Build plugin tables through createTableSql(), so optional columns get defaults

[mastacontrola]

Companion to fogproject #1275 (merged), which added the seam. GH-1245.

The problem

A column declared NOT NULL with no DEFAULT is only mandatory if something enforces it. For nine years nothing did — PDODB cleared sql_mode on every connection, so the server downgraded the error to a warning and substituted an implicit zero. Removing the clear turned every one of those declarations into a real constraint, and an INSERT that omits one now fails with error 1364.

FOG's schema step 348 repairs the tables an install already has. It cannot repair ours: createSql() runs as step 0 of each plugin's own schema(), so a plugin installed after the migration gets a table built the old way — every optional column mandatory again.

The change

FOGManagerController::createTableSql() takes the identical arguments and fills a default into every NOT NULL column that has none, leaving bare only:

• the primary key and auto-increment column,
• anything the model declares required,
• anything whose name ends in ID — an INSERT that forgets the row it hangs off should fail, not make a silent orphan.

A default a manager passes explicitly always wins, so nothing already stated here changes. 25 call sites, one token each.

Verification

Against the live 1.6 install (prove_plugin_table_defaults_16.php in the fogproject scripts calls every manager's createSql(), creates the table under a scratch name, and reads the result back out of information_schema):

	before	after
tables built, DDL accepted	23/23	23/23
columns carrying a default	29	60
columns left bare	83	52

Each of the 52 is bare for one of the three stated reasons. Negative control — the wrapper filling nothing — puts 31 columns back to optional-and-still-mandatory.

WolbroadcastManager builds its table inside install() rather than createSql(), so the probe can't call it; its call site is switched the same way and is covered by the new test.

Tests

New tests/tables-carry-column-defaults.test.php — asserts no manager calls Schema::createTable() directly and every table-builder goes through the wrapper. It strips comments first, because two managers mention Schema::createTable() in prose explaining how it names its indexes. 49 checks; two mutations, two caught (a reverted manager, and a scan that reaches nothing).

The stub gains a pass-through createTableSql(). It deliberately does not reimplement the rule: the real one only adds defaults, so a pass-through cannot mask an assertion about a default a manager sets deliberately — which is exactly what oidc-provider-safety checks (opEnabled, opJITProvision, opAllowAPI must ship off) — and a copy of the rule would only give it somewhere to drift from.

Suite: 8/8.

Shipping

Needs a release, and then FOG_PLUGINS_VERSION bumped in fogproject to match. fogproject #1275 is already merged, so the seam is present on working-1.6.

🤖 Generated with Claude Code