#11180 suites — trunk p2j vs 11180c¶

Side-by-side comparison of the two recorded runs in this directory:

Run	Engine	Report
trunk	/home/sbi/projects/p2j, brz trunk rev ca@goldencode.com-20260807191255 (2026-08-07)	"p2j/RESULTS.md":p2j/RESULTS.md
11180c	/home/sbi/projects/11180c, brz revno 16697, rev-id sbi@goldencode.com-20260731102243 (2026-07-31)	"11180c/RESULTS.md":11180c/RESULTS.md

Both runs used the identical harness — same abl/launcher.p ?prog= driving, same click coordinates, same client reaping, no server restarts, 27 program runs each — so the differences below are engine-attributable, subject to the confound in §6.

Expected results throughout are the native OpenEdge baselines in docs/testsuite/*.md and docs/screen-testsuite.md. FWD behaviour is never the expectation.

---

1. Verdict tallies¶

Suite		trunk	11180c
File-logged (10)	Pass / match	3	6
	Partial	2	2
	Fail	5	2
On-screen (17)	Match	4	11
	Partial / deviates	1	2
	Undetermined	2	0
	Fail	8	2
	OE inconclusive	2	2

2. File-logged suite — per test, with the exact log delta¶

< = trunk, > = 11180c. Diffs taken between p2j/logs/ and 11180c/logs/.

#	Test	trunk	11180c	Log delta
1	tc11180d_false_steal	FAIL	FAIL	logs identical
2	tc11180e_dest_browse_leave	FAIL	PARTIAL	< 08 FOCUS is now: <none/invalid> → > 08 FOCUS is now: FILL-IN/fC
3	tc11180f_btn	PARTIAL	PASS	> 13 fSib LEAVE added
4	tc11180f_btn_browse	PARTIAL	PARTIAL	> 13 fSib LEAVE added
5	tc11180f_btn_browse_ovl	PASS	PASS	> 13 fSib LEAVE added
6	tc11180f_btn_browse_sep	FAIL	PASS	09 frmB ENTRY → 09 fDest LEAVE; settled focus <none/invalid> → FILL-IN/fSib; + 14 fSib LEAVE, 15 frmB LEAVE, 16 frMain ENTRY
7	test-11180-f-redirect-to-target	FAIL	PASS	< 08 FOCUS is now: <none/invalid> → > 08 FOCUS is now: FILL-IN/fT
8	test-11180-g-teardown-apply	PASS	PASS	identical
9	test-11180-g-teardown-hide-apply	PASS	PASS	identical
10	test-11180-g-teardown-mouse	FAIL	FAIL	identical

Four logs are byte-identical (1, 8, 9, 10) — those code paths are untouched by 11180c.

The _sep change at line 09 is a substitution, not an addition: trunk's cosmetic frmB ENTRY is replaced by the fDest LEAVE that the OE baseline requires there.

3. On-screen suite — per probe¶

Probe	trunk	11180c	Movement
A -entry-choose-leave	FAIL	MATCH	Fixed — teardown pair fires twice; originating defect gone
B1 -reentrant-leave-fillin	MATCH	MATCH	unchanged
B2 -reentrant-leave-browse	MATCH	MATCH	unchanged
B3 -reentrant-leave-frameleave	undetermined	MATCH	trace unchanged; verdict resolved by evidence, not by a fix (§5)
B4 -reentrant-leave-delete	undetermined	DEVIATES	behaviour changed — 12 → 14 lines, abandoned-target LEAVE pair added
B5 -reentrant-leave-browse-caller	FAIL	MATCH	Fixed — Effect B correct, focus fiZ
B6 -reentrant-leave-browse-nodelete	MATCH	MATCH	unchanged
B7 test-11180-1	PARTIAL	DEVIATES	focus fixed (<invalid> → fiThird), but LEAVE pair added
C1 -d-false-steal	FAIL	MATCH	Fixed — focus fB
C2 -e-dest-browse-leave	FAIL	MATCH	Fixed — Effect B + focus fC
C3 -e-mouse-dest-browse-leave	FAIL	FAIL	unchanged
C4 -f-redirect-to-target-gui	FAIL	MATCH	Fixed — focus fT; residual frmB ENTRY also gone
C5 -g-teardown-apply-gui	consistent	consistent	unchanged (OE inconclusive)
C6 -g-teardown-hide-apply-gui	MATCH	MATCH	unchanged
C7 -g-teardown-mouse-gui	diverges	diverges	unchanged (OE inconclusive)
C8 -h-redirect-to-sibling-gui	FAIL	MATCH	Fixed — fA LEAVE emitted in OE order; both residuals gone
C9 -i-mouse-frame-leave-gui	FAIL	FAIL	unchanged

4. What the comparison shows¶

Six probes and three file tests went FAIL → MATCH, and every one sits on the APPLY 'ENTRY' path (Site A). The common mechanism: focus now settles on a real widget instead of <invalid>, and the abandoned-target LEAVE is emitted where OE emits it. The originating CounterAct defect (group A) is resolved.

Nothing on the mouse path moved. C3, C9, C7 and file test 10 are unchanged, and three of the four byte-identical logs are here. This matches the source: FocusManager.java is byte-identical between the two trees, so Site B received no change. docs/screen-testsuite.md §1 requires the two engine sites to agree; after 11180c they diverge more than before — Site A now emits Effect B, Site B still emits nothing and lets the clicked destination win.

Two probes moved the wrong way (B4, B7): the abandoned-target LEAVE pair is now emitted where OE emits nothing. Both are new in 11180c. The underlying reason is symmetrical and worth stating plainly — neither engine implements OE's both-Browse gate on that emit:

Probe	Old focus → target	OE	trunk	11180c
B5 -browse-caller	browse → browse	emits	silent ✗	emits ✓
C2 -e-dest-browse-leave	browse → browse	emits	silent ✗	emits ✓
B4 -reentrant-leave-delete	fill-in → browse	silent	silent ✓	emits ✗
B7 test-11180-1	browse → fill-in	silent	silent ✓	emits ✗

Trunk is uniformly silent, so it accidentally satisfies the two cases where OE is silent and fails the two where OE speaks. 11180c is uniformly loud, so it satisfies the opposite pair. Probes 4 and 7 exist precisely to pin this gate down. Focus outcomes are unaffected in both directions; the cost of the 11180c side is spurious LEAVE triggers on widgets that were never entered.

The same-frame sibling-redirect path is not affected on either engine: tc11180f_btn and tc11180f_btn_browse correctly emit no fDest LEAVE, and _sep correctly does.

5. Two verdict changes that are not behaviour changes¶

B3 and B4 were scored "undetermined" on trunk only because the probes' own dynamic overlay hides the trailing -- FOCUS now = … line on the FWD web canvas (see p2j/RESULTS.md §D3). In the 11180c run their event identities were resolved from line ordering — frame-before-widget is the ENTRY pair, widget-before-frame the LEAVE pair, as fixed by B2/B5 where the full text is legible and by the probe source, which registers only ENTRY/LEAVE loggers for those widgets.

• B3's trace is identical across the two runs. Its upgrade to MATCH is better evidence, not a fix.
• B4's trace genuinely changed (12 → 14 lines), so its move to DEVIATES is a real behaviour change.

6. Confound¶

11180c is revno 16697 (2026-07-31); the trunk tested is 2026-08-07. The two trees therefore differ by roughly a week of trunk drift plus the #11180 work, and the 119-line ThinClient.java diff contains both. The attributions in §4 are supported by the history entry (1105, HC 20260723 / SBI 20260731) and by the shape of the code (pendingEntryTarget, the destroyFrame() recovery, the downMouseButtonNewFocus relaxation), but a strictly clean A/B would compare 11180c against its own merge base rather than current trunk.

7. Open items after 11180c¶

1. Site B (FocusManager) has no corresponding change — the mouse path still drops the clicked target and never emits Effect B (3 failing probes, 1 inconclusive).
2. The abandoned-target LEAVE fires without the both-Browse gate — 2 probes over-emit (§4).
3. The head-less twins of review points #1 and #2 still fail where their GUI twins pass — tc11180d vs -d-false-steal, tc11180e vs -e-dest-browse-leave. Same topology, opposite outcome, depending on whether the transfer is driven by a synchronous top-level APPLY chain or from inside the WAIT-FOR loop. The file-logged suite therefore currently understates the fix and should be resolved before it is used as the regression gate.

---

Note: this directory lives under deploy/, which is regenerated by ant deployapp and removed by ant clean.all. An earlier rebuild already deleted p2j/logs/*.log once (restored from the run's scratchpad). Copy this tree elsewhere if it needs to survive a rebuild.