Inside WeChat’s Tinker: Fast Android Hot‑Patch Techniques Explained

Why Hot Patches Are Needed

Hot patches allow apps to be updated without reinstalling, enabling rapid, dynamic fixes.

While hot patches reduce release time and avoid user reinstall, they have limitations: patches target a single client version, cannot modify AndroidManifest, and cannot guarantee 100% success for code or resources.

Suitable Scenarios

1. Lightweight and fast upgrades – Small patches (≈300 KB) can be delivered within a day to over 70% of users, far quicker than traditional releases.

2. Remote debugging – Patches can be sent to specific devices to fix issues that cannot be reproduced locally.

3. Data statistics – Patches enable A/B testing on the same user base without requiring multiple app installations.

Evolution of WeChat’s Hot‑Patch Technology

WeChat evaluated several solutions and built its own framework, Tinker, after finding limitations in Dexposed, AndFix, and Qzone.

AndFix

AndFix uses a native hook via dalvik_replaceMethod to replace method implementations. It cannot add or remove fields because class layout is fixed at load time, limiting its use to small fixes. It offers immediate effect but introduces compatibility complexity.

Qzone

Qzone (implemented in the open‑source Nuwa) replaces classes via a custom classloader, providing better stability but requiring instrumentation that can affect performance on both Dalvik and ART. Instrumentation may increase verify/optimize time and can cause address mismatches on ART.

WeChat’s Tinker

Tinker combines full‑dex replacement with a binary‑diff approach. Using a BsDiff‑like algorithm (DexDiff), it generates a small patch containing only the differences between the old and new dex files. At runtime, the patch dex is merged with the original dex to produce the updated version. This approach reduces patch size (e.g., a 13 MB dex can be patched with a 6.6 MB diff) but adds ROM overhead (≈1.5× the modified dex size) and requires a background synthesis process that consumes time and memory.

The framework also includes version management, process handling, security checks, and supports both Gradle and command‑line integration.

Current Deployment Practices

WeChat’s hot‑patch system includes three delivery channels:

Pull channel – periodic checks for updates.

Version‑specific push – rapid rollout to all users within an hour.

User‑specific push – targeted remote debugging.

Management tools provide real‑time monitoring of patch adoption, crash rates, and usage metrics with minute‑level granularity.

Success rates are around 98.5% for Qzone‑based patches and 95.5% for Tinker, with ongoing efforts to improve reliability.

Future Work

WeChat aims to modularize its components and open‑source Tinker and related tools, facilitating reuse across multiple apps and encouraging community contributions.

References

Dexposed GitHub

AndFix GitHub

Nuwa GitHub

Qzone implementation analysis

Instant Run (English)

Instant Run (Chinese translation)

Buck exopackage introduction

Hot‑Patch scheme comparison article