Preserving Comments

Comments are formatter foes. AST drops them. CST keeps them.

Strategies:

Position-based:

• Each comment has a position (line, col).
• Associate with the AST node before/after based on lines.

Trailing vs leading:

• Leading comment: # comment then node.
• Trailing comment: node # comment.

When formatting:

Edge cases:

• Block comments inside expressions: hard. Some formatters punt and keep verbatim.
• Multi-line comments: indent properly.
• Comments separating logical groups: detect blank-line patterns.
• Disabling: // prettier-ignore skip the next node.

Doc strings (Python) and docstrings (other langs):

• Often preserved exactly (don't reflow).
• Some formatters detect and skip.

Embedded code in comments (jsdoc):

• Don't touch interior.

Best practice: use a CST-aware library. Don't roll your own comment association.

Real CSTs:

• libCST (Python): Python-specific, comment-aware.
• ts-morph (TypeScript): wraps TypeScript AST + lossless source.
• tree-sitter: language-agnostic.