Issue: #1722 Commits: f9cc73b0e (fix) + 099a44704 (regression suite) Branch: fix/issue-1722 (pushed; the CI merge-prs.yml lands it) Status: VERIFIED-FIXED — regression suite 31/31 PASS (bash tmp/verify_1722.sh, exit 0) Direct successor of: #1627, which fixed the non-whitelisted doAction and listed this exact blank 200 under "Known, deliberately NOT fixed here".

reqMgrSystemCommands::$guiOpWhiteList advertises 7 actions, but renderGui()'s switch ($argsObj->doAction) handles only 5. The remaining two — delete and checkConnection — passed #1627's whitelist validation (they are whitelisted), fell out of the switch with no branch taken, left $renderType at its initial 'none', and were then swallowed by the second switch's default: break;. The request ended as HTTP 200 with a 0-byte body: a blank page, and — for delete — not a single Event Viewer row, so the failure left no trace anywhere.

Unpatched, — the title was the raw URL, no markup, no error, no way back.

1. Neither shipped theme can post those values. The only value the edit form ever posts is {$gui->operation} (gui/templates/tl-classic/reqmgrsystems/reqMgrSystemEdit.tpl:129,131 and the dashio copy at :129,131), and $actionOperation (reqMgrSystemEdit.php:40-42) only ever yields doUpdate / doCreate / ''. A repo-wide grep for checkConnection finds zero occurrences in any reqmgrsystems .tpl.
2. The modernized screen does not use this controller at all: the wrench button calls checkConnection(id) (gui/templates/reqmgrsystems/reqMgrSystemView.html:182,185,274), which is served by the BFF api/reqmgrsystems/index.php:294-330 → tlReqMgrSystem::checkConnection().
3. delete has no method on the command class at all, and the themes use doDelete.

Why it breaks now: step 7. The white list and the case list were never in sync — a 1.9.20 defect inherited as-is. What 2.0.1 changed is the contract: #1627 established the house rule that "this controller cannot do what you asked" is answered with a graceful 302 back to the list (reqMgrSystemEdit.php:150-153). #1722 was the one remaining way to answer the same situation without that rule. checkConnection used to be reachable from the legacy edit form and delete used to be a real method; both entry points disappeared and nothing was updated here.

A second, latent path into the same sink: #1627's hardening if (is_null($opObj)) { break; } (reqMgrSystemEdit.php:54-57) also leaves $renderType === 'none'. Unreachable today (all 5 case values are real methods), but it is the same defect and the same fix closes it.

lib/reqmgrsystems/reqMgrSystemEdit.php — the default: branch of the $renderType switch (the sink of step 7) now does what init_args() does for a non-whitelisted action:

      default:
        // Refs #1722: $renderType is still 'none' here, which means this controller
        // can not render the requested action. …
        tLog('reqMgrSystemEdit.php - requested action is not renderable - Value:' .
             $argsObj->doAction . ' - File: ' . basename(__FILE__) . ' - Function: ' . __FUNCTION__,
             'ERROR');
        $base = isset($_SESSION['basehref']) ? $_SESSION['basehref'] : '/';
        header('Location: ' . $base . 'gui/templates/reqmgrsystems/reqMgrSystemView.html', true, 302);
        exit();

• The $renderType switch, not the doAction switch. It is the single choke point every unrenderable path already funnels through, so one branch closes both step 5 (no case) and the latent step-4b variant (case matched but $op is null). A default: on the outer switch would have left the latter still blank.
• 302 to the modernized list, byte-identical to init_args() (:163-169). The two rejections then look the same to the user, to a log reader and to a test. A 4xx/500 would be worse (it turns a cosmetic dead-end into an error); die() with a message is worse still (raw text in the user's face, breaks the frame-based layout).
• One tLog ERROR row. Measured: delete produced 0 rows of any kind, which is why the blank page was undiagnosable. A one-line reason makes the next occurrence greppable, and it matches the sibling rejection at init_args():166.
• Rejected — remove delete/checkConnection from the white list. Would make #1722 unreachable rather than graceful, and would rewrite the command class' contract (the list also sanitizes input for other consumers) for a minor dead-URL gain. Larger blast radius than the bug.
• Rejected — add a case 'checkConnection' and render it. The modernized list already renders connection_status from the BFF (api/reqmgrsystems/index.php:294-330, reqMgrSystemView.html:182,185); a second renderer would be dead code for a URL no button emits.
• Rejected — also fix initGuiBean()'s Undefined array key "checkConnection". That is #1628, a separate defect with its own issue; mixing it in would break the one-bug-per-run rule.

Unchanged. The new branch performs no command and no write — it logs and redirects. The controller still sits behind checkRights() → reqmgrsystem_management (reqMgrSystemEdit.php:219-222, applied by testlinkInitPage() at :15 before initScript()), so an unauthorised user still never reaches renderGui(). Nothing that was previously unreachable became reachable.

Regression suite #1722, bash tmp/verify_1722.sh — 31/31 PASS (exit 0). It creates and deletes its own reqmgrsystems row, so it is re-runnable on a fresh import.

Negative control: with only the controller file reverted to origin/sebiboga, the very same harness reports 24 PASS / 7 FAIL (exit 1). All 7 are genuine behaviour: 3 × delete (status / Location / log), 3 × checkConnection (status / Location / log) and the 9-value sweep, which catches all three blank-200 values (delete, checkConnection&id=1, delete&id=1) in one assertion. No assertion in the harness is self-referential — a first draft ended with a grep for the new log string, which could only prove the string existed; it was replaced by that sweep during code review precisely so the negative control proves behaviour. Every shared behaviour passes before and after — that is the anti-regression proof that the new default: never fires during normal use. php -l is asserted unconditionally, so the count is exactly 31 with or without php on the PATH.

• #1721 — ?doAction=edit&id=<missing id> still logs 5 × E_WARNING Trying to access array offset on null in the compiled reqMgrSystemEdit.tpl.php (unguarded reqMgrSystemCommands.class.php:164, getByID() → null). Re-confirmed during this run that it does not occur for an existing id (0 warnings for the row the harness created).
• #1628 — E_WARNING Undefined array key "checkConnection" at reqMgrSystemCommands.class.php:72. Still present, now alongside the new ERROR row.

bash tmp/verify_1722.sh        # 31/31 PASS expected, exit 0