Library and command line tool for converting XML files to json


LTS Haskell 18.28:2.0.1
Stackage Nightly 2022-02-07:2.0.1
Latest on Hackage:2.0.1

See all snapshots xml-to-json appears in

MIT licensed by Noam Lewis
Maintained by [email protected]
This version can be pinned in stack with:xml-to-json-2.0.1@sha256:cdb650659931c9a8ebbdc6cb6a77a5b8337af964be7483f9e88ba67cbe59f87a,2166

Module documentation for 2.0.1


Fast & easy library & command line tool for converting XML files to JSON.

Heads up! The project has been split into two projects. See xml-to-json-fast for a low-memory-usage version that has less features.



xml-to-json converts xml to json. It includes a Haskell library and a command-line tool.

xml-to-json ships with two different executables:

  1. xml-to-json-fast (“fast”) uses a lot less memory, but you can’t control the output. Can be used on XML files of any size.
  2. xml-to-json (“classic”) provides some control over json output format, but uses a lot of memory. Suitable for smaller files.

“Fast” xml-to-json-fast

The so-called “fast” version (which uses a lot less memory) has been forked into a separate project, xml-to-json-fast.

“Classic” xml-to-json

The fully featured “classic” xml-to-json provides compact json output that’s designed to be easy to store and process using JSON-based databases, such as mongoDB or CouchDB. In fact, the original motivation for xml-to-json was to store and query a large (~10GB) XML-based dataset, using an off-the-shelf scalable JSON database.

When using “classic” xml-to-json, the input XML must be valid.

Currently the xml-to-json processes XMLs according to lossy rules designed to produce sensibly minimal output. If you need to convert without losing information at all consider something like the XSLT offered by the jsonml project. Unlike jsonml, this tool - xml-to-json - produces json output similar (but not identical) to the xml2json-xslt project.

Implementation Notes

xml-to-json is implemented in Haskell.

As of this writing, xml-to-json uses hxt with the expat-based hxt-expat parser. The pure Haskell parsers for hxt all seem to have memory issues which hxt-expat doesn’t.


Note for Windows users: Only local files, not URLs, are supported as command line arguments. This is because curl doesn’t compile on my (windows + cygwin) machine out-of-the-box.

To install the release version: Since xml-to-json is implemented in Haskell, “all you need to do” is install the latest Haskell platform for your system, and then run:

cabal update
cabal install xml-to-json

To install from source: Clone this repository locally, and then (assuming you have Haskell platform installed) run cabal install:

cd xml-to-json
cabal install


Basic usage

Just run the tool with the filename as a single argument, and direct the stdout to a file or a pipe:

xml-to-json myfile.xml > myfile.js

Classic xml-to-json: Advanced Usage

Use the --help option to see the full command line options.

Here’s a (possibly outdated) snapshot of the --help output:

Usage: <program> [OPTION...] files...
  -h      --help                      Show this help
  -t TAG  --tag-name=TAG              Start conversion with nodes named TAG (ignoring all parent nodes)
  -s      --skip-roots                Ignore the selected nodes, and start converting from their children
                                      (can be combined with the 'start-tag' option to process only children of the matching nodes)
  -a      --as-array                  Output the resulting objects in a top-level JSON array
  -m      --multiline                 When using 'as-array' output, print each of top-level json object on a seperate line.
                                      (If not using 'as-array', this option will be on regardless, and output is always line-seperated.)
          --no-collapse-text=PATTERN  For elements with tag matching regex PATTERN only:
	  			      Don't collapse elements that only contain text into a simple string property.
                                      Instead, always emit '.value' properties for text nodes, even if an element contains only text.
                                      (Output 'schema' will be more stable.)
          --no-ignore-nulls           Don't ignore nulls (and do output them) in the top level of output objects

Example output

Input file:

<?xml version="1.0"?>
  <Test Name="The First Test">
    <SomeText>Some simple text</SomeText>
    <SomeMoreText>More text</SomeMoreText>
    <Description Format="FooFormat">
Just a dummy
<!-- comment -->
Xml file.
  <Test Name="Second"/>

JSON output using default settings:

{"Tests":{"Test":[{"Name":"The First Test","SomeText":"Some simple text","Description":{"Format":"FooFormat","value":"Just a dummy\n\nXml file."}},{"Name":"Second"}]}}

Formatted for readability (not the actual output):

            "Name":"The First Test",
	    "SomeMoreText":"More text",
            "SomeText":"Some simple text",
               "value":"Just a dummy\n\nXml file."

Note that currently xml-to-json does not retain the order of elements / attributes.

Using the various options you can control various aspects of the output such as:

  • At which top-level nodes the conversion starts to work (-t)
  • Whether to wrap the output in a top-level JSON array,
  • Whether or not to collapse simple string elements, such as the element in the example, into a simple string property (you can specify a regular expression pattern to decide which nodes to collapse)
  • And more…

Use the --help option to see a full list of options.

Performance of “classic” xml-to-json

The “classic” xml-to-json cannot operate on large files. However, it is fast when operating on multiple small files. For large XML files, the speed on a core-i5 machine is about 2MB of xml / sec, with a 100MB XML file resulting in a 56MB json output. It took about 10 minutes to process 1GB of xml data. The main performance limit is memory - only one single-threaded process was running since every single large file (tens of megabytes) consumes a lot of memory - about 50 times the size of the file.

A few simple tests have shown this to be at least twice as fast as jsonml’s xlst-based converter (however, the outputs are not similar, as stated above).

Currently the program processes files serially. If run in parallel on many small XML files (<5MB) the performance becomes cpu-bound and processing may be much faster, depending on the architecture. A simple test showed performance of about 5MB / sec (on the same core-i5).