fix(ios): stop ForEach indices race in row lists and validate driver port range

[datlechin]

Summary

Two distinct production crashes hit by TestFlight users on TableProMobile build 10/11. Both fixed in this PR.

Crash 1 — Array._checkSubscript / "Index out of range"

Stack trace (build 11, iPhone 15, iOS 26.4.1):

Thread 0 Crashed:
0  Swift runtime failure: Index out of range
1  specialized Array._checkSubscript
2  specialized Array.subscript.getter
3  closure #1 in DataBrowserView.rowList.getter (DataBrowserView.swift:259)
4  ForEachChild.updateValue() (ForEach.swift:2068)

Root cause: ForEach(rows.indices, id: \.self) { index in ... rows[index] ... }. SwiftUI captures rows.indices once but rows is a computed property reading viewModel.legacyRows. When the array shrinks during a SwiftUI update pass (page change, filter, deletion), ForEach still iterates the stale indices and rows[index] traps. CLAUDE.md flags this exact pattern as a "Performance Pitfall".

Fix: introduce IndexedRow: Identifiable wrapper (Helpers/IndexedRow.swift) and iterate [IndexedRow] instead. SwiftUI diffs by stable id; if rows shrink, the missing ids are removed cleanly without ever subscripting a stale offset. Applied to all three sites: DataBrowserView, QueryEditorView, RowDetailView.

Crash 2 — Swift runtime failure: Not enough bits to represent the passed value

Stack trace (build 10, iPhone 16):

Thread 6 Crashed:
0  Swift runtime failure: Not enough bits to represent the passed value
1  MySQLActor.execute(_:) (MySQLDriver.swift:297)
2  MySQLDriver.execute(query:) (MySQLDriver.swift:56)
3  DataBrowserView.loadData(isInitial:) (DataBrowserView.swift:526)

Root cause: mysql_real_connect(..., UInt32(port), ...) where port: Int. If the connection's port is negative or > UInt32.max, UInt32(port) traps. The form parses port via Int(portString) ?? 3306 — a typed -1 slips through. Same shape exists in RedisDriver (Int32(port)).

Fix: validate port is in 1...65_535 before each driver-level integer cast and throw a readable connectionFailed instead of trapping. Applied to MySQL, Redis, and PostgreSQL (consistency — PG was string-interpolating so it didn't crash, but it silently passed garbage to libpq).

Test plan

• Crash 1 manual repro: open a table with many rows, scroll, then quickly switch tables / change filter — no crash on rapid re-render.
• Crash 2 manual repro: edit a connection, set port to -1 or 99999999999, save, tap connect. Expect a readable error sheet, no crash.
• Regression: normal flows (open table, paginate, run query, view row detail, swipe through rows in RowDetailView) work as before.

Files changed

• TableProMobile/Helpers/IndexedRow.swift (new) — Identifiable wrapper
• TableProMobile/Views/DataBrowserView.swift — use IndexedRow.wrap
• TableProMobile/Views/QueryEditorView.swift — use IndexedRow.wrap
• TableProMobile/Views/RowDetailView.swift — use IndexedRow.wrap
• TableProMobile/Drivers/MySQLDriver.swift — port range guard before UInt32(port)
• TableProMobile/Drivers/RedisDriver.swift — port range guard before Int32(port)
• TableProMobile/Drivers/PostgreSQLDriver.swift — port range guard (parity / readable error)
• CHANGELOG.md — Fixed entries

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.