zenmemes

New post

Feature plan: Step 1

Site Text Search Step 1 Solution Assessment

Problem Statement

Users need a site search path that lets them enter text and find matching visible forum content without leaving the application.

Option A: Add simple SQL-backed search over the existing read model

Pros:

  • Fits the current PHP application and SQLite read-model architecture.

  • Keeps the first user-facing version small by querying existing posts and threads content.

  • Avoids database migrations and new services.

  • Easy to expose as a normal page with graceful empty-query and no-result states.

Cons:

  • Relevance ranking will be basic.

  • Substring matching can get slower as content grows.

  • Search behavior may differ from the existing related-content ranking path.

Option B: Add SQLite full-text search for public site search

Pros:

  • Better suited to user-entered text search than ad hoc LIKE matching.

  • Can provide more useful ranking, tokenization, and snippet behavior.

  • Keeps search local to SQLite without adding an external service.

Cons:

  • Requires read-model/index changes and rebuild considerations.

  • Adds migration and compatibility risk around SQLite FTS availability.

  • More upfront work before proving the desired page and workflow.

Option C: Introduce a reusable search service boundary first, backed by the existing read model

Pros:

  • Lets the public search page and future callers share one search contract.

  • Can start with simple SQL and later move to FTS behind the same boundary.

  • Aligns with the existing internal related-content search pattern.

  • Keeps UI, route, and search behavior testable without committing to a permanent ranking engine.

Cons:

  • Slightly more design work than directly querying inside a page handler.

  • Initial relevance still depends on the first backing strategy.

  • The boundary must stay narrow to avoid premature abstraction.

Recommendation

Recommend Option C.

Brief justification:

  • The main uncertainty is product behavior and ranking quality, while the app already has SQLite read-model content and precedent for a search service.

  • A small shared boundary supports the first public search page now and leaves room for SQLite FTS later without reshaping the route or template.