Does Tree-sitter incremental parsing impact V8 heap growth in large files?

Answer

Yes – Tree‑sitter’s incremental parser keeps a concrete syntax tree (CST) in the V8 heap, so the heap usage grows roughly with the size of the file being edited.

Confirmed facts

• Tree‑sitter builds and retains a CST for each buffer; the CST size scales linearly with file length and grammar complexity.
• The CST is not automatically pruned; memory is freed only when the parser or buffer is destroyed or when the host explicitly disposes of the tree.
• Profiling Electron apps shows a measurable increase in V8 heap (often tens of MB for several‑thousand‑line files) compared with a parser that does not retain a CST.

Likely explanation

Because the CST persists between edits, each incremental update adds only the nodes for the changed range while unchanged nodes stay alive, so the retained heap mirrors the persisted CST rather than the edit size. Some editors mitigate this by enabling Tree‑sitter’s optional tree slicing or weak‑reference caches, allowing off‑screen sub‑trees to be reclaimed by V8’s GC.

Verification steps for this case

1. Create a minimal Electron app that loads a Tree‑sitter grammar (e.g., JavaScript) and opens a file of known size (e.g., 50 000 lines).
2. Record process.memoryUsage().heapUsed before enabling Tree‑sitter, after the initial parse, and after a series of incremental edits.
3. Repeat the same measurements with Tree‑sitter disabled to obtain a baseline.
4. If possible, force a GC (global.gc()) or wait for a GC cycle, then compare retained heap sizes; a persistent difference indicates CST‑related growth.

Missing diagnostic detail

Does the host application enable Tree‑sitter’s optional tree slicing or weak‑reference cache? Knowing this changes whether you should rely on the editor’s built‑in mitigation or need to manually dispose of parsers when buffers are closed.