Kconfig reference

CONFIG_ZXSD enables the descriptor-driven runtime. Schemas are compiled on the host and are never parsed on the target.

Resource limits
Option	Default	Meaning
`CONFIG_ZXSD_MAX_INPUT`	1048576	Maximum bytes in a NUL-terminated document.
`CONFIG_ZXSD_MAX_DEPTH`	32	Maximum nested elements; affects parser and streaming stack storage.
`CONFIG_ZXSD_MAX_PARTICLES`	32	Maximum direct-term counters or interned history particles per frame.
`CONFIG_ZXSD_STREAM_CHILD_CAPACITY`	64	Child tokens retained by models requiring history; incremental models do not consume this capacity.
`CONFIG_ZXSD_STREAM_MODEL_CAPACITY`	128	Shared continuation/frontier slots for nested models. Zero removes the pool and selects bounded child history for non-flat models.
`CONFIG_ZXSD_STREAM_NAME_CAPACITY`	128	Bytes for a local name, namespace URI, or prefix, including the NUL.
`CONFIG_ZXSD_MAX_ATTRIBUTES`	32	Maximum non-namespace attributes and descriptors per type.
`CONFIG_ZXSD_MAX_NAMESPACES`	16	Maximum simultaneously active namespace bindings.
`CONFIG_ZXSD_MAX_IDENTITY_VALUES`	64	Maximum key, unique, or keyref tuples retained per constraint.
`CONFIG_ZXSD_MAX_TEXT_LENGTH`	1024	Maximum decoded lexical value and per-frame streaming text buffer.
`CONFIG_ZXSD_XML_TOKEN_CAPACITY`	4096	Maximum incomplete raw-XML token retained across input fragments.
`CONFIG_ZXSD_MAX_PATTERN_LENGTH`	256	Maximum UTF-8 byte length of one XSD regular expression.
`CONFIG_ZXSD_MAX_REGEX_NODES`	512	Fixed parser capacity and principal pattern-validation stack cost.
`CONFIG_ZXSD_MAX_REGEX_DEPTH`	32	Maximum group and character-class subtraction nesting.
`CONFIG_ZXSD_MAX_SCHEMA_COMPLEXITY`	4096	Maximum aggregate runtime-visible descriptor count.

CONFIG_ZXSD_NO_HEAP defaults to enabled. Schemas without identity constraints remain fully usable. A schema requiring identity tuple storage returns ZXSD_ERR_RESOURCE_LIMIT unless validation is invoked with an application allocator.

CONFIG_ZXSD_BINDINGS builds the reader and writer for structures generated with zxsdc --bindings; see Typed bindings. It is disabled by default and adds no RAM: reading uses a caller-provided arena and writing streams to a caller-provided sink.

Choose limits from the largest trusted schema and document the device accepts, then exercise the boundaries with tests. See Resource measurements for the relationship between limits and memory.

Generate schema recommendations

Run the host compiler before configuring Zephyr:

$ python3 zxsdc device.xsd -o generated --zephyr-config
$ west build -b qemu_cortex_a53 path/to/application -- \
    -DEXTRA_CONF_FILE=/absolute/path/generated/device.conf

The option retains the ordinary validator output and additionally writes device.conf and device.limits.h. The generated validator source includes the limits header. It emits a compile-time error when active ZXSD_* macros are below the known schema requirements. On Zephyr, those macros are defined from the corresponding CONFIG_ZXSD_* values for the library and application.

The fragment contains active assignments for descriptor sizes: particles, attributes, runtime schema complexity, name storage, regex AST nodes, pattern UTF-8 bytes, and regex nesting. It also supplies an element-depth envelope when the compiled graph is finite, accounting for possible dynamic derived types and descriptor-group nesting. All emitted types and roots contribute, including unused named declarations. A requirement exceeding a Kconfig range fails generation explicitly rather than being clamped.

The child-history recommendation accounts for incremental model selection. Flat models and nested sequences/choices can use history capacity 1 even with unbounded repetitions. Flat-only profiles disable the shared model pool to avoid reserving unused RAM. Other profiles require it to be enabled and leave its size as a commented application budget (default 128 slots). Size it for simultaneously active nested models and alternative occurrence partitions, then test representative documents and exhaustion. The pool is shared across XML depths, not reserved per frame. Its per-event epsilon-work budget is the square of its capacity. Exhaustion returns ZXSD_ERR_STREAM_LIMIT.

Commented assignments are budgets to review and uncomment as appropriate:

XML input bytes, lexical text bytes, namespace bindings, and raw token bytes cannot be bounded from XSD alone. Comments, arbitrary prefixes, unused namespace declarations, entity spelling, and lexical whitespace remain instance choices. A maxLength facet counts characters, list items, or decoded binary bytes; it does not bound raw XML or every lexical spelling.
Recursive type graphs, recursion introduced through xsi:type, and wildcard/anyType subtrees require an application depth limit.
Identity tuple counts need an input budget and caller-owned scratch storage in zero-heap mode. Schema-defined field/path sizes without a Kconfig option are reported as C compile definitions and checked by the generated header.
Attribute recommendations cover declarations. Add headroom for accepted xsi:* attributes and wildcard attributes. Name storage includes schema names/URIs, standard XML/XSD/xsi URIs, and NUL, but extra-long input prefixes can require a larger value.

The fragment is a sizing aid for the documented runtime limits, not a promise to accept every XML spelling that satisfies the XSD. Stack sizes, caller-owned arenas, HTTP buffers, TLS, and networking remain application configuration.

For multiple top-level schemas, generate one combined profile:

$ python3 zxsdc samples/zephyr/weather/nominatim.xsd \
    samples/zephyr/weather/met_weather.xsd -o generated --zephyr-config

This writes generated/nominatim.conf using the combined model: currently 45 particles, 14 attributes, depth 6, history capacity 1, and no model pool. Explicit local -I search directories and includes/imports work as in normal generation. If separate profiles are merged manually, take the component-wise maximum (including enabling/sizing the model pool if any profile needs it); successive Zephyr fragments otherwise overwrite earlier assignments.

Build-time generation can retain the checks with:

zxsd_generate(TARGET app SCHEMA device.xsd ZEPHYR_CONFIG)

This helper runs after Zephyr’s Kconfig evaluation. It generates the fragment as a build artifact but does not apply it to the current configuration. Regenerate the reviewed profile before reconfiguring when schemas change; the compile-time checks catch known settings that have become insufficient. Add BINDINGS to both invocations when typed bindings are needed; the fragment then also enables CONFIG_ZXSD_BINDINGS.

Pattern parsing is allocation-free and uses a fixed automatic workspace. With the default 512-node capacity, reserve approximately 11 KiB in the validating thread in addition to the normal validation stack. Lower CONFIG_ZXSD_MAX_REGEX_NODES when the compiled schemas use smaller patterns; the samples/zephyr validation and bindings samples copy their schema-derived values, including this capacity, into prj.conf.