Douyin HarmonyOS: 50% Dev Efficiency Boost via Indexing, Memory & Build Optimizations

Core Challenges in Large-Scale HarmonyOS Engineering

Douyin's HarmonyOS app uses a monorepo multi-module architecture to support multi-team collaboration, unified dependency versions, and simplified build complexity. However, as the codebase grew, the team faced severe efficiency bottlenecks: slow project sync, high memory consumption, and long full-build times, impacting daily debugging, feature iteration, and release cadence.

Three Core Optimizations

1. Code Indexing Time Reduced 80% (10 min → 2 min)

In high-frequency scenarios like project startup and branch switching, DevEco Studio must build a full code index to enable search, completion, and navigation. For Douyin's massive codebase, full indexing took 10 minutes. Root cause analysis revealed the original TSC compilation mechanism created separate Program objects per module, causing redundant object creation due to differing module contexts and isolation rules.

The teams jointly redesigned the TSC compilation mechanism, introducing a project-level unified Program architecture that centrally schedules global file dependencies, symbol information, and shared dependency packages. After optimization, full indexing time dropped from 10 minutes to 2 minutes — an 80% improvement — eliminating the indexing bottleneck.

2. DevEco Studio Memory Usage Reduced 65% (20 GB → 7 GB)

Large projects require the IDE to continuously load runtime data, full code indexes, and syntax trees. Previously, Douyin's project consumed 20 GB of RAM, causing system pressure when developers ran browsers, office apps, and communication tools simultaneously.

The teams analyzed the LSP cross-platform architecture and optimized the language service implementation. On the server side, they abandoned full-file syntax tree parsing, instead performing partial parsing for non-visible files and loading only core signature nodes. On the client side, they refactored capability logic based on the PSI architecture, optimizing node invocation for high-frequency operations like find references, safe delete, and rename. These changes cut memory usage from 20 GB to 7 GB, a 65% reduction, resolving device lag and memory pressure.

3. Build Efficiency Improved 30% (4.5 min → 3 min)

Build profiling showed compilation time concentrated in three phases: dependency installation, language compilation, and compression/packaging. The teams implemented project-level fine-grained control: unified lockfile management to avoid duplicate dependency downloads and installations, streamlining pre-build steps. For language compilation, they merged syntax-checking passes, reducing object creation and intermediate artifacts. For packaging, they enabled incremental .so compression, reusing already-compressed native libraries. Overall build time fell from 4.5 minutes to 3 minutes, a 30% gain, accelerating the version build-and-release pipeline.

Outcome and Future Direction

All optimizations are now integrated into DevEco Studio 26. The improvements not only support Douyin's rapid HarmonyOS iteration but also provide a reusable reference for other large-scale HarmonyOS applications. The case demonstrates that boosting development efficiency at scale requires deep collaboration across engineering systems, toolchains, and real-world application practices. Both teams plan to continue iterating on large-project capabilities, further lowering the barrier for massive HarmonyOS apps and strengthening the ecosystem's engineering foundation.