fix(engine): partial SEEK keeps Found() true when deleted rows are hidden

[russimicro]

Table::seek_key re-derives the exact flag whenever deleted records are hidden, and it did so by padding the search key out to the full index key width with spaces:

del_key.resize(idx->key_length(), ' ');
ck.resize(del_key.size(), ' ');
exact = (ck == del_key);

A partial (prefix) key can never satisfy that — the pad bytes get compared against the stored suffix. For a tag built as CON+DOC+STR(SEQ,6,0), a 10-byte CON+DOC seek compared "2P 972299" + " " against "2P 972299" + " 1", so the cursor landed on the right record but AdsIsFound reported 0. It only happens with SET DELETED ON, which is what made it look arbitrary. The same padding delimited the equal-key run in the walk_to_last (AdsSeekLast) block.

Clipper, Harbour DBFCDX and SAP ACE all treat a shorter seek key as a prefix match, and legacy xBase code relies on it: SEEK con+doc followed by IF !FOUND() over a detail table reported every document as missing on the ADS path while DBFCDX found them.

CdxIndex::seek_key itself was already correct (covered by cdx_prefix_seek_test); the defect was strictly the Table-layer re-derivation. The fix compares only the bytes the caller supplied; a key longer than the index key keeps the historical truncation. It applies to both ADS_CDX and ADS_ADT, since it sits in the shared engine layer.

Test

abi_prefix_seek_deleted_test — prefix / full / absent key under both AdsShowDeleted states, plus a group whose first line is deleted (the prefix must still find the surviving line) and a fully deleted group (must miss).

Verified that the test actually catches the defect: built against the tree with the fix reverted, it fails on exactly the two prefix-with-DELETED-ON assertions.

build	unit suite	this test
without the fix	1241 / 1248	FAIL — 56/58 assertions
with the fix	1242 / 1248	PASS — 58/58 assertions

The remaining failures are the pre-existing Remote: / networking cases (abi_pritpal_lock_test, embedded TCP server), identical with and without this change in our environment.

Real-data validation

Harbour 3.2 + rddads + this DLL, against a production table (1,674 records, 6 tags, 16-byte tag 1). 18/18 across CDX and ADT/ADI, both DELETED states, including the non-regression case (absent prefix must miss):

format	RDD	DELETED	partial (10)	full (16)	absent prefix
CDX	ADS	OFF / ON	SI / SI	SI / SI	NO / NO
CDX	DBFCDX	OFF / ON	SI / SI	SI / SI	NO / NO
ADT/ADI	ADS	OFF / ON	SI / SI	SI / SI	NO / NO

ADS now matches DBFCDX in every cell. Before the fix, ADS / DELETED ON / partial was the only failing one.

Found while running a Harbour/FiveWin ERP on OpenADS: an ADS/CDX company could not read any movement document. Validated end-to-end in the ERP with no application change.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[FiveTechSoft]

Integrated into main as part of the PR stack merge (9e7ac41). Cherry-picked onto post-CI-fix main, conflict-resolved (CHANGELOG / CMakeLists), and verified locally. Thank you @russimicro.