refactor(plugins): unify the package PluginKit on the framework contract (R-001)

[datlechin]

Summary

Phase 1 (R-001): the SPM TableProPluginKit (Packages/TableProCore/Sources/TableProPluginKit) was a stale, divergent copy of the Xcode-framework TableProPluginKit (Plugins/TableProPluginKit). This promotes the package to the framework's (current, ABI v16) content so the two trees are byte-identical except Info.plist. The physical collapse to one source is the follow-up (7.2).

How to review

Don't read the 2000-line diff line by line. Instead:

1. Run diff -qr Plugins/TableProPluginKit Packages/TableProCore/Sources/TableProPluginKit — it should report only Info.plist. That proves "package == framework now."
2. Review only the three real logic changes below.

The three logic changes

• #if canImport(AppKit) gates DocumentInspectorPlugin in both trees (the only macOS-only file; InspectorWindowFactory uses NSDocument/NSWindowController). macOS compiles it unchanged; iOS excludes the block. The protocol itself (documentClass: AnyClass) is platform-neutral.
• QueryResult.init(from: PluginQueryResult) converts PluginCellValue -> String? (null→nil, text→value, bytes→uppercase hex). This is the entire downstream cascade of PluginQueryResult.rows becoming [[PluginCellValue]]. Mobile never constructs PluginQueryResult nor calls this init, and QueryResult.rows/Cell are unchanged, so there is no mobile data-path migration. Covered by a new test (mapsCellValuesToLegacyStrings).
• Two package-only files removed:
  • CompletionEntry.swift — not abandoned; relocated. The framework defines CompletionEntry inside SQLDialectDescriptor.swift, so once that file is adopted, the standalone copy would be a duplicate definition. CompletionEntry is still used by the desktop autocomplete.
  • PluginPagedResult.swift — dead (zero references repo-wide; tied to the dropped legacy paging API).

Drift baseline trimmed (.github/duplicate-contract-baseline.txt): all pluginkit: entries removed except Info.plist.

Risk Addressed

• R-001: duplicate, drifting PluginKit contracts (a plugin could compile against one and load against another).

Verification

• swift build / swift test --package-path Packages/TableProCore (host; 62 tests).
• swift build --triple arm64-apple-ios17.0 --sdk <iphoneos> — package cross-compiles for iOS (proves AppKit gating + iOS-safety).
• CI Run iOS Tests pass — full mobile build/tests green.
• xcodebuild -scheme TablePro build (macOS app).
• scripts/audit-refactor-health.sh --check (drift gate green).

Notes

• ABI bump required: no. The framework (what plugins link) is unchanged except the macOS-no-op #if canImport(AppKit) gate; currentPluginKitVersion stays 16. No plugin rebuild.
• Forward note: the bytes → hex conversion in QueryResult.init(from:) is currently callerless (kept deterministic, pinned by test). Whoever wires the desktop path (8.2b) through this init should revisit it — the desktop wants full PluginCellValue fidelity, not hex String?.
• Unblocks 7.2 (collapse the two trees into one physical source) and 8.2b (desktop import TableProModels once there is a single TableProPluginKit module).

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cc8ec90751

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Make byte parameter literals dialect-specific

When a driver relies on this default executeParameterized implementation, .bytes parameters are substituted as X'...'; I checked the plugin tree and Oracle/BigQuery do not provide their own executeParameterized overrides, while both can surface binary values as PluginCellValue.bytes. Row edits or user-parameterized queries involving RAW/BYTES columns on those drivers will therefore emit a literal form those dialects do not accept, so the save/query fails despite the new binary cell support. Please either require those drivers to override byte binding or move byte literal generation behind a dialect-specific hook.

Useful? React with 👍 / 👎.