How an incremental JSON parser fixes SSE stream fragmentation in chat apps

·2 views

A developer building a streaming chat interface discovered that their JSON parser was crashing because Server-Sent Events (SSE) does not guarantee each event contains a complete JSON object. The SSE protocol flushes whatever bytes are ready, meaning a single JSON object can be split across multiple events or several objects can arrive in one event. The naive approach of calling json.loads() on each event's data field failed whenever a fragment lacked proper opening or closing braces. The fix involved using Python's json.JSONDecoder.raw_decode to build an incremental parser that buffers incoming chunks and only yields fully formed objects. This approach allows split fragments to reassemble naturally and handles multiple concatenated objects arriving in a single chunk.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

What Is Data Sniffing and How Businesses Can Defend Against It

Data sniffing, also called packet sniffing, is the covert interception of data packets as they travel across a network, and attackers use it to steal passwords, financial data, and confidential communications. Unlike many cyberattacks, sniffers operate silently and can be difficult to detect, making them a persistent threat on both corporate and public networks. Businesses face serious risks including customer data leaks, executive credential theft, internal communication exposure, and supply chain compromise. A single infected or rogue device on a network can give an attacker broad access to sensitive information affecting employees, customers, and partners. Experts recommend a layered defence strategy that includes encryption, strong identity verification, network segmentation, anomaly monitoring, user education, and regular security testing.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Auto-Select GitHub PAT by Repo Owner Using gh CLI and .gitconfig

Developers managing multiple GitHub accounts over HTTPS often face authentication errors when Git defaults to the wrong Personal Access Token (PAT) for a given repository. A proposed solution embeds a custom credential helper directly into the .gitconfig file, eliminating the need for external scripts or manual account switching. The helper uses the repository owner extracted from the remote URL to run 'gh auth token --user OWNER', fetching the correct PAT automatically. Setting 'credential.useHttpPath = true' is essential so Git passes the full repository path to the helper, enabling owner extraction. An empty 'helper =' line before the custom helper ensures any pre-existing credential managers are reset, preventing conflicts on platforms like Git for Windows.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

MCP-powered math engine gives AI dungeon masters accurate tabletop dice mechanics

A developer has built a dedicated tabletop math engine using the Model Context Protocol (MCP) to fix unreliable dice roll simulations in LLM-based Dungeons and Dragons sessions. Rather than asking language models to handle deterministic arithmetic, the system offloads calculations — including advantage/disadvantage rolls, proficiency bonuses, and damage modifiers — to specialized external tools. This approach prevents so-called hallucinated randomness, where LLMs approximate probabilistic outcomes instead of computing them correctly. The MCP server exposes individual tools such as simulate_roll_outcome, resolve_ability_check, and calculate_damage, allowing the AI agent to focus on narration while receiving mathematically sound results. The developer also cautions against large data payloads per request, noting that memory exhaustion errors can occur when too much information is processed in a single tool execution.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why a Confirm Button Alone Cannot Secure AI Coding Agents

Security researchers warn that approval dialogs in AI coding agents create a false sense of security when the displayed action differs from what the system actually executes. Symlinks or other layers of indirection can cause a file write to land at a different path than the one shown to the user, meaning the human approves one operation while the machine runs another. The core problem is that most agent workflows collapse three distinct objects — user intent, the model's description, and the runtime's resolved operation — into a single prompt. Experts argue that approvals must bind to computed facts such as the resolved resource path, tool version, and credentials, and should be invalidated if any material detail changes before execution. A structured approval envelope covering preparation, display, binding, execution, verification, and recording is proposed as a more reliable alternative to broad session-level permissions.

0 comments Read more at DEV Community