Insights, tutorials, and stories from the world of software development.
If a subscriber registers only after an event has already been published, it will miss that data. To prevent required data from being lost, the system needs a clear delivery contract: wait for all modules to subscribe, replay the latest state, or store history for replay.
This article explains that the key question is not whether to use a static class or a regular class, but how the event’s scope and lifecycle should be managed.
This article explains the performance problem of Draw Call in Unity and compares two common batching techniques: Static Batching and GPU Instancing. It describes how Static Batching combines static meshes into one large mesh at build time, making it fast but heavy on RAM and impossible to move, while GPU Instancing lets the GPU draw many copies of the same mesh with one Draw Call, using far less memory and allowing movement. The post also includes a comparison table, practical use cases, and pro tips on when to use each method, including direct rendering with Graphics.DrawMeshInstanced and Graphics.DrawMeshInstancedIndirect.
FixedUpdate() keeps the physics simulation step stable, not the render frame rate, so it should not be used for every game task. The article explains why input, UI, camera, and animation belong in Update() or LateUpdate(), while physics work like Rigidbody movement belongs in FixedUpdate(). It also shows a simple pattern: read and store player input in Update(), then apply it to physics in FixedUpdate(), and warns about the spiral of doom when heavy logic is added to physics ticks.
A deep dive into the Prometheus Lab RAG chatbot architecture, covering the full pipeline from offline indexing to online query handling: collecting data from portfolio pages, blog posts, and internal assets; splitting text into chunks with sanitization; generating 1536-dimensional embeddings with text-embedding-3-small; storing chunks in Supabase with zero-downtime atomic swaps; performing semantic search with pgvector and cosine similarity; streaming answers with NDJSON; and applying security measures such as prompt-injection protection, access control, rate limiting, validation, and output sanitization.
This article explains why a large Texture Atlas can outperform many small textures in Unity: the main benefit is reducing texture/material state changes and improving batching, not automatically saving VRAM. It uses a simple GPU printing analogy, shows how atlases can cut draw call overhead, and notes that memory usage depends on packing, padding, mipmaps, and other settings. It also warns that an oversized atlas can backfire by increasing memory and load time, then gives practical advice for splitting atlases and a Unity checklist for testing with Frame Debugger and Memory Profiler.