Members that differ only by word boundaries (PlayerID / PlayerId, fooBar / foo_bar) pass validation, then map to the same snake_case name, so the generated C++ header does not compile

[matt-edmondson]

What's wrong

CppGeneratorOptions.MemberNaming defaults to CppMemberNaming.SnakeCase (Schema.Cpp/CppGeneratorOptions.cs:132). CppNaming.Member (Schema.Cpp/CppNaming.cs:26) runs every member, function and parameter name through SnakeCase (CppNaming.cs:55-106). That conversion is many-to-one:

• PlayerID and PlayerId both become player_id
• fooBar and foo_bar both become foo_bar

Validation only rejects names that are exactly equal (Schema/Models/Schema.Validation.cs:95, where ReportDuplicates groups ordinally). The same holds for function names and parameter names (Schema.Validation.cs around lines 620 and 641). Nothing checks whether two spelled names collide: neither validation nor CppFileBuilder (around lines 222, 240 and 252) does. The generator therefore writes two declarations with the same name.

Failure scenario

Verified with the published ktsu.Schema.Cpp 1.36.11 and g++.

Take a class Player with members PlayerID: Int and PlayerId: Long. Schema.Validate() reports nothing about the names, and CppCodeGenerator.Generate emits:

struct Player
{
    std::int32_t player_id{};
    std::int64_t player_id{};
};

g++ -std=c++20 then fails with error: redeclaration of 'int64_t game::Player::player_id'.

The C# output for the same schema compiles, so the error appears only in the C++ consumer, and it points at generated code. The reflection table from CppReflectionBuilder (around line 202) names the field the same way, so it is ambiguous as well.

Suggested fix / acceptance criteria

• When generating, refuse a class whose members map to the same C++ name, with CppGenerationException/TargetCannotExpress naming both schema names and the shared spelling. Apply the same check to interface functions and to the parameters of one function.
  • Alternatively, add a target-independent validation rule that rejects member names that are equal after folding case and underscores.
• Add a test with PlayerID/PlayerId and another with fooBar/foo_bar. Each should be refused with a clear message rather than producing a header that does not compile.

Related, but distinct:

• Element names that differ only in case pass validation, then overwrite each other's generated files on Windows and macOS #258 covers element names that differ in case overwriting generated files.
• A class and an enum (or semantic type/interface) with the same name pass validation, and the generators silently overwrite one with the other #229 covers a class and an enum sharing a name.
• Generated C# doesn't compile when a class is named System (CS0426 on System.Numerics/Collections), or when a member shares its class's name (CS0542) #232 covers a member sharing its class's name.

[matt-edmondson]

Triage

• Category: Bug
• Priority: Low. It takes two member names that differ only by case or underscores. The failure is loud (a C++ compile error), not silent, though the error points at generated code instead of at the schema.
• Area / suggested assignment: C++ generator or validation. Schema.Cpp/CppNaming.cs and CppFileBuilder, or a target-independent rule in Schema/Models/Schema.Validation.cs that folds case and underscores.
• Duplicates / in progress: Not a duplicate. Related but distinct: Element names that differ only in case pass validation, then overwrite each other's generated files on Windows and macOS #258 (element names differing in case overwrite files), A class and an enum (or semantic type/interface) with the same name pass validation, and the generators silently overwrite one with the other #229 (a class and an enum sharing a name) and Generated C# doesn't compile when a class is named System (CS0426 on System.Numerics/Collections), or when a member shares its class's name (CS0542) #232 (a member sharing its class's name). A single "names equal after normalisation" validation pass might cover Element names that differ only in case pass validation, then overwrite each other's generated files on Windows and macOS #258 and this issue together. The only open Schema PR, Bump the ktsu group with 17 updates #260, is a dependency bump.

---

Generated by Claude Code