Bugfix Issue 1738 Array Typed Parameter Trim TypeError - sebiboga/testlink-upgraded GitHub Wiki

Bugfix: array-shaped value for an input parameter raises an uncaught PHP 8 TypeError through trim() (Issue #1738)

Issue: an array-shaped query/body value (?k[]=x) for an input parameter reaches PHP's string/number functions and raises an uncaught TypeError → HTTP 500, 0-byte body, zero events rows (no observable log trail). The shared input layer is used by every TestLink controller through R_PARAMS() / P_PARAMS() / G_PARAMS(), so one bracketed parameter name is enough to kill a request from any screen.

Root cause

Chain (each hop file:line, verified on HEAD before the fix):

  1. PHP materialises ?k[]=x (and ?k[][]=x) as an array in $_GET / $_POST / $_REQUEST.
  2. lib/functions/inputparameter.class.php:153 — fetchParameter() reads the raw value.
  3. lib/functions/inputparameter.class.php:162-167 — the guard added for #1720 drops an array only when the validation type is a scalar one; it deliberately keeps arrays for tlArrayValidationInfo (ARRAY_INT / ARRAY_STRING_N).
  4. lib/functions/inputparameter.class.php:308 → tlStringValidationInfo::normalize() → trim() (:330): ltrim()/rtrim()/trim()/tlSubStr() have no type guard.
  5. lib/functions/inputparameter.class.php:424 → tlIntegerValidationInfo::normalize() → intval(trim($value)): same unguarded shape.
  6. lib/functions/inputparameter.class.php:501 → tlArrayValidationInfo::normalize() calls the member validator's normalize(); a nested array element still reaches steps 4/5.
  7. TypeError extends Error — it is not an Exception, so the catch (Exception) / catch (Throwable) guards at the call sites do not turn it into a graceful message; the failure happens before tLog(), hence 0 events rows.

Why it breaks now: on PHP 5/7 trim(array) only emitted a warning and coerced; PHP 8 turns it into a fatal TypeError. The #1720 fix closed the scalar case one level above the root cause; the guard the issue itself asked for (in trim()) was never added, so nested array-typed parameters remained fatal.

Fix (approach)

Add the is_scalar() guard the issue suggested at the three sites that hand a value to PHP's string/int functions. null is returned for a non-null non-scalar, so the parameter (or the array member) reads as absent — matching the contract the existing per-screen workarounds (shimReqScalar() in lib/reqmgrsystems/reqMgrSystemEdit.php, reqSpecEditDropShapedAction() in lib/requirements/reqSpecEdit.php) already assume.

  1. tlStringValidationInfo::trim() (:330) — if ($value !== null && !is_scalar($value)) return null; before the switch. Makes trim() itself never fatal for every caller, including direct / out-of-tree construction and array members.
  2. tlIntegerValidationInfo::normalize() (:430) — the same guard before intval(trim($value)).
  3. tlArrayValidationInfo::normalize() (:501) — unset() a non-scalar member so it never reaches the member normalizer (and the member validator never throws a spurious validation error for it).

Why "reads as absent" (null) rather than coercing to '': '' would silently pass the min/max-length validators; null is what the existing workarounds already return.

Rejected alternatives: a fourth fetch-layer guard (would again miss nested members and any direct tlInputParameter construction — the duplication the issue already laments); a global set_exception_handler (too broad for a parameter-shape defect).

Files changed

  • lib/functions/inputparameter.class.php — is_scalar() guards in tlStringValidationInfo::trim(), tlIntegerValidationInfo::normalize() and tlArrayValidationInfo::normalize() (+21 lines).

Verification

  • php -l lib/functions/inputparameter.class.php → clean.
  • Shared-layer CLI matrix: scalar STRING_N / INT / INT_N / CB_BOOL with ?k[]=x → null (unchanged, already fixed by 60b823d8d); ARRAY_INT nested ?k[][]=x → TypeError … trim(): Argument #1 … array given before, array() after; ARRAY_STRING_N nested no longer fatal.
  • Live HTTP (temporary docroot probe, removed after the test): ?doAction=hi&nested[]=7 → 200 {hi,[7]}; ?doAction[]=x → 200 {null,null}; ?nested[][]=x → 200 {null,[]}; ?doAction=++spaced++&nested[]=1&nested[]=2 → 200 {"spaced",[1,2]}. No 0-byte 500.
  • events rows with log_level<=2: 0 before → 0 after (no new Error/Warning).
  • Regression suite Regression — Issue #1738 (14 cases) 14/14 PASS; suite gate 7/7 (bash ai/verify_test_suites.sh).
  • No user-facing string added → no i18n bundle touched.

Related

  • #1720 — the scalar-parameter manifestation of the same defect (fixed at the fetch layer by 60b823d8d / aa53eab6e); this issue completes it at the trim() root cause.
  • #1731 (reqMgrSystemEdit.php), #1732 (listTestCases.php), #1736 (reqSpecEdit.php) — per-screen is_scalar() workarounds for the same shape.