[ddehilster]

What

A new Help-view page, Large Files (Help/markdown/vscode/largefiles.md), linked from Help home. It covers the two kinds of large file an analyzer meets.

Large inputs

• Why size matters: the engine builds one parse tree per input. Measured memory and time per MB, and the 2 GB input limit.
• Zone mode (zone / endzone in analyzer.seq, engine 4.2.47): skim the document, read it one zone at a time, then wrap up.
  • A worked one-zone-per-form example for a tax return package: skim passes that build _form zones from Form … / SCHEDULE … header lines, a per-form pass, and an endzone summary.
  • How to write passes for zone mode: @CODE runs per zone, whole-document work goes after endzone, and what -DEV writes.
• Splitting a file into pieces streamed through -STDIN.
• The -TIMELIMIT and rec pass limits.

Large dictionaries and KBs

• A short summary of lazy loading: …full.dict / …full.kbb, tokenizer and any-pass lookups, loadkbb / loaddict, and compiled analyzers. It points to the existing Lazy Loading page for the detail.

Checked

• The example's four pass files were copied out of the page and run exactly as written on engine 4.2.47, against a made-up three-form package. They produce the expected per-form lines and the package total.
• check-help-docs.ps1: no broken links or images. The one orphaned page it reports, New_in_Current_Version, was already orphaned on main.

Follow-up

To list the page in the sidebar Help view next to "Lazy Loading", vscode-nlp/src/helpView.ts needs one entry: { label: 'Large Files', page: 'vscode/largefiles', icon: 'file-binary' }. Until then the page is reachable from Help home.

🤖 Generated with Claude Code