and remove exactly 51 existing Expense API exclusions. Preserve exactly nine non-API Expense exclusions; all 19 previously retained API exclusions are now removed. - Enable the existing `enableCleanTestCodeunitExecution` boolean. Use existing DisabledTests selectors throughout discovery, ordinary execution and clean-codeunit execution/reruns. No app-name allowlist, new selection setting or runner implementation/test change. - Include the six-line license-safe WorkDate helper needed by PerDiem, five query-safe policy URL compositions, four independent response clears, and the existing unlimited-approval fixture. Preserve all nine Activity Log tests and upstream assertions. - Consolidate microsoft#11453's exact assigned-user filter (no cross-table range compression), true fallback fixture and regressions. Upstream wildcard quoting alone did not fix the range-gap case. - Consolidate microsoft#11451's unique denied-approval error/permission capture and microsoft#11454's existing repeated-delta, missing-header-permission, Released-status and real deletion-total regressions, including its internal test-only permission set. - Do not reintroduce upstream repairs from microsoft#11654 (three permission-role tests and posted zero-amount fixture), the obsolete date test removed by microsoft#12074, or the production indirect-Modify change already supplied by microsoft#11333. microsoft#11452's duplicate deletion helper is absent; existing callers use the upstream shared helper. - All seven artificial Expense exclusions introduced by the previous uptake are absent here and at every general checkpoint. This is not blanket re-enablement of Expense tests. This focused review diff has21 paths, using the two existing API exclusion manifests. No NAV selectors, local NST, BC-ExpenseAgent changes, unrelated new test scope or production permission changes. Exact head: `bd3bbc04e13965f967a26eef435bdbf0cbc09ad7`; tree: `2b8c47b138846f63bd49d47188f54ea7e86c1f24`. ## Expense-first sequencing using existing exclusions The same **193 method-specific temporary exclusions (135 APIV1 +58 APIV2 across12 codeunits)** now live in the existing `src/DisabledTests/_Exclude_APIV1__Tests/_Exclude_APIV1__Tests.DisabledTest.json` and `src/DisabledTests/_Exclude_APIV2__Tests/_Exclude_APIV2__Tests.DisabledTest.json`. No separate sequencing files remain. Original entries retain their order/style/content; no new duplicate or wildcard/codeunit-wide exclusion is introduced. Unrelated pre-existing APIV2 duplicates are preserved rather than mixed with this change. microsoft#11860 removes the temporary additions from these same existing manifests alongside its already-reviewed authentication uptake. All193 temporary stage entries are removed in microsoft#11860, but it makes **192/193 target methods eligible**: the independent `139739::TestDeleteInUse` exclusion remains until the VAT fixture correction in PR [microsoft#11224](microsoft#11224) removes it. From microsoft#11224 onward all193 are eligible; eligibility is not runtime success. All general/downstream cumulative trees are **exactly byte-identical** to the preceding checkpoint; the full checkpoint has none of these193 methods excluded. No app-name allowlist, selection setting, runner/authentication/test/production change or Logiq exclusion is added. **Why these methods started executing:** they already required Disabled isolation and were IntegrationTest codeunits. Ordinary typed selection in TestSuiteMgt332–352 selects None|Codeunit; the old extra Disabled pass in RunTestsInBcContainer was UnitTest-only. The clean-execution switch newly reaches Disabled IntegrationTest codeunits and still honors the existing JSON exclusions. These193 methods were not listed in those exclusions. This is a pre-existing selection gap, not a newly introduced product auth bug or proof they never ran in any historical configuration. A complete67-codeunit audit preserves existing UnitTest/Legacy behavior and the ten independently handled Logiq integration tests. Historical baseline run37215976231 at head69df41756d918a516a3c868ca66f45cbfd320428 independently corroborates this: zero of the exact193 methods appear across five inspected W1 result artifacts containing37,626 testcases (Integration, both Legacy buckets, default/unit and Uncategorized). This evidence is limited to that W1 baseline, not every country or historical run. The completed alternate-path audit found no other ordinary configured baseline BCApps lane: APIV1/APIV2 are outside Legacy buckets, Disabled-unit fallback retains UnitTest filtering, discovery skips test procedures, and ordinary PR/CI/CD/rerun routes do not bypass those constraints. Manual or explicitly untyped execution remains possible; no universal historical or NAV absence is claimed. **Expense scope is unchanged:**51 API reenables, nine non-API exclusions, baseline six cases and all consolidated regressions. The two legacy Spend Requests methods remain conditional on not CLEAN30. Focused microsoft#12325 review:21 files, including the two existing exclusion manifests. General microsoft#11860 review:159 files. The official NAV `Disable-NAVALTest` helper was inspected: it has no destination/app-file parameter, writes NAV's App/DisabledTests using per-codeunit filenames and sorts entries. It cannot safely preserve these BCApps files. A bounded JSON relocation preserved original prefixes/order and checked exact identities, then exercised the unchanged real loader. No new shared helper/framework or NAV selector change was made. ## Validation and presentation evidence **Pester117/117 passed independently at the new Expense/general/full heads.** Actual existing-loader checks confirm identical effective193-method selection after relocation and preserved51/9 Expense scope. All nine downstream Git tree hashes remain exactly unchanged. Fresh exact-head CI is pending; old-head successes are not substituted for current runtime evidence. Historical run37330893173 at head88881be reached193 general methods:171 genuine401 failures and22 nominal passes (17 bare-ASSERTERROR cases can accept the wrong error; five local fixtures). Separately, all51 Expense methods passed in13 inspected countries (663 results), and all63 touched API methods yielded819 results; Activity coverage was117/198 at that checkpoint. These are bounded historical results, not a full/current matrix pass. The preceding stage-fix runs hit50 hosted-runner acquisition cancellations; other jobs were active, so this was not a claimed global outage. New pushes schedule fresh CI, not an AL retry. The general SQL-pool NRE and missing warning-reference artifacts remain separate unresolved limitations. Durable session presentation notes: **api-test-enablement-presentation-notes.md**, with source-pinned selection proof, fix/coverage inventory, helper limitations and historical-vs-current evidence boundaries. Fresh runtime proof must still establish the51 Expense cases and actual193-case suppression at Expense stage. [AB#646383](https://dynamicssmb2.visualstudio.com/1fcb79e7-ab07-432a-a3c6-6cf5a88ba4a5/_workitems/edit/646383) (link only). Native stack #12327 remains microsoft#12325 → microsoft#11860 → microsoft#11224 → microsoft#11225 → microsoft#11226 → microsoft#11227 → microsoft#11228 → microsoft#11229 → microsoft#11230 → microsoft#11322. Protected merged microsoft#10085/microsoft#11862/microsoft#11891 and validation microsoft#11892 are untouched; validation-only microsoft#11861/microsoft#11455 remain Do Not Merge. No new main integration, native-group/base change, PR merge, queue operation, Actions cancellation or manual retry was made. All prior heads remain ancestors; source publication used backups and an atomic forward-only push with explicit leases. Exact current head: `bd3bbc04e13965f967a26eef435bdbf0cbc09ad7`; source tree: `2b8c47b138846f63bd49d47188f54ea7e86c1f24`. --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 3952f078-a881-4da8-ad96-13b727e48a91
Fixes AB#653125
Scope
Assign a generated number series only when the test-created vendor-payment journal batch lacks one. Keep production numbering behavior unchanged.
Related umbrella646383.
Evidence and limits
Two observed CZ failures in saved result set 2: CU139843.TestCreateVendorPaymentWithoutDocNo and CU139867.TestApplyVendorEntryAppliesPayment. There are 18 direct test callers of the shared helper, but only these two failures are evidenced by those saved results.
Only this layer's original fix/exclusion patch is reviewed here (2 paths); the patch is unchanged by Expense-first restructuring.
Exact head:
202c33bd4f85fa13eab6c1b73a35518347940293; tree:74300ae9d5678a4d432452e90cfc7e6a597195f6.Expense-first sequencing using existing exclusions
The same 193 method-specific temporary exclusions (135 APIV1 +58 APIV2 across12 codeunits) now live in the existing
src/DisabledTests/_Exclude_APIV1__Tests/_Exclude_APIV1__Tests.DisabledTest.jsonandsrc/DisabledTests/_Exclude_APIV2__Tests/_Exclude_APIV2__Tests.DisabledTest.json. No separate sequencing files remain. Original entries retain their order/style/content; no new duplicate or wildcard/codeunit-wide exclusion is introduced. Unrelated pre-existing APIV2 duplicates are preserved rather than mixed with this change.#11860 removes the temporary additions from these same existing manifests alongside its already-reviewed authentication uptake. All193 temporary stage entries are removed in #11860, but it makes 192/193 target methods eligible: the independent
139739::TestDeleteInUseexclusion remains until the VAT fixture fix PR #11224 removes it. From #11224 onward all193 are eligible; eligibility is not runtime success. All general/downstream cumulative trees are exactly byte-identical to the preceding checkpoint; the full checkpoint has none of these193 methods excluded. No app-name allowlist, selection setting, runner/authentication/test/production change or Logiq exclusion is added.Why these methods started executing: they already required Disabled isolation and were IntegrationTest codeunits. Ordinary typed selection in TestSuiteMgt332–352 selects None|Codeunit; the old extra Disabled pass in RunTestsInBcContainer was UnitTest-only. The clean-execution switch newly reaches Disabled IntegrationTest codeunits and still honors the existing JSON exclusions. These193 methods were not listed in those exclusions. This is a pre-existing selection gap, not a newly introduced product auth bug or proof they never ran in any historical configuration. A complete67-codeunit audit preserves existing UnitTest/Legacy behavior and the ten independently handled Logiq integration tests.
**Expense scope is unchanged:**51 API reenables, nine non-API exclusions, baseline six cases and all consolidated regressions. The two legacy Spend Requests methods remain conditional on not CLEAN30. Focused #12325 review:21 files, including the two existing exclusion manifests. General #11860 review:159 files.
The official NAV
Disable-NAVALTesthelper was inspected: it has no destination/app-file parameter, writes NAV's App/DisabledTests using per-codeunit filenames and sorts entries. It cannot safely preserve these BCApps files. A bounded JSON relocation preserved original prefixes/order and checked exact identities, then exercised the unchanged real loader. No new shared helper/framework or NAV selector change was made.Validation and presentation evidence
Pester117/117 passed independently at the new Expense/general/full heads. Actual existing-loader checks confirm identical effective193-method selection after relocation and preserved51/9 Expense scope. All nine downstream Git tree hashes remain exactly unchanged. Fresh exact-head CI is pending; old-head successes are not substituted for current runtime evidence.
Historical run37330893173 at head88881be reached193 general methods:171 genuine401 failures and22 nominal passes (17 bare-ASSERTERROR cases can accept the wrong error; five local fixtures). Separately, all51 Expense methods passed in13 inspected countries (663 results), and all63 touched API methods yielded819 results; Activity coverage was117/198 at that checkpoint. These are bounded historical results, not a full/current matrix pass. The preceding stage-fix runs hit50 hosted-runner acquisition cancellations; other jobs were active, so this was not a claimed global outage. New pushes schedule fresh CI, not an AL retry. The general SQL-pool NRE and missing warning-reference artifacts remain separate unresolved limitations.
Durable session presentation notes: api-test-enablement-presentation-notes.md, with source-pinned selection proof, fix/coverage inventory, helper limitations and historical-vs-current evidence boundaries. Fresh runtime proof must still establish the51 Expense cases and actual193-case suppression at Expense stage.
Related umbrella646383. Native stack #12327 remains #12325 → #11860 → #11224 → #11225 → #11226 → #11227 → #11228 → #11229 → #11230 → #11322. Protected merged #10085/#11862/#11891 and validation #11892 are untouched; validation-only #11861/#11455 remain Do Not Merge. No new main integration, native-group/base change, PR merge, queue operation, Actions cancellation or manual retry was made. All prior heads remain ancestors; source publication used backups and an atomic forward-only push with explicit leases.
Exact current head:
202c33bd4f85fa13eab6c1b73a35518347940293; source tree:74300ae9d5678a4d432452e90cfc7e6a597195f6.