Olive Young Resolves 10-Year Source-Deployment Mismatch Using Decompilers and AI

Olive Young's back-office system accumulated source code discrepancies between the version control server and the production server over 10 years of operation. The primary causes were procedural gaps, including missed incremental deployments via ANT and emergency fixes via SFTP that were not reverted to version control.

Recovery Strategy Using Decompilers and AI

The development team proceeded with decompiling .class files from the production server to compare them with the source in version control. Initially, they attempted to overwrite the version control source with the decompiled code but halted due to issues with lost comments and changed variable names. Subsequently, they introduced a method of cross-verifying decompilers such as CFR, JAD, and Fernflower, and using AI to classify source differences into high, medium, and low risk levels.

4-Step Standard Procedure and Verification

The recovery work was standardized into the following 4 steps.

1. Collection and Build: Prepare the production server classes and the compilation results of the version control source.
2. Decompilation Under Identical Conditions: Use the same decompiler to eliminate comparison noise.
3. Pinpoint Manual Application: Refer to AI analysis results but selectively patch only logic differences.
4. Recompilation and Comparison Verification: Use AI to compare the hash values and bytecode of the recompiled classes to confirm identity.

Results and Implications

Approximately 10% of the 1,000 WAS files and 20% of the 4,800 FE files were inconsistent, but a final consistency rate of 99% was achieved. While AI excelled in pre-analysis and classification, its reliability was low for code with complex framework dependencies, so final judgment and verification were performed by engineers. Following this work, Olive Young improved its deployment structure by migrating from ANT to Gradle, controlling SFTP access, and establishing an AI-based early warning system.