Skip to content

performance.md

Latest commit

 

History

History
56 lines (39 loc) · 2.26 KB

performance.md

File metadata and controls

56 lines (39 loc) · 2.26 KB

JSONMap Performance Notes

This document describes the current large-file graph behavior and the benchmark command used for v1.8 validation.

Large-File Mode

JSONMap switches to large-file graph mode when the active document is larger than 1 MiB.

Expected behavior:

  • Graph parsing and transformation run outside the React render path through the worker-backed processing pipeline.
  • Large JSON input emits a shallow preview graph first, then a deeper complete graph.
  • The complete large-file graph is capped to 500 nodes and 50 children per branch.
  • Branches beyond the configured depth or child limit are marked as deferred and can be loaded on demand.
  • Scalar labels are shortened for readability.
  • Full scalar values are omitted from graph node payloads in large-file mode to avoid duplicating large strings across UI state.
  • Full JSON content remains available in the editor, versioning system, export flow, and subtree actions.

Benchmark Fixtures

The graph benchmark uses generated in-memory JSON fixtures so the repository does not store large blobs.

Fixture Target size Purpose
1 MiB 1,048,576 bytes Minimum large-file mode boundary.
5 MiB 5,242,880 bytes Medium local API payload or exported dataset.
20 MiB 20,971,520 bytes Stress fixture for local-first graph processing.

Running Benchmarks

Run:

pnpm benchmark:graph

The command generates all fixtures in memory, runs processGraphRequest, and prints a Markdown table with:

  • generated byte size,
  • processing duration,
  • emitted processing stages,
  • preview and complete graph node counts,
  • deferred branch count,
  • large-file mode status.

Local Sample

Example measured on the maintainer's local macOS development machine on 2026-04-29.

Fixture Bytes Duration Stages Preview nodes Complete nodes Deferred branches Large-file mode
1 MiB 1,062,923 44.4 ms preview -> complete 4 57 51 yes
5 MiB 5,331,280 40.6 ms preview -> complete 4 57 51 yes
20 MiB 21,386,297 91.1 ms preview -> complete 4 57 51 yes

These numbers are hardware-dependent. Treat them as a local sanity check, not a public performance guarantee.