Bugfix Issue 1571 DatetimeCf Wiped On Save

Bug fix — Issue #1571: datetime design CF value wiped on save + E_WARNING "Undefined array key input"

Symptom

Design-time DATETIME (type 10) Custom Field values are silently wiped to empty whenever the test case version is saved through the Test Case Editor (modern tcEdit.html and the legacy editor alike), and a PHP E_WARNING is emitted on every such save. Measured on the freshly-imported DB (fixture recreated per FIX-ISSUE §2): after the save, SELECT value FROM cfield_design_values WHERE node_id=9 AND field_id=4 returned nothing (row DELETED) and the events table gained log_level=2 (WARNING) E_WARNING "Undefined array key \"input\"" ... lib/functions/cfield_mgr.class.php - Line 1973.

Reproduction (this run, fresh DB)

1. Seeded fixture /tmp/opencode/fixture1571.sql: TCEDDemo (tproject 6), testsuite node 7, testcase tcA node 8, tcversion 9 (v2), design CFs f1 Severity (type 6/list), f2 Priority Flag (type 5/checkbox), f3 Notes (type 20/textarea), f4 DateTime (type 10/datetime); cfield_design_values node 9 = f4 1767638400 plus f1/f2/f3 values.
2. curl login (admin/admin) + POST /api/testcasesedit/?action=update with the legacy hash shape produced by collectCustomFields() (gui/templates/testcases/tcEdit.html:577): custom_field_10_4_input='2026-09-23 00:00:00', custom_field_10_4_hour=12, custom_field_10_4_minute=30, custom_field_10_4_second=45.
3. Pre-fix: HTTP 200, f4 row deleted, E_WARNING logged. Post-fix: HTTP 200, f4 row kept (1868531445), zero new events rows.

Root cause chain (lib/functions/cfield_mgr.class.php)

• The BFF persists design CF values through design_values_to_db() → _build_cfield() (:825, :834).
• _build_cfield (:1907-1937) iterates the hash one key at a time. For a date/datetime field the hash carries one key per part (_input/_hour/_minute/_second). The pre-fix loop reset $the_value = null (:1925) and assigned a fresh single-element array on every suffix key: $the_value[$dummy[$last_idx]] = $value; (:1929). Iterating input → hour → minute → second therefore left $the_value = ['second' => 45] — the last suffix wins.
• The datetime branch (:1981-1995) then read $value['input'] on that ['second'=>45] map → PHP 8 E_WARNING Undefined array key "input" → $cfield[$field_id]['cf_value']=''.
• Back in design_values_to_db (:875-878): the row exists (rowCount>0) and the value is empty → the branch issues DELETE FROM cfield_design_values. Silent loss of the persisted value.
• Any page that posts only the bare _input key (no time selects) kept the value, which is why the failure only manifested on pages submitting the full date+time split.

Second latent defect exposed by the fix

Once the merge fix lets the datetime branch run with a real map, mktime() (:1994) received the raw string parts. Numeric strings coerce fine under PHP 8.3, but the reported ISO-shaped input 2026-09-23 00:00:00 under the d/m/Y locale parses to year = "23 00:00:00" (non-numeric) → TypeError → HTTP 500. This was previously unreachable (the E_WARNING short-circuited to an empty cf_value before mktime). Fixed with (int)/intval() casts — behavior-identical for legitimate values, immune to non-numeric remnants.

Fix approach

1. Merge date-part suffixes (cfield_mgr.class.php:1927-1945): each date part now accumulates into a single per-field map. cf_value is pre-seeded from $cf_map (:1884-1887, 'cf_value'=>''), so is_array($cfield[$field_id]['cf_value'] ?? null) distinguishes "first part of this field" (→ init the map with the part) from "subsequent parts" (→ merge into the map). Non-suffix keys keep the exact legacy $the_value = $value overwrite path (#0008347 comment preserved).
2. PHP 8 int hardening of date (:1970-1975) and datetime (:1981-1994) branches: all mktime() arguments go through (int)/intval(). No change in behaviour for the numeric values the UI produces; no crash for anything else.

Alternatives considered and rejected

• Clamping/validating $value['input'] against the locale format and rejecting ISO-shaped input server-side — out of scope: the UI never sends it (the showCal shim writes the localized format), and rejecting input was not the reported defect.
• Rewriting _build_cfield around a real date parser — too large an architectural change for a late-phase PHP 8 compat repo; the merge + cast is the smallest change that preserves every legacy branch.
• The multi-owner CF keying (custom_field_<type>_<id>_<owner>_<part>) keeps its pre-existing $cfid_pos semantics (per-field, last owner wins) — verified unchanged via a CLI harness; out of scope.

Verification (measured)

• php -l lib/functions/cfield_mgr.class.php → no syntax errors.
• Issue's exact repro POST → HTTP 200, cfield_design_values node 9 = f4 1868531445, f1 High, f2 Yes, f3 some notes; events table empty.
• Empty datetime input → f4 row removed (intended clear semantics), others intact, no warning.
• Realistic localized round-trip in the modern editor (browser): 23/09/2026 + 12/30/45 rendered from the stored timestamp; native-picker overlay → 28/09/2026; hour select → 14; Save → f4 1790605845 == mktime(14,30,45,9,28,2026); events only audit INFO (audit_login_succeeded).
• CLI harness for edge shapes: date single-_input → mktime(0,0,0,9,22,2026); datetime single-_input → midnight timestamp, no warning; execution-page owner shape (custom_field_10_5_234_* with cf_map=null) → field 5 mktime(8,15,0,9,22,2026), string field intact.
• Browser console on tcEdit.html: no JS errors from the fix; two pre-existing cosmetic 404s for the legacy calendar.gif/trash.png theme images in the injected datetime markup (present pre-fix, PHP-only change left them as-is).

Files changed

• lib/functions/cfield_mgr.class.php — merge suffix keys + int-cast the mktime() args (23 insertions, 12 deletions).
• tmp/TLU_Test_Cases.md — Regression — Issue #1571 suite (7/7 PASS).
• CHANGELOG — [KEY BUGFIX] - #1571 entry.
• Wiki mirror page Bugfix-Issue-1571-DatetimeCf-Wiped-On-Save.md (with screenshots) + this docs mirror.

Result

Issue #1571 fixed error-free and pushed (fix/issue-1571); regression suite 1571 7/7 PASS; Event Viewer clean.