Bugfix Issue #1904 — addRelation() tcversion-id input mode is dead

testcase::addRelation() documents that $source_id/$destination_id may be "ID of source test case version or test case", but calling it with tcversion ids never inserted a relation: the destination open-guard resolved the version id as a test-case id, found nothing, emitted 2 E_WARNINGs (+ 1 PHP 8.3 false-to-array deprecation) and returned related_tcase_not_open with zero rows written.

Symptom

Calling addRelation($tcversionSrc, $tcversionDst, $type, $authorId) with the two tcversion ids of two active test cases:

• returned array('status_ok' => false, 'msg' => 'related_tcase_not_open');
• inserted no row in testcase_relations;
• logged three PHP diagnostics:

  E_WARNING  Trying to access array offset on false ... testcase.class.php:2977
  Deprecated Automatic conversion of false to array is deprecated ...:2992  (PHP 8.3)
  E_WARNING  Undefined array key "is_open"            ... testcase.class.php:8235
  

The version-id input mode advertised by the docblock was effectively dead.

Reproducing it by hand

Fixture: php tmp/fixtures_1904.php (project Rel 1904, two ACTIVE test cases, prints tcaseA/tcversionA/tcaseB/tcversionB). Repro: php tmp/repro_1904.php.

Before the fix (version-id pair):

=== CASE 1: tcversion ids (the dead documented mode) ===
PHP Warning:  Trying to access array offset on false in .../testcase.class.php on line 2977
PHP Deprecated:  Automatic conversion of false to array is deprecated in ...:2992
PHP Warning:  Undefined array key "is_open" in .../testcase.class.php on line 8235
return={"status_ok":false,"msg":"related_tcase_not_open"}
rows_after_case1=0

The test-case-id control call worked: relation_added, 0 warnings, 1 row.

Root cause

addRelation() (lib/functions/testcase.class.php:8189) accepts test-case or tcversion ids but, at the time of the report, resolved only test-case ids:

if( $nu['node_type_id'] == $this->node_types_descr_id['testcase'] ) {
    // resolve BOTH endpoints to latest active tcversion ids
    $safeID['s'] = intval($tcvSet[$safeID['s']]['tcversion_id']);
    $safeID['d'] = intval($tcvSet[$safeID['d']]['tcversion_id']);
}

When the source is a tcversion id this block is skipped (by design — "Will check only source") so the ids stay as versions, which the INSERT then stores correctly. The failure is the open-guard:

$dummy = $this->get_by_id($destination_id, self::LATEST_VERSION);   // :8234 (old)
if(($dummy[0]['is_open']) == 1) { ... }

get_by_id() treats its first argument as a test-case id (WHERE NHTCV.parent_id = {$id}), so a version id matches no test case and no row is returned. Under self::LATEST_VERSION with a scalar id the failure becomes array(false) (get_by_id wraps fetchFirstRow()'s false at :2918-2919), so the render loops inside get_by_id produce the :2977/:2992 diagnostics and the guard produces the :8235 one.

Why it breaks NOW

Not a regression — the guard has always been test-case-id-only; the docblock (:8176-8184) advertises both modes. File-follows: while verifying the #1903 fix (issue #1903), identical output with and without that fix — confirmed unrelated.

Blast radius

Production callers of addRelation() (both pass test-case ids, both unaffected):

caller	file:line	path
legacy controller	lib/testcases/testcaseCommands.class.php:1293	tcEdit.php ?doAddRelation
modern BFF	api/testcases/index.php:3219	POST ?action=add_relation

The same get_by_id($id, LATEST_VERSION) shape appears in getRelations() (:7904, :7939) but its inputs are test-case ids — out of scope.

The fix

Branch the open-guard on the destination node type (lib/functions/ testcase.class.php:8207 ff.):

• test-case id → the legacy get_by_id($destination_id, LATEST_VERSION) check, unchanged (now null-safe with is_array() guards so an absent test case is rejected without warnings);
• tcversion id → SELECT is_open FROM tcversions WHERE id = $safeID['d'] — the exact version row the INSERT stores (this revives the documented mode);
• anything else / missing (id not present in nodes_hierarchy) → clean related_tcase_not_open with no get_by_id() call, hence no warnings.

The destination node type is read once before the resolution block:

$destNodeType = null;
$sqlDest = " SELECT node_type_id ... WHERE id = {$safeID['d']}";
$nuDest = current((array)$this->db->get_recordset($sqlDest));
if( !empty($nuDest) ) { $destNodeType = intval($nuDest['node_type_id']); }

The source-guard is also hardened with !empty($nu) so a missing source id no longer warns either.

Alternatives rejected: fixing get_by_id() alone would have silenced the two internal diagnostics but left the semantic bug (version id fed as a test-case id); per-id full resolution (test-case→latest-active-version on each side) would be more complete for mixed pairs but changes behaviour for the production test-case-id callers — out of scope of a minor latent fix.

i18n

No user-facing strings changed. relation_added, related_tcase_not_open, relation_already_exists are pre-existing feedback codes already localized in the BFF screen and legacy locale/*/strings.txt.

Verification

#	Check	Result
1	php tmp/verify_1904.php matrix (version/version, tcase/tcase, duplicate, is_open=0, missing dest)	PASS — 5/5, every case warn=0
2	BFF live (POST /api/testcases/index.php?action=add_relation, admin session): call #1	PASS — HTTP 200 relation_added, 1 row
3	BFF identical call #2	PASS — HTTP 422 relation_already_exists, still 1 row
4	Browser tcView.html?tcase_id=<A>	PASS — "RELATIONS WITH OTHER TEST CASES (1)" → "is related to REL1904-2"
5	events after the pass	PASS — 0 new Error/Warning rows
6	browser console	PASS — no JS errors (only the expected 422 resource)
7	php -l lib/functions/testcase.class.php	PASS — clean
8	bash ai/verify_test_suites.sh (TLU_REQUIRE_SUITE="Issue #1904")	PASS — 7 PASS / 0 FAIL

Regression suite: ## Regression — Issue #1904 in tmp/TLU_Test_Cases.md (10/10 rows for this fix).

Files changed

file	change
lib/functions/testcase.class.php	destination node-type lookup + open-guard branching (+37/−5)
tmp/fixtures_1904.php, tmp/repro_1904.php, tmp/verify_1904.php	fixture / repro / regression matrix
tmp/TLU_Test_Cases.md	## Regression — Issue #1904 suite
CHANGELOG	[KEY BUGFIX] line for #1904
docs/screenshots/issue-1904-addRelation-tcversion-mode.png	tcView showing the added relation