Bugfix — Issue 1567 — SQL 1066 "Not unique table/alias 'EB'" in exec tree when a combined bug+platform filter matches nothing

Issue: #1567 Branch: fix/issue-1567-eb-alias-collision Status: FIXED & VERIFIED (2026-09-22)

Symptom

Running the execution-navigator tree through the modern BFF with a linked-bug filter whose bug id has no matching execution_bugs row produces the MySQL error

ERROR 1066 (42000) at line 1: Not unique table/alias: 'EB'

Measured pre-fix (fresh DB, fixture tmp/fixtures_1562.php, admin/admin, tproject_id=1, testplan_id=2, build_open=1, platform_id=1):

• GET /api/execnavigator/index.php?action=init&tplan_id=2&tproject_id=1&filter_bugs=X-1&setting_platform=999&setting_testplan=2 → responded [200] with the legacy DB Access Error debug page (not JSON);
• events table got two DATABASE rows (log_level=1): 1066 - Not unique table/alias: 'EB' - /* Class:testplan - Method: getLinkedForExecTree sqlUnion - executions */ SELECT … JOIN execution_bugs EB ON EB.execution_id = E.id … JOIN execution_bugs EB ON EB.execution_id = E.id … AND EB.bug_id IN ('X-1');
• standalone SQL harness: same query piped to mysql → ERROR 1066 (42000).

Severity Low (crafted filter_bugs input only; the modernized screen exposes no bugs filter input), as reported.

Root cause chain

1. helper_bugs_sql() — lib/functions/testplan.class.php:1087-1106 — builds the bug filter join as JOIN execution_bugs EB ON EB.execution_id = E.id with the alias hardcoded to EB (:1103) and stores it as $my['join']['bugs'] (assigned at :6526-6527).
2. getLinkedForExecTree() (lib/functions/testplan.class.php:6166) emits $my['join']['bugs'] in the exec branch at two positions:
   • :6277 — in the first join block (before LEBBP and before JOIN executions E) — spurious;
   • :6293 — after JOIN executions E, annotated // need to be here because uses join with E table alias — intended.
3. When bug_id is set, $my['join']['bugs'] is non-empty at BOTH positions → duplicate alias EB in a single query → MySQL 1066 (execTree() → fetchRowsIntoMap(), lib/functions/execTreeMenu.inc.php:146).
4. The not_run UNION branch is NOT part of the collision: it is nulled whenever bug_id is set (:6211/:6215, return :6298). This corrects the report's original "UNION branch" hypothesis — see the issue comment trail.

Regression source

Upstream commit ff21e99201 (2020-04-04, "new class build.class.php" big testplan.class.php refactor) ADDED the spurious pre-LEBBP bugs join at :6277. The pre-refactor function emits the bugs join only once, at the post-E position (git show ff21e99201^:lib/functions/testplan.class.php).

Blast radius

• join['bugs'] occurrences: :6237 (not_run branch — harmless), :6277 (BUG), :6293 (intended). getLinkedForExecTreeCross/getLinkedForExecTreeIVU do not emit it.
• Consumers: execTree() ← tlTestCaseFilterControl::build_tree_menu() (execution_mode) ← BFF api/execnavigator/index.php and legacy lib/execute/execNavigator.php.

Fix (minimal)

Removed the spurious duplicate from the exec branch:

lib/functions/testplan.class.php
-                     $my['join']['bugs'] .        (pre-LEBBP block, :6277)

The intended post-E bugs join (:6293) is kept, so bug-filter semantics are unchanged (this also removes a latent forward-reference of E.id). No other files touched. Alternatives rejected: renaming the alias to EB2 (keeps a redundant join) or keeping only the pre-LEBBP join (requires MySQL-only forward reference).

Verification

• SQL unmatched (X-1): exactly ONE JOIN execution_bugs EB …, executes cleanly, 0 rows.
• SQL matching (BUG-123): returns the executed TC (exec_status=p).
• BFF end-to-end: filter_bugs=X-1&setting_platform=999 → 200 {"status":"ok"} empty tree; filter_bugs=BUG-123&setting_platform=1 → 200 status:ok 1 child; baseline unchanged.
• Event Viewer: events max id unchanged after post-fix runs — no new Error/Warning rows.
• Regression suite: tmp/TLU_Test_Cases.md → Suite 1567 — PASS.

Test matrix

Case	Pre-fix	Post-fix
bug_id unmatched (SQL)	ERROR 1066	OK, 0 rows
bug_id matching (SQL)	— (never ran)	OK, returns TC
BFF bug+platform filter	DB Access Error + DATABASE 1066 events	200 status:ok, empty tree
BFF no-filter baseline	OK	OK (unchanged)
Event Viewer	1066 ERROR rows	no new rows