XML to CSV: you choose what a row is

The one that repeats at the level you care about — usually the child of the root, like <item> inside <inventory> or <entry> inside <feed>. The dropdown ranks every path in your document by how often it occurs and shows how many leaf columns it would produce, so the choice is a fact about your file rather than a guess. Choosing an element that occurs once gives a single very wide row, which is a useful signal that you picked a container rather than a record.

By their path from the record element, joined with a separator: <stock><units>420</units></stock> inside a record becomes a column named stock/units. Depth is not limited, so a three-level structure produces headers like shipping/address/postcode, and attributes hang off the same path with an @ — price/@currency. This keeps distinct values in distinct columns, which is the point of flattening; a converter that emits stock as one blob has not flattened anything.

You choose between three answers, and the numbering is decided across the whole file rather than per record. Numbered columns give tag[1] and tag[2]; joining puts every value in one cell separated by a string you pick; keeping the first discards the rest and counts what it dropped. The document-wide rule matters: if one record has two tags and the next has one, the single tag still lands in tag[1] rather than in a column called tag, so the column keeps meaning the same thing all the way down.

The header is the union of every path seen, in the order they first appear, and records missing a path get the placeholder. XML has no obligation to make sibling elements identical — an optional <discontinued> appears on two products out of 300 — while a table has to be rectangular, so the columns are as wide as the widest record. The report names every sparse column, which is the quickest way to spot an optional field you did not know the feed had.

Yes, as columns marked with @, and you can switch them off in one click. Attributes usually carry the identifiers — sku, id, currency, status — so dropping them is the fastest way to lose the key you would join on later. They are flattened by the same path rule as elements, so an attribute on a nested child becomes price/@currency rather than colliding with a top-level one.

A field is quoted when it contains the delimiter, a double quote, a CR, an LF, or a leading or trailing space, and any quote inside it is doubled — that is RFC 4180 section 2, and it is why a product description with a comma does not shift every column to its right. CRLF is the line ending the same memo specifies and what Excel expects on Windows; LF is the option for anything Unix-shaped downstream. Long descriptions with real line breaks inside them stay one record, quoted.

Not as one table, and no converter should pretend otherwise. A document holding <customer> and <invoice> elements describes two entities with different columns, and forcing them into one sheet gives you a header with every field of both and half the cells empty. Convert once per record type — change the dropdown, export again — which is the same thing you would do with two database tables.