Performance SLOs
Status
- Type: Normative
- Stability: Draft
Purpose
To be industry-standard, FerroTeX must meet responsiveness expectations.
This document defines target Service Level Objectives (SLOs).
Latency Targets (interactive)
On representative medium projects:
- completion p95: <= 100 ms
- hover p95: <= 100 ms
- definition/references p95: <= 200 ms
- semantic tokens full p95: <= 300 ms (with caching)
These targets are advisory until implementation benchmarks exist.
The “Speed” Aspiration
To achieve “Incremental Everything,” FerroTeX aims for:
- 50ms re-analysis for documents up to 500 pages.
This ensures that keystroke-level diagnostic feedback is indistinguishable from local editing latency, even in large monographs or theses.
Parsing and Indexing
- incremental parse update: proportional to change size
- index update on save: bounded by file size + dependency fanout
Memory
- avoid duplicating large document strings
- bounded caches with eviction policies
Cancellation
All long-running LSP requests MUST be cancellable.
Benchmarking
Benchmarks must be recorded as per docs/development/benchmarks.md.