XML to CSV Converter

XML is the older of the two big data formats and still the one many systems speak: RSS and Atom feeds, sitemaps, SOAP web-service responses, exports from ERP and accounting systems, government open-data releases, product feeds for marketplaces, Office and LibreOffice files under the hood, OFX 2 bank statements, configuration files such as Maven's pom.xml. The shape is always the same: one root element and, somewhere inside it, an element that repeats once per record, with the values in child elements, in attributes, or both.

That repeating element is what a table needs, and finding it is the only real decision in the conversion. The converter counts how often each element name occurs under one parent, prefers elements that have children or attributes over plain values, and among those the one closest to the root: in an RSS feed that is item rather than category, in a catalog product rather than tag. The select shows every element that repeats and how many rows it would give, plus the root for a one-row table.

Every value in a row gets a column named by its path from the row element. Take one product from a catalog with attributes, a nested dimensions element and two tag elements:

• Attributes are marked with @ so a column never collides with a child element of the same name; turn "Attributes as columns" off to leave them out entirely.
• A child element that repeats anywhere in the document is indexed in every row, so the header reads tag[0] for a product with one tag as well; the record with the most repeats sets the number of columns.
• Columns are the union of what all rows contain, in the order they first appear. A row missing an element gets an empty cell; an empty element such as <note/> gives an empty cell too.
• Values are taken exactly as written: 007 stays 007, 1e3 stays 1e3, true stays the word true. CSV has no types, so nothing is converted.

CSV is a grid of text. Everything that XML can express beyond that is flattened or dropped, and it is worth knowing which is which. Nesting becomes dot paths in the header, attributes keep the @ marker, and that is the whole record of the document's structure. Comments, processing instructions and the XML declaration are dropped. Entities are decoded, so &amp; comes out as &, and CDATA sections become plain text. Whitespace around values is trimmed.

Mixed content, text interleaved with child elements as in <p>Hello <b>world</b>!</p>, does not survive: the text pieces are joined into one cell and their position relative to the child elements is lost. Namespace prefixes stay part of the names (dc:creator), but namespace declarations are treated as ordinary attributes. Element order within a record becomes column order, which is fine for data files and meaningless for documents.

• The file is UTF-8 without a byte order mark, with CRLF line endings as RFC 4180 describes and with double quotes around any field that contains the delimiter, a quote or a line break (inner quotes doubled). Most tools read that as is.
• Excel on Windows assumes the system code page when you double-click a CSV, so ä, é or 中 can come out garbled. Use Data › From Text/CSV, choose 65001: Unicode (UTF-8), and while you are there set columns like SKU or ZIP to Text so 00123 keeps its zeros and 1e3 is not turned into 1000.
• In countries where the comma is the decimal separator (Germany, France, Spain, Italy and others) Excel expects a semicolon between fields; choose Semicolon here or the whole row lands in one cell.
• Google Sheets (File › Import) detects the delimiter and reads UTF-8; Numbers and LibreOffice Calc open the file directly and ask about the delimiter.
• Tab-separated output pastes straight into an open sheet: copy the text, click a cell, paste.