DataToolsLab
XML Tools

XML Split

Split a document into separate files, one per repeating child element under a chosen parent.

Loading tool…

What is XML Split?

The opposite of merging: a document holding many records becomes a file per record. This is what turns a nightly batch export into something a queue can consume one message at a time.

How it works

The repeating element is detected or specified by path. Each matched element becomes its own document: the original root — and any ancestors between it and the matched element — are rebuilt around it, so every part keeps its structure, its namespace declarations and any sibling fields that explain the batch. Names come from a key attribute when you name one, otherwise from a numbered prefix. A path that matches nothing is reported rather than producing empty files.

  1. Paste the document. Drop in XML, a schema, a feed or an XPath expression. Everything is parsed in the page: no upload, no server round trip, and the tools keep working with the network switched off.
  2. Set the options that match your document. Indent unit, whether to keep comments, which dialect, which output method. The defaults are the safe ones — nothing that changes meaning is enabled for you.
  3. Read the result, then copy or download it. Results appear as you type, copy straight to the clipboard, and download with a sensible filename (.xml, .xsd, .csv). Reset returns every field to its default.

Examples

A minified document from an API response

Paste one line of XML and the Formatter indents it, wraps long start tags one attribute per line and normalises empty elements — without touching text inside mixed content.

An XPath that works in Chrome and not in your servlet

The XPath Tester evaluates against a real tree and reports the axis or function it cannot support, instead of returning zero nodes and letting you guess why.

A document that will not parse

The Validator gives line and column for every mismatched tag, stray ampersand and duplicate attribute, with the reason in words: which open element was expected, which was found.

A schema that only half matches your data

The XSD Validator marks each violation with the path of the offending node, and the Schema Generator goes the other way — inferring a starting-point XSD from real documents.

Common mistakes

Assuming a pretty-printer cannot change meaning

Whitespace between elements is often insignificant, but inside mixed content it is not: `<p>Hello <b>world</b></p>` must stay on one line. Formatting here re-indents element-only content and leaves mixed content alone.

Treating a validation pass as proof of interoperability

The DTD and XSD validators implement honest subsets. A document reported valid here can still be rejected by a full validator that understands identity constraints, substitution groups or XSD 1.1 assertions — each page lists what it does not cover.

Reading the whole document because the parser did

A DOCTYPE is not inert. If your parser resolves external entities, pasting an untrusted document is enough to read a file or call an internal URL. The XXE Risk Checker shows what a document asks for; the fix is in the parser configuration.

Forgetting that attributes and elements are not interchangeable

`<id>1</id>` and `id="1"` mean different things to every schema and every XPath. When converting to CSV or YAML the distinction is preserved (`@id`), because flattening it away is the change that costs an afternoon later.

Frequently asked questions

Why do parts keep the original root element?

Because a bare `<shipment>` element with no root is not a well-formed document and most consumers will reject it. Rebuilding the ancestors means each part can be handed to the same code that processed the original — switch to bare mode if you really only want the element.

How are the file names chosen?

From the attribute you name — `SHP-1001` gives `shipment-1-SHP-1001.xml`, so a record can be found without opening anything. Without one, parts are numbered. Duplicate key values are reported, because two files with the same name is a silent data loss.

Is there a limit on how many parts?

Only the browser's memory, since every part is generated in the page. A few thousand parts is comfortable; beyond that, split with a command-line tool and keep the key-attribute naming convention, which is the part worth copying.

Online