onnx_verify.h#

Schema-free structural validation for onnx_proto messages.

The functions declared here validate that a protobuf is internally consistent on its own terms: required fields are set, names are unique, the graph is in single static assignment (SSA) form and topologically sorted, tensor payload sizes match the declared shape/dtype, and so on. None of these checks needs an operator-schema registry: nothing here depends on the semantics of a specific op_type.

For schema-aware validation (operator input/output arity, attribute constraints, type/shape inference), use onnx_lib::checker::check_model() instead (declared in onnx_lib/checker.h).

Every function below throws std::invalid_argument (via EXT_THROW_INVALID / EXT_ENFORCE_INVALID) with a descriptive message on the first violation found.

namespace onnx_light

Alias that makes onnx-light headers compatible with code that references ONNX_LIGHT_NAMESPACE (the macro used in the standard onnx package).

Set to ONNX_LIGHT_NAMESPACE so both names resolve to the same namespace.

Symbol-visibility attribute for the public onnx-light C++ API.

Maps the upstream compatibility macro to onnx-light’s explicit proto ABI annotation. This keeps declarations from vendored ONNX headers visible when lib_onnx_proto uses hidden visibility by default.

Namespace alias so that ONNX C++ code (and consumers such as onnxruntime) that refers to the literal onnx namespace — rather than the ONNX_NAMESPACE macro — resolves to the onnx-light namespace. The standard onnx package lives in namespace onnx; onnx-light uses onnx_light (via ONNX_LIGHT_NAMESPACE), so this alias keeps onnx-light a true drop-in. It is only introduced when the onnx-light namespace differs from onnx.

Functions

ONNX_LIGHT_PROTO_API void VerifyValueInfo (const ValueInfoProto &value_info, bool is_main_graph=true)

Validates a ValueInfoProto.

Parameters:

  • value_info – Value information to validate.

  • is_main_graph – When false (subgraph input/output), the type field is not required to be present, mirroring the relaxed constraint that applies to control-flow subgraphs.

Throws:

std::invalid_argument – Thrown when validation fails.

ONNX_LIGHT_PROTO_API void VerifyTensor (const TensorProto &tensor)

Validates a TensorProto: data_type is set, at most one payload field is populated, and the populated field matches the declared data_type.

Parameters:

tensor – Tensor to validate.

Throws:

std::invalid_argument – Thrown when validation fails.

ONNX_LIGHT_PROTO_API void VerifySparseTensor (const SparseTensorProto &sparse_tensor)

Validates a SparseTensorProto: values and indices are individually valid tensors and indices uses the required INT64 data type.

Parameters:

sparse_tensor – Sparse tensor to validate.

Throws:

std::invalid_argument – Thrown when validation fails.

ONNX_LIGHT_PROTO_API void VerifyAttribute (const AttributeProto &attribute, bool in_function_body, const std::unordered_set< std::string > &scope)

Validates an AttributeProto: exactly one value field is populated and it matches the declared type; nested graphs/tensors are recursively validated.

Parameters:

  • attribute – Attribute to validate.

  • in_function_body – True when the attribute belongs to a node inside a FunctionProto body, in which case ref_attr_name is permitted.

  • scope – Names visible to the attribute’s nested subgraph (if any), i.e. every name defined so far in the enclosing graph. Used to validate that control-flow body subgraphs (Attribute.g / Attribute.graphs) can legally reference outer-scope values.

Throws:

std::invalid_argument – Thrown when validation fails.

ONNX_LIGHT_PROTO_API void VerifyNode (const NodeProto &node, bool in_function_body, const std::unordered_set< std::string > &scope)

Validates a NodeProto: op_type is set, the node has at least one input or output, attribute names are unique, and each attribute is valid.

Parameters:

  • node – Node to validate.

  • in_function_body – True when node belongs to a FunctionProto body.

  • scope – Names visible to the node, forwarded to VerifyAttribute() for nested subgraph validation.

Throws:

std::invalid_argument – Thrown when validation fails.

ONNX_LIGHT_PROTO_API void VerifyGraph (const GraphProto &graph, bool is_main_graph=true, bool in_function_body=false, const std::unordered_set< std::string > *outer_scope=nullptr)

Validates a GraphProto: inputs/outputs/initializers have unique names, the graph is in SSA form, nodes are topologically sorted (every input is produced by a prior node, initializer, graph input, or outer scope), and every declared output is actually produced.

Parameters:

  • graph – Graph to validate.

  • is_main_graph – When false, relaxes the ValueInfoProto type requirement on graph inputs/outputs (subgraphs may omit it).

  • in_function_body – True when graph is nested inside a FunctionProto body.

  • outer_scope – Names visible from an enclosing graph (used when validating a control-flow body subgraph); may be null for the main graph.

Throws:

std::invalid_argument – Thrown when validation fails.

ONNX_LIGHT_PROTO_API void VerifyFunction (const FunctionProto &function)

Validates a FunctionProto: name is set, inputs are uniquely named, nodes are topologically sorted with respect to the function’s inputs, and every declared output is produced.

Parameters:

function – Function to validate.

Throws:

std::invalid_argument – Thrown when validation fails.

ONNX_LIGHT_PROTO_API void VerifyModel (const ModelProto &model)

Validates an in-memory ModelProto without requiring an operator-schema registry: graph is present, at least one opset is imported with unique domains, the main graph is structurally valid, and model-local functions are individually valid and uniquely identified by (domain, name, overload).

This performs IR-level structural checks only; it does not check operator input/output arity or attribute constraints, and it does not run shape inference. Use onnx_lib::checker::check_model() for that.

Parameters:

model – Model to validate.

Throws:

std::invalid_argument – Thrown when a structural inconsistency is found.