[smoltafet]

Fixes #422.

Updated after @wdavidw's feedback, twice over: from_line keeps counting physical lines by default (no breaking change), and the comment-skipping behavior now lives on from_line itself instead of a separate option. from_line takes two forms:

parse(input, { comment: '//', from_line: 3 })
// counts physical lines, as before

parse(input, { comment: '//', from_line: { position: 3, skip_comments: true } })
// starts emitting at the 3rd non-comment line, however many
// comment lines precede or surround it

An integer is normalized to an object with the integer as position and skip_comments set to false; in an object, skip_comments is optional and defaults to false.

Implementation notes:

• The from_line activation check subtracts the number of full comment lines seen so far only when skip_comments is true. That count is taken when a comment starts (new state.fullCommentLines), not at the record delimiter: while emission is disabled by from_line, __onField short-circuits and the record/field state at the delimiter no longer reflects the line's content, which would misclassify lines with trailing comments as comment lines.
• A line is a full comment line when nothing but the comment was seen on it: no field characters were appended and no quoted field was opened (tracked as state.lineHasContent). A trailing comment after content does not make the line a comment line. info.comment_lines and error line numbers still use physical lines, unchanged.
• Normalization validates the object form (position must be a positive integer, skip_comments a boolean); the normalized option is typed in options.d.ts and the API types test covers both forms.
• This also fixes a pre-existing state leak in the same code path: when the from_line transition (or the skip_empty_lines skip) fired on a line ending with a trailing comment, the early continue skipped the state.commenting = false reset, so the whole next line was silently swallowed as a comment. On master, parse('a,b # comment\n1,2\n3,4', { comment: '#', from_line: 2 }) returns only ['3','4'] and reports comment_lines: 1; it now returns both records and comment_lines: 0. Both branches now reset commenting before continuing.
• to_line is intentionally left counting physical lines, as in the issue's request; happy to align it too if you prefer.

Tests: eight new tests in packages/csv-parse/test/option.from_line.js covering the object form (leading comment lines using the issue's example, comments between records, boundary positioning, trailing comments, columns: true after comments, validation of the object form, skip_comments unset, and a default-behavior test showing comment lines are counted for a plain integer). Full csv-parse suite: 637 passing, 2 pending (pristine master: 629 passing, 2 pending); tsc --noEmit and eslint are clean. Default behavior was also compared against pristine master on the distinguishing cases and is unchanged.

[smoltafet]

Hi @wdavidw, could you take a look at this when you have a moment? Happy to fix up anything that needs changing. Thanks!

[wdavidw]

Hi @smoltafet, I have been receiving a flood of PR recently, alimented by AI, and I don't have the time to review them. I quickly looked at it recently. If I understand your proposition correctly, counting lines, even those containing comments, is on purpose. The idea is that if someone open a document and found the CVS records start in line 200, then parsing shall start there, whichever comes before. Another argument is that the PR introduces a breaking change. What could be done is the introduction of a new option, something like from_line_skip_comments.

[smoltafet]

@wdavidw thanks, that makes sense. I reworked it so from_line counts physical lines by default like before, and the comment skipping is now opt-in through a new from_line_skip_comments option. Tests pass locally (full csv-parse suite, 637 passing). Happy to rename the option if you prefer something else.

[wdavidw]

I am currently thinking that from_line could take 2 forms:

• an object, with 2 properties eg position: number and skip_comments: boolean
• an integer, normalized to an object with the integer being the position number and skip_comments set to false

what do you think ?

[smoltafet]

@wdavidw good call, I went with the object form. from_line now takes {position, skip_comments}, and an integer gets normalized to that shape with skip_comments: false, so the separate option from the last push is gone. Tests pass locally (full csv-parse suite 637 passing, tsc and eslint clean).