zxsd — XSD compiler and bounded C validator

zxsd compiles a supported XML Schema (XSD) subset into immutable C descriptors. The target application links those descriptors with a portable C99 runtime to validate XML without parsing the schema at runtime.

• zxsdc runs on the build host and generates deterministic C source and headers.
• libzxsd validates complete documents, raw byte fragments, or parser events.
• Optional generated bindings read and write typed C structures using caller-owned storage.
• Resource limits are fixed at compile time; zero-heap builds are supported.

Python, filesystem access, networking, and an XSD parser are never required on the target.

Status: draft/experimental. zxsd implements a documented XSD 1.0 subset, not the complete specification. Unsupported schema constructs fail explicitly. During 0.x, upgrades require rebuilding the runtime and generated code.

Documentation · XSD support matrix · Getting started · Typed bindings · Zephyr · Compatibility

Quick start

Generate a schema-specific validator:

python3 zxsdc tests/device.xsd -o generated

The generated header exposes a validation entry point:

#include "device.zxsd.h"

zxsd_result_t result = device_validate(xml);
if (result.code != ZXSD_OK) {
    /* result.message, result.offset and result.path describe the error. */
}

The installed CMake package can generate and compile the validator as part of the application build:

find_package(zxsd CONFIG REQUIRED)

add_executable(application main.c)
zxsd_generate(TARGET application SCHEMA device.xsd)
target_link_libraries(application PRIVATE zxsd::zxsd)

Add BINDINGS to zxsd_generate() to generate typed C structures and read/write functions. The host compiler is also available as python -m pip install zxsdc. See Getting started for source builds, installation, and non-CMake integration.

What it validates

The supported subset includes:

• simple and complex types, derivation, lists, unions, and facets;
• sequences, choices, all, groups, occurrences, and wildcards;
• namespaces, schema composition, substitution groups, and recursive types;
• xsi:nil, xsi:type, defaults, fixed values, and abstract declarations;
• unique, key, keyref, and the restricted XSD identity XPath language;
• a bounded XSD regular-expression subset with Unicode categories and blocks.

The exact contract, including partial and unsupported features, is maintained in SUPPORT.md. Compiler acceptance alone is not treated as conformance.

Runtime modes

• Document validation: validate a NUL-terminated XML document with one call.
• Incremental XML: feed arbitrary byte fragments from a socket, file, or HTTP client.
• Event validation: send expanded start, attribute, text, and end events from another parser.
• Typed bindings: validate and materialize selected schemas into generated C structures.
• Streaming bindings: receive selected typed subtrees while validating the document once.

All modes use the same generated schema descriptors. Inputs that exceed a configured depth, text, identity, model, or input limit fail with a structured error instead of growing memory without a bound.

Embedded use

The runtime is portable C99 and is tested on Linux, macOS, Windows, and Zephyr. ZXSD_NO_HEAP=ON removes standard-library allocation; schemas with identity constraints can use an explicit application allocator. zxsdc --zephyr-config generates Kconfig recommendations and compile-time limit checks from the schema.

Zephyr uses the allocation-free XML backend by default. Mini-XML 4 is an optional adapter for validating an existing Mini-XML tree, on the host or on Zephyr when a Mini-XML module provides CONFIG_MXML.

See the host sample, Zephyr samples, and resource-usage guide for complete integrations.

Conformance

Project-owned positive and negative tests gate every documented feature. An optional W3C XSD 1.0 harness inventories the official suite while keeping passed, failed, unsupported, expected-failure, and skipped results distinct.

See Conformance and test coverage.

zxsd is licensed under Apache-2.0.

Assistance

Development assisted by GPT-5.6 Sol.