REDDY

Reddy is a self-built Reddit client written in Flutter, targeting Android, iOS, Windows, Linux, and web from a single codebase. Instead of Reddit's OAuth API, it authenticates via a WebView login flow and persists the resulting session cookies locally, which lets the app read a real logged-in feed without needing an approved API app. State management runs on GetX (RxList/Rx observables), with Hive used for structured local storage (search history) and SharedPreferences for auth session and settings.

The app went through two build phases: an initial feature build (feed, subreddit browsing, search, comments, video/gallery playback), followed by a dedicated audit-and-fix pass that hunted down and resolved a set of real correctness and performance bugs that had shipped silently.

Key technical achievements include:

• True infinite scroll, not "load more": the original feed replaced its entire post list on every "next page" fetch, so scrolling further silently lost earlier posts. Rebuilt around a growing, de-duplicated RxList<RedditPostModel> in HomeController, a scroll listener that prefetches ~800px before the bottom, and a real SliverList.builder (swapped out from a non-lazy ListView nested in a CustomScrollView) for genuinely lazy rendering.
• Session persistence and re-authentication: added an AuthController + RedditAuthService backed by SharedPreferences that reloads cookies before first frame, added a "Re-authenticate" flow that clears the cookie jar to support account switching, and centralized 401 handling so an expired session cleanly routes back to login instead of leaving the app broken.
• Gallery post rendering: Reddit gallery posts (is_gallery + media_metadata/gallery_data) weren't parsed at all, falling back to a raw URL. Added parsing into an ordered galleryImages list and a swipeable PageView carousel with a "1/N" counter.
• Killed an N+1 request pattern: every post in a feed page was firing an extra, un-awaited HTTP request for its author's profile during JSON parsing — 25 posts meant 25 extra requests before a single frame rendered, a common trigger for Reddit rate-limiting. Moved author avatar fetches to be lazy (triggered only when a post scrolls into view) and cached in-memory by username with in-flight de-duplication.
• Fixed silently-dropped cookie rotation: the client was reading response.headers['Cookie'] (a request header) instead of set-cookie (the actual response header), so Reddit's rotated session cookies were never being persisted — fixed and wired through AuthController.updateCookie.
• Memory-aware image decoding: every image loaded via Image.network with no disk cache, and even after switching to cached_network_image, images were decoded at full source resolution regardless of display size (a 4000×3000 image shown in a 400px card decoded ~10x more pixels than were ever rendered). Set memCacheWidth/memCacheHeight based on actual on-screen size × device pixel ratio, and wrapped feed items in RepaintBoundary so scrolling no longer forces every post card to repaint together.
• Site-wide search with sort-by: added a search mode that detects r/<subreddit> shorthand versus a general query, runs general queries against /search.json with the same infinite-scroll feed UI, and added a Best/Hot/New/Top/Rising sort menu — transparently remapping sort values Reddit doesn't actually support on certain endpoints (e.g. controversial isn't valid on /search.json and is mapped to top; best isn't a real per-subreddit listing and is mapped to hot).
• Removed a live security issue: an unused, unrouted test screen was found to be posting raw session cookies to a public beeceptor.com webhook; deleted along with other logging that leaked cookies/response bodies to the console.

The result is a Reddit client that behaves closer to reddit.com's own search/sort semantics than a naive wrapper, with feed scrolling, image memory, and session handling tuned for actual mobile constraints rather than desktop-first defaults.