It's an argument against preprocessors for programming languages.

Tree-sitter's error handling is constrained by its intended use in editors, so incrementality and efficiency are important. For diffing/merging, a more elaborate parsing algorithm might be better, for example one that uses an Earley/CYK-like algorithm but attempts to minimize some error term (which a dynamic programming algorithm could be naturally extended to.)

Interesting idea. Tree-sitter's trade-off (speed + incrementality over completeness) makes sense for editors but you're right that for merge/diff a more thorough parser could be worth the cost since it's a cold path, not real-time. We only parse three file versions at merge time so spending an extra 50ms on a better parse would be fine. Worth exploring, thanks for the pointer.