specview.php PHP 8 Null-Argument TypeErrors Fix (Test Automation report)

Issue #428

Reports and Metrics → Test Automation (lib/results/testAutomationSpec.php) died with a raw fatal TypeError on any test project whose filtered test spec resolves to an empty set — e.g. every test case is Manual, so the "execution type = Automated" filter matches nothing:

Fatal error: Uncaught TypeError: array_diff_key(): Argument #2 must be of
type array, null given in lib/functions/specview.php:919

and, after guarding that first statement, a second fatal one hop later:

Fatal error: Uncaught TypeError: count(): Argument #1 ($value) must be of
type Countable|array, null given in lib/functions/specview.php:1554

Both statements relied on the PHP 7 behaviour "warn and return null"; on the PHP 8.3 runtime of TestLink 2.0.1 they are uncaught TypeErrors → HTTP 500. The same count($test_spec) guard exists in three places (the two tree-view spec builders and the flat one), so all three were fixed together.

Root cause

Chain (verified against revision 385ca3317, where both fatals were still reproducible):

1. testAutomationSpec.php:31-36 calls genSpecViewFlat() with $filters = array('exec_type' => TESTCASE_EXECUTION_TYPE_AUTO).
2. getTestSpecFromNode() initializes $allowedSet = null (lib/functions/specview.php); it is only assigned when an execution_type / cfields filter matches something. With zero automated TCs, filter_tcversions_by_exec_type() returns null ⇒ $emptySet = true, $test_spec = null.
3. The unconditional array_diff_key($tcversionSet,$allowedSet) then received (array, null) → TypeError #1.
4. Once guarded, execution continues; genSpecViewFlat() calls count($test_spec) on the legitimately-null return of getTestSpecFromNode() → TypeError #2.

The fix — approach

Two behaviour-preserving guards (landed on the default branch in commits 764d625e1 + 16ba44c5e; this run reproduced both fatals pre-fix, verified the post-fix render end-to-end and hardened the remaining warnings on the same path — see below):

// keep null flowing into the existing !is_null($setToRemove) guard:
// "no filter active => remove nothing"
$setToRemove = is_null($allowedSet)
               ? null
               : array_diff_key($tcversionSet,$allowedSet);

// equivalent to count() for arrays, null-safe:
if (!empty($test_spec)) { ... }

Alternatives rejected:

• (array)$allowedSet cast — diffs against an empty set, so $setToRemove becomes the whole $tcversionSet and the entire spec gets blanked (wrong results instead of a crash);
• guarding only the report controller — the two sibling count() sites would have kept dying for other screens.

Verification (measured)

Reproduction against a fresh DB (project Issue 428 Repro, suite Manual Suite, TCs Manual TC One/Two exec_type=MANUAL, plan Plan 428):

Case	Pre-fix	Post-fix
Report, all-manual project	HTTP 500, fatal at specview.php:919	HTTP 200, empty result set
Report, +1 automated TC (Auto TC One)	(unreachable before)	HTTP 200, lists exactly the automated TC
Shared-path smoke planAddTC.php?edit=testsuite&id=2&tplan_id=7	—	HTTP 200, all TCs listed
Event Viewer	—	no new Error/Warning entries

Note: calling the report without &format=0 exits silently at displayMgr.php:87-90 (tlog('Parameter format is not defined') + exit()); that silent exit masked my first reproduction attempt and is worth knowing when testing any report screen directly.

Follow-up hardening (same render path)

Fixing the fatals unmasked 5 downstream E_WARNINGs that had been cut short by the early script death; they are fixed under issue #641:

• $gui->tplan_name initialized in initializeGui() (lib/results/testAutomationSpec.php) — used by buildMailCfg();
• 'platforms' => false added to the $useFilter initializer (lib/functions/specview.php) — key was read but never declared;
• is_null($tcase_memory) checked before the array dereference in buildSkeletonFlat() — mirrors the sibling buildSkeleton() order.

Regression suite

tmp/TLU_Test_Cases.md → Suite 428 — Regression — Issue #428.