Skip to content
This repository was archived by the owner on Oct 21, 2026. It is now read-only.

Bump the corpus to c874a95 (corpus bump 41) - #399

Merged
StefanMaron merged 10 commits into
mainfrom
corpus-bump-41
Oct 21, 2026
Merged

StefanMaron merged 10 commits into
mainfrom
corpus-bump-41

Conversation

@StefanMaron

Copy link
Copy Markdown
Owner

What changed

  • Corpus pinned to c874a956 (corpus master, merge of corpus PR 698 = corpus bump 41: writers QJ and QK, plus the workflow uses: fix for the move to aldevtools). The oracle files are copied unchanged from the green run of PR 698 (39550736920). That run's head d0d77cd0 has the same tree as c874a956. The master push run of c874a956 (39556438011) was still running when I set the pin.
  • Done with tools/corpus_bump.sh 39550736920. Then I moved the pin from d0d77cd0 to c874a956 and reran --update-baseline for BC 27, 28, 29 with the shared HOME. Harness counts, both coverage maps and the pinned-columns header are regenerated. The oracle README names the new corpus repository (aldevtools).
  • Note on the task brief: it named run 39505560294. That is the bump 36 run (head 55f2db65, tree 62268a97), not bump 41. I used the bump 41 run instead.
  • Bump 41 adds 8 cases per major, the same on every major: QjPartPropagation.al (8 QJ_ cases). QkSetupRows.al is a fixture with no tests. QK changed existing tests in 5 codeunits (78603, 70535, 67755, 84363, 86504) to write their own setup rows. No must-not-compile case and no version guard changed.
  • Runtime changes for what the bump pins:
    • QJ, part update propagation. The assumptions part-update-propagation-both-refreshes-host and part-saverecord-in-validate-keeps-xrec are removed. The code cites pages/lt-qj-part-propagation/QjPartPropagation.al: host re-read in every UpdatePropagation mode (:508, :545, :582), unsaved host edit saved first (:623, :661, :699), SaveRecord-only card action (:511), xRec after SaveRecord in a subform OnValidate (:733, :735, :763, :764, :766). One branch is still unpinned: a NEW part row with UpdatePropagation = Both (the QJ tests use a stored row). It keeps its behavior under the narrower assumption part-update-new-row-both-refreshes-host.
    • QK, CaptionClass areas 1 and 101. Without test data these were refused (no demo General Ledger Setup row). They are now answered when the running test wrote General Ledger Setup before any caption of those areas was resolved in the session. Base Application's codeunits 342 and 343 are SingleInstance and keep the row they read first. Only what QK writes is answered: area 101 type 0 (LCY Code) and area 1 1,1 (Global Dimension 1 Code). Other types stay refused (QdCaptionClassAreas.al:267, :269, :311).
    • General Ledger Setup is a demo table now. A Count or IsEmpty of it before the test wrote to it is refused. Without this, QF_GLSetup_CorpusRow (QdCaptionClassAreas.al:804) would go from not run to fail.
    • The demo-table write check is per test now, not per session. The QK tests put the demo values back after they run. On BC that restores the demo row; the runtime is left with the install-code row. A write by an earlier test therefore no longer counts.
  • The Count/IsEmpty demo-row refusals from PR 393 do not refuse any QK test.

Numbers (pass / fail / not run)

Before: main at 35df818 with the old pin 0075e2f; baseline lines 13891 / 13929 / 13957 (BC 27 / 28 / 29).

  • With only the new pin (same code as main): BC 27 13890/17/3115 of 17022, BC 28 13928/17/3166 of 17111, BC 29 13949/17/3239 of 17205.
  • With the runtime changes, after merging main at 61bcbad (PR 398): BC 27 13892/17/3113, BC 28 13930/17/3164, BC 29 13951/17/3237. Flagged passes 506/509/508 -> 430/433/432.
  • New baseline cases per major (3): 84363 ClearAll_InBaseAppPage_ExtensionGlobal (QK writes its setup rows), and 86504 Literals_FieldCaption and Expressions_FieldCaption on every major. Baseline lines after: 13894 / 13932 / 13953.
  • 7 BC 29 baseline cases left the baseline in one dedicated commit, because QK changed their oracle on purpose (the commit names them): 67755 PW_DumpPlain_Lines01To18, PW_DumpPlain_Lines19To36; 70535 PW_Dump_Lines001To016, 017To032, 081To096, 097To112, 113To128. They now write a Company Information row with a Picture and address fields. The runtime refuses them: the dataset header's CompanyLogo encoding and the company address format are not built.
  • No other baseline case stopped passing. No case that ran before fails now. OnPrem and DotNet: no change.
  • probes-needed.md table rows: 634 -> 633 (2 rows deleted, 1 narrower row added).

Cases that fail, by cause (17 per major)

  • New, QJ open sequence (8 per major, 86889 QjPartPropagation.al). All 8 fail at the first observation: the trigger log of the page open, before the propagation part the runtime implements. In a part's OnAfterGetRecord on open, BC gives xRec blank (x.t=0), the runtime the same row. The runtime also refreshes the part once more after the host's second read, and the subform runs OnAfterGetCurrRecord twice. This is the same open-path xRec gap as QiHostSaveParts.al. It belongs to the lane working on pages and host SaveRecord with parts, so I left it.
  • Not new: 86836 QiHostSaveParts.al (8 per major) and 86826 QI_TableFieldExpression_FormatInCode (1 per major), as in PR 393.

Cases that are refused, by cause (QK codeunits, BC 29)

  • 14: CompanyLogo of the dataset header (Company Information has a Picture): 67755 (5), 70535 (8), 78603 (1). BC 29 only. The oracle pins the base64 of QK's 1x1 PNG.
  • 2: company address lines of the dataset header (78603, "Address 2" and "City" set).
  • 1: 86504 QF_GLSetup_CorpusRow, Count of General Ledger Setup before the test wrote (the corpus database's demo row).

Needs a decision (next lane, not built here)

  • CompanyLogo and the company address format of the BC 29 dataset header. They win back the 7 cases that left the baseline, plus 78603. The logo is pinned only for one 1x1 PNG, so whether BC passes every picture's bytes through unchanged is not pinned.
  • The QJ open sequence (xRec in a part's OnAfterGetRecord on open, the extra part refresh): for the pages and parts lane. Fixing it should turn the 8 QJ cases green.
  • New part row with Both (part-update-new-row-both-refreshes-host): needs a probe.

Assumptions

  • Removed: part-update-propagation-both-refreshes-host, part-saverecord-in-validate-keeps-xrec (bump 41 pins them).
  • Added: part-update-new-row-both-refreshes-host (the unpinned new-row branch of the removed one, same behavior as before).

Gates

  • ./check.sh green after merging origin/main. corpus_resolve_deps --ignored: 4 of 4 pass.

https://claude.ai/code/session_01MifNJNPViEXwK4p83TH2DQ

Corpus master c874a956 is the merge of corpus PR 698 (bump 41, writers QJ and QK, plus the
workflow fix for the move to aldevtools). The oracle files are copied unchanged from the green
PR run 39550736920, whose head d0d77cd0 has the same tree as c874a956. Done with
tools/corpus_bump.sh, then the pin moved from d0d77cd0 to c874a956. The oracle README names
the new corpus repository.

84363 ClearAll_InBaseAppPage_ExtensionGlobal now passes on BC 27 and 28 (QK writes its setup
rows). The BC 29 baseline follows in the next commits.
Writer QK changed these tests on purpose: they now write their own Company Information row
(misc/lt-qk-setup-rows/QkSetupRows.al), with a Picture and address fields, instead of reading
the corpus database's demo row. The old passes were of the old tests. The runtime now refuses
them: the dataset header's CompanyLogo encoding and the company address format are not built.

  67755 PW_DumpPlain_Lines01To18
  67755 PW_DumpPlain_Lines19To36
  70535 PW_Dump_Lines001To016
  70535 PW_Dump_Lines017To032
  70535 PW_Dump_Lines081To096
  70535 PW_Dump_Lines097To112
  70535 PW_Dump_Lines113To128
84363 ClearAll_InBaseAppPage_ExtensionGlobal now passes on BC 29 too (QK writes its setup rows).
… c874a95

No platform column became pinned; only the commit in the header changes.
pages/lt-qj-part-propagation/QjPartPropagation.al is on corpus master now. The two assumptions that
flagged its behavior until the bump landed are removed, and the code cites the file's lines:
- part-update-propagation-both-refreshes-host: a part's Update(false) after its code changed the
  host row makes the host read its row again in every UpdatePropagation mode (:508, :545, :582);
  an unsaved host edit is saved first (:623, :661, :699); a card action that only called
  SaveRecord reads OnAfterGetRecord alone, then the parts (:511).
- part-saverecord-in-validate-keeps-xrec: SaveRecord in a subform OnValidate keeps xRec; for a new
  row xRec stays the blank new row (:733, :735, :763, :764, :766).

One branch stays unpinned: a NEW part row with UpdatePropagation = Both (the QJ tests use a stored
row). It keeps its behavior under the narrower assumption part-update-new-row-both-refreshes-host.
…Ledger Setup (writer QK)

Corpus bump 41 (misc/lt-qk-setup-rows/QkSetupRows.al) makes Literals_FieldCaption and
Expressions_FieldCaption of QdCaptionClassAreas.al write their own General Ledger Setup and
Dimension rows. Without test data, a caption of areas 1 and 101 was refused because the runtime has
no demo setup row. It is now answered when the running test wrote General Ledger Setup before any
caption of those areas was resolved in the session (Base Application's resolvers are SingleInstance
and keep the row they read first), and only for what QK writes: area 101 type 0 (LCY Code) and
area 1 '1,1' (Global Dimension 1 Code). Other types stay refused.

General Ledger Setup joins the demo tables: a Count or IsEmpty of it before the test wrote to it is
refused (QdCaptionClassAreas.al:804 QF_GLSetup_CorpusRow, which would otherwise fail). What an
earlier test wrote no longer counts for the next test: the QK tests put the demo values back, which
leaves the runtime with the install-code row.

Two new baseline cases per major: 86504 Literals_FieldCaption and Expressions_FieldCaption.
@StefanMaron
StefanMaron enabled auto-merge October 21, 2026 07:48
@StefanMaron
StefanMaron merged commit 6536b24 into main Oct 21, 2026
7 checks passed
@StefanMaron
StefanMaron deleted the corpus-bump-41 branch October 21, 2026 08:03
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant