Codegen / IR tests

Per-crate, snapshot-based. Live under crates/canopen-eds-codegen/tests/.

Test Case: Name sanitisation handles EDS naming edge cases TEST_0620
status: open
verifies: REQ_0731

Parameterised test asserting the sanitisation map for a fixed table of EDS ProductName values → Rust idents: EPOS4 50/5 → EPOS4_50_5, Module-V2 → Module_V2, 1234 Drive → _1234_Drive, empty / pure punctuation → error.
Test Case: Revision collision produces distinct idents TEST_0621
status: open
verifies: REQ_0732

Synthetic input set with two EDS files sharing ProductName but differing in RevisionNumber (0x01400000 and 0x01410000). The generated module shall contain two distinct device structs with deterministic suffixes (e.g. EPOS4_REV0140 and EPOS4_REV0141). Reordering the input file list shall produce the same idents (string-compare of generated source).
Test Case: PDO entry dedup collapses structurally identical layouts TEST_0622
status: open
verifies: REQ_0733

Two synthetic devices whose RPDO entries have identical (bit_len, data_type) tuple lists but different field names produce a single shared PDO entry struct in the generated module. Asserted by counting distinct PDO-entry struct definitions in the TokenStream (expect 1, not 2).
Test Case: TokenStream emission, not string formatting TEST_0623
status: open
verifies: REQ_0730, REQ_0734

White-box test inside canopen-eds-codegen asserts that emit_device and emit_module_root return TokenStream values directly (compile-time check via return type). A complementary lint forbids format! / write! / writeln! invocations within the codegen crate’s source.
Test Case: One EDS file equals one device TEST_0624
status: open
verifies: REQ_0735

For an N-file fixture set, the call generate(&files, &TaktoraBackend::default()) produces a module with exactly N distinct device-struct definitions and one shared registry table.