Large files

Choose a loading path based on whether the user needs editing and language analysis. You need representative files and a production browser build to measure your integration.

1. Decide which work the file needs

The core stores a fully loaded editable document. A file that fits memory may still cost too much to parse, tokenize or analyze on each edit. A plain core editor has no language analysis until you attach providers or plugins. Singapore's tree-sitter and TypeScript plugins have no automatic document-size cutoff.

Choose and enforce an analysis policy in your host application. You can keep large documents editable while leaving expensive analysis plugins detached. Fregat's host defaults to a 10 Mi UTF-16 code-unit cutoff for syntax and language-server analysis. That cutoff belongs to Fregat's settings and host policy.

The paged package provides a read-only path for files you want to inspect without loading the whole file into the editable buffer. Editable paging remains planned work.

2. Measure the editable path

Load your file in a plain core editor first. Measure open time, typing, scroll and retained memory. Add syntax, language servers, minimap and host subscriptions one at a time to find the cost your application adds.

A code-unit count differs from file bytes. Emoji occupy two UTF-16 units, and UTF-8 byte counts depend on the characters. Include both when reporting results.

3. Use the paged viewer when it fits the task

Consult the generated @singapore-editor/paged reference for its file source and loading APIs. Keep the file source alive for the view lifetime. Verify decoding, navigation and cleanup with your actual file format.

Result and limits

The editable core and read-only paged viewer solve different tasks. A size ceiling says that a test file opened; it says nothing about whether language analysis or typing met an interactive budget. Existing measurements show large-file typing can miss a 60 Hz frame, and analysis can add substantial latency.

There is no supported universal 1 GB editable-file claim. Public editor-level comparisons with Monaco and CodeMirror are still missing. The research report records the measured ceilings and their limitations.

If it doesn't work

Typing stalls after adding language features

Check the cutoff your host enforces, and measure each plugin's work separately. Detach expensive analysis above the file size your measurements support. For log inspection, the read-only paged view may fit better.

The file opens but memory keeps growing

Check document, worker and file-source disposal. Compare retained memory after closing the view against a plain-text baseline.

Continue with performance and workers.