Every Dashboard load for a non-existent / stale test plan id wrote two E_WARNING rows to the events table:

E_WARNING
Trying to access array offset on null - in .../lib/functions/testplan.class.php - Line 5877
source: PHP

HTTP stays 200 — the symptom only shows up in the Event Viewer / events table (log_level 2, activity=PHP).

Reproduce at HEAD (measured on the fresh-import CI box, zero nodes_hierarchy / testplans / testprojects rows): with an authenticated session, one browser load of index.php?tproject_id=1&tplan_id=2 — the Dashboard main frame (gui/templates/mainpage/mainPage.html) fires GET /api/mainpage/ and GET /api/mainpage/bugsTested, and each request appends one E_WARNING row:

• curl -b <cookies> "http://localhost:8082/api/mainpage/index.php?tproject_id=1&tplan_id=2" → HTTP 200, 1 new row
• curl -b <cookies> "http://localhost:8082/api/mainpage/index.php/bugsTested?tproject_id=1&tplan_id=2" → HTTP 200, 1 new row

• lib/functions/tree.class.php:220 get_node_hierarchy_info() returns $result = !is_null($rs) ? $rs[0] : null; — a node id with zero rows yields null, not an empty array.
• lib/functions/testplan.class.php:5875-5877 helperConcatTCasePrefix($id) does $io = $this->tree_manager->get_node_hierarchy_info($id); then list($prefix,$garbage) = $this->tcase_mgr->getPrefix(null,$io['parent_id']); — dereferencing ['parent_id'] on null throws the PHP 8 warning.
• lib/functions/logger.class.php:1435-1460 watchPHPErrors() persists every E_WARNING into events (log_level=2).

Trigger path: mainPage.html → getBugsTestedData() (api/mainpage/index.php:316) → getAllExecutionsWithBugs() (testplan.class.php:7575) → getLinkedTCVersionsSQL() (testplan.class.php:7066, callsite 7184) → helperConcatTCasePrefix().

Blast radius: single shared choke point — all 8 helperConcatTCasePrefix callsites (tlTestPlanMetrics.class.php:2293,2414,2557,3065,3220,3286; testplan.class.php:6003,7184) are covered by the one guard. This is the testplan-side twin of the fixed #1557 (testproject::isIssueTrackerEnabled()).

Committed on branch fix/issue-1558-helperconcat-null-guard, commit 0f6d0550c:

// Get test case prefix
$debugMsg = 'Class:' . __CLASS__ . ' - Method: ' . __FUNCTION__;
$io = $this->tree_manager->get_node_hierarchy_info($id);

// A node id that does not exist (stale/deep-link request) makes
// get_node_hierarchy_info() return null; without a reset guard the
// $io['parent_id'] dereference raises a PHP 8 E_WARNING that lands in
// the events table (api/mainpage dashboard passes stale tplan ids).
// Bailing out with a null prefix yields the same empty-prefix concat the
// code emits for a project with no prefix, and the SQL stays valid.
// Refs #1558
$prefix = null;
if( !is_null($io) && isset($io['parent_id']) )
{
  list($prefix,$garbage) = $this->tcase_mgr->getPrefix(null,$io['parent_id']);
}
$prefix .= $this->tcaseCfg->glue_character;
$concat = $this->db->db->concat("'{$prefix}'",'TCV.tc_external_id');

Guard semantics: $prefix is initialised to null; getPrefix() is only called when the node exists. On the skip path $prefix .= $glue_character produces the same glue-only prefix string the code already emits for a project with no prefix configured, and ADOdb Concat() always returns a string (implode($concat_operator, ...)), so the caller's SQL ($fullEID AS full_external_id) stays valid in all cases. Valid ids resolve their prefix exactly as before (getPrefix(null,$io['parent_id']) unchanged).

Rejected alternatives: (a) returning a bare null / empty-string $concat — would break the SQL-string callers on the stale-id path; (b) throwing a user-visible "unknown plan" dialog — out of scope, the stale-id request is log-noise only; (c) changing get_node_hierarchy_info() / get_recordset() to return [] — a database-layer contract change with a far wider blast radius. The fix is minimal and mirrors the already-shipped #1011 getPrefix() guard and the #1557 isIssueTrackerEnabled() guard.

• lib/functions/testplan.class.php — helperConcatTCasePrefix(), lines 5871-5887: null-safe guard around the getPrefix call.
• tmp/TLU_Test_Cases.md — regression suite "Regression — Issue #1558", 4/4 PASS.

No user-facing strings touched → no i18n bundle changes (i18n not applicable).

All checks on branch fix/issue-1558-helperconcat-null-guard, PHP 8.3, MySQL testlink fresh-import schema.

• R1/R2 — stale id 2 (zero-row DB), DELETE FROM events first: GET /api/mainpage/index.php?tproject_id=1&tplan_id=2 and GET /api/mainpage/index.php/bugsTested?... → HTTP 200 each; events count 0 (pre-fix: 1 E_WARNING per call — liverepro events ids 3..8).
• R3 — full browser page load index.php?tproject_id=1&tplan_id=2 (navbar + aside + Dashboard main frame: GET /, /bugsTested, /assigned, /notifications/count) → HTTP 200 frameset, Dashboard renders its empty state, events 0 (pre-fix: 2 E_WARNING rows per load).
• R4 — valid-plan smoke: seeded nodes_hierarchy (project id 1, plan id 2) + testprojects prefix PROJ + testplans; CLI harness with error_reporting(E_ALL) and endpoint curls → valid id yields CONCAT('PROJ-',TCV.tc_external_id), stale id 999 yields CONCAT('-',TCV.tc_external_id) with zero PHP warnings; endpoints HTTP 200, events 0. Fixture removed afterwards (DB back to fresh).
• R5 — Event hygiene: after the whole matrix events = 0 Error/Warning rows; Event Viewer screen shows all levels 0 (issue-1558-eviewer-clean.png); php -l lib/functions/testplan.class.php → No syntax errors detected.
• Code review (subagent) over commit 0f6d0550c: APPROVE — $concat remains a valid SQL expression in all three paths (valid id / non-existent id / null parent_id), no new warnings introduced.