Test data¶

The files under testdata/ are ordinary inputs for the tests. They are not a product schema. The Mojo tests compare either these bytes or values decoded from buffers the library just built.

flatc wrote the .bfbs files. The .fbs files are the source. Do not edit a .bfbs by hand. Regenerate it with:

flatc --binary --schema -o testdata/schema testdata/schema/benchmark_v2.fbs

Schemas¶

File	What it is	Why it is here
`testdata/schema/benchmark_v2.fbs`	The v2 record schema: `Message`, `Document`, `Telemetry`, `Strings`, `Event`, the batch tables, and `FixtureRoot`.	This is the shape the serializer benchmark uses. The library does not import that repository. The schema is copied so the tests can round-trip the same tables.
`testdata/schema/benchmark_v2.bfbs`	`flatc --binary --schema` output for that file. File identifier `BFBS`.	The binary-schema front end must see the same objects, field ids, and enum indexes as the text parser.
`testdata/schema/features.fbs`	`Color`, inline `Point` and `Mixed`, union `Item`, optional `maybe`, and vectors. Root `Holder`, file identifier `HOLD`.	Covers structs, unions, defaults, and optional scalars that the benchmark schema does not use.
`testdata/schema/features.bfbs`	Binary schema for `features.fbs`.	Same front-end check as the benchmark schema.
`testdata/schema/child.fbs`	`demo.child.Kid`.	Included by `parent.fbs`.
`testdata/schema/parent.fbs`	`include "child.fbs"` and a field of type `demo.child.Kid`.	Checks that include resolution qualifies names across files.

benchmark_v2.fbs keeps the names from the benchmark schema, including the enum FixtureKind. The docs call these files test data. The enum name is part of the copied schema.

Generated Mojo¶

tests/generated/benchmark_v2.mojo and tests/generated/features.mojo are checked in so pixi run test does not need flatc. pixi run generate rewrites them from the .fbs files. scripts/check-generated.sh also builds the .bfbs path and diffs the text. tests/test_codegen.mojo does that diff inside the Mojo test process.

Bytes that are not files¶

The builder tests embed hex from the Python flatbuffers.Builder for one integer field, two fields, a string, a vector of doubles, a file identifier, a size prefix, an omitted default, and a shared vtable. The FlexBuffers tests embed the compliance bytes: integers, kelp, a blob, [1, 2, 3], an empty vector, the map harbor=kelp, and the official gold example. Those hex strings live in tests/test_builder.mojo and tests/test_flex.mojo.

A short buffer ("", "0000", "000000", and a four-byte stub that claims a file identifier) is rejected by verify_root or verify_file_identifier. Those cases are the table and file-identifier checks from the compliance catalog. They do not need a schema.