A location-keyed trivia table (built by associate).

An empty trivia table (for a printing pass that carries no comments — e.g. a binary-input conversion, or the dry pass that only populates collect).

An opaque set of source locations — the collect/seen tables. It keys on the start/end byte offsets only (cheap to hash and compare, no filename-string traversal), which the location-keyed lookup on every printed atom used to be dominated by. A caller only ever fills the one associate hands its collect function; all lookups happen inside Trivia.

mark set loc records loc in set. Lets associate's collect build its set by walking the document it is about to print, instead of driving a discarded dry print pass whose only effect is the same get-time collect.

associate ~collect ctx associates trivia to locations. The second component holds the leftover comments that no location owns (trailing comments, or every comment when there are no locations); the caller prints them as tail trivia.

collect records the locations the printer will actually look up (see get) into a set of this function's making — it is the language's Output.collect (a walk of the laid-out document, or a dry printing pass). The association covers the spans in that set that are also parse nodes: a comment would otherwise either attach to a node the printer skips (and be lost) or to a looked-up span that is no source construct at all — a conversion also stamps output nodes with token spans and with Ast.dummy_loc.

Create a new trivia context.

report_item ctx kind location content reports a comment or an annotation.

report_newline ctx reports a newline.

report_token ctx pos records that a meaningful token ending at byte pos has been encountered on the current line.

record_pos ctx loc records that a node spans loc, without building the wrapper. For a node whose record is not an Ast.annotated — an instruction, which carries hints besides its location — and so cannot go through with_pos.

with_pos ctx loc v wraps v with location loc, recording the span as record_pos does.

drop_in_ranges ctx ranges removes every comment whose anchor falls within one of the half-open byte ranges [start, end). Used after conditional specialization splices out a branch: the comments inside the removed source span are discarded rather than re-attaching to a surviving neighbour. The ranges and comments are each sorted once and swept together in a single pass.

The empty association (before, within and after all empty).

get trivia ~seen loc returns the trivia associated with loc, with de-duplication: it returns dummy_assoc for None, a missing location, or a location already present in seen; on the first real hit it records the location in seen and returns its association. De-duplication is a no-op for formatters (each parser location occurs once) and prevents a comment from being emitted repeatedly when conversion replicates one source location onto several output nodes.

Drop blank-line entries at the end of the list, so emitted tail trivia does not leave spurious blank lines at the end of the file.

retarget ~src ~dst ctx rewrites the delimiters of every comment collected in ctx from the src syntax to the dst syntax (line-comment prefix and block-comment delimiters), leaving blank lines untouched. Block delimiters are balanced in stored content, so a global swap preserves nesting.

An annotation is dropped: it is kept as verbatim source text, which only the WebAssembly text format can spell, so a conversion to Wax loses it rather than pasting (@…) into a Wax file.

Called by a cross-format conversion, on the context whose trivia will be replayed onto the converted AST — the rewrite belongs to the translation, not to printing. It touches only comment text, never an anchor or a kind, so it commutes with associate.