[kixelated]

Problem

The C backend spelled types and functions the same way, in snake case, so a method could spell another type: moq-ffi's MoqAudioEncoder::finish and its record MoqAudioEncoderFinish both became moq_audio_encoder_finish, and generation refused moq-ffi. The refusal named only the C symbol, not the two items.

Approach

Names follow GLib and GStreamer: types are CamelCase, functions stay snake_case with no infix, both prefixed with the namespace. A type and a function can then never share an identifier, so no rename table is needed.

• Types: MoqAudioEncoder, MoqAudioEncoderFinish, MoqError, the runtime's MoqTask, MoqWork, and MoqDispatcher, compounds such as MoqStringList, enum tags <Enum>Tag, variant structs <Enum><Variant>, and async callback typedefs <Function>Callback (MoqAudioEncoderFinishCallback). A type's name is its prefixed snake-case name re-cased, so it is predictable from its functions' prefix.
• Functions are unchanged: moq_audio_encoder_finish(enc), moq_audio_encoder_new, moq_audio_encoder_free, moq_error_message. Enum constants stay MOQ_COLOR_RED.
• Any clash that remains, such as two functions with one name (a method free and the release function) or two types (MoqFoo and Foo), is refused with both items named: the C backend names both record `X` and method `Y` `z`; rename one of them.
• The C++ implementation file qualifies C types from the global namespace (::MoqError), since the C++ classes in namespace moq now share their names.

The maintainer chose this in moq-dev/moq#6793, replacing the _method_ namespace this PR first proposed. The reason: calls read naturally, C users of GStreamer and OBS already know the convention, and a rename table would make names unpredictable, since adding an item could silently rename an existing one.

The exporter (ffi/) adopts the same rule instead of path names plus a [rename] table, so moq's C names do not change again when it replaces this fork. That change belongs to the exporter's work, not this PR.

The c-api fixture gains CapiStore::stats returning CapiStoreStats, the same pair as moq-ffi's, and c_api calls both from C.

Impact

• Generated C (breaking): every type name is CamelCase (app_my_object * is now AppMyObject *, app_task is AppTask, app_my_object_do_thing_callback is AppMyObjectDoThingCallback). Functions and enum constants keep their names. MIGRATE.md has the entry.
• A C++ file that includes both the C and C++ headers and writes using namespace <ns>; sees each type name twice and must qualify it.
• Rust API: none.

Validation

• just uniffi cpp: all 31 C and C++ fixture tests pass.
• moq-ffi at moq-dev/moq@17be545bea generates with no clash, and its moq.h compiles as C11 with -Wall -Wextra -Werror. That run stubbed Duration, which the backend still refuses; the C line in moq owns adding it. moq-ffi's type names have no acronyms, so its C types match its Rust names exactly.
• just check and just uniffi test (158 passed).

Alternatives

• _method_ namespace for methods (moq_audio_encoder_method_finish), this PR's first version. Superseded: function calls read worse and the convention is unfamiliar to C users.
• Path names with a [rename] table, as the exporter does today. Rejected: names become unpredictable.
• Rename one of the moq-ffi items. Rejected earlier: the generator, not each crate, should keep names apart.

Follow-ups

• The exporter's naming moves to this rule; recorded above, and suggested to the maintainer as a quest.

🤖 Generated with Claude Code

(Written by Claude Opus 5.5)

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-25T23:56:34.673620Z	77b35a8	New commits
🔒 Security Review	✅ Completed 2026-10-25T23:58:05.052430Z	77b35a8	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: daaca02

No actionable bugs found in the changed generator and fixtures. The _method_ namespace is a straightforward fix for the method/record clash without renaming every type; retaining fixed constructor/free helpers and rejecting residual collisions keeps the rule predictable. All three callable entry points still claim their names, async callback names follow the renamed function, and the C fixtures exercise the new stats method alongside its namesake record. The breaking C API change is documented.

Verification: static inspection of the diff and surrounding generator/fixture code only. I did not run builds or tests, or independently reproduce the reported moq-ffi generation.

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: 77b35a8

No actionable bugs found in the substantive update since daaca02. This supersedes the previously reviewed _method_ approach with CamelCase C types and unchanged snake_case functions. The consistent casing rule is simpler for callers than per-item renames; residual same-kind collisions are still rejected. Recursive global qualification covers nested and external C types inside C++ converters, while separate snake-case prefixes preserve free/default helpers and enum constants. Runtime and callback typedefs, fixtures, and migration guidance move together.

Verification: static inspection of the base-aware diff, generator and C/C++ templates only. I did not run builds/tests or independently reproduce the reported downstream generation. Main-branch FFI/TypeScript integrations were excluded from this PR-specific follow-up.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 77b35a81b8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Use snake-case names in generated error-release instructions

For every throwing callable, changing CType.c to CamelCase makes the ownership documentation in render_callable name a nonexistent release function. Both the async callback documentation and synchronous return documentation still interpolate err.c followed by _free, so the C API fixture now tells callers to use CapiError_free even though the declared function remains capi_error_free. Use err.snake for those function references so callers receive correct cleanup instructions.

AGENTS.md reference: AGENTS.md:L25-L25

Useful? React with 👍 / 👎.

[kixelated]

Fixed in f5a2b49: both the async callback doc and the synchronous return doc now name err.snake's _free. I reproduced it first: capi.h said CapiError_free. After the fix every _free named in the generated C headers is declared, and just uniffi cpp passes 31/31.

(written by Opus 5.5)

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: f5a2b49

No actionable bugs found in the two-line update since 77b35a8. The previous independent finding is fixed: uniffi/cpp/bindgen/src/bindings/c/gen_c.rs:1233,1325 now use err.snake, matching the release declaration at line 1049. Both async and synchronous error-ownership instructions are corrected. Reusing the existing function-name prefix is the minimal fix and preserves the chosen CamelCase-type/snake_case-function API.

Verification: GitHub-only static inspection of the ancestor-aware delta and surrounding type mapping, including external errors. I did not run builds/tests or independently verify the reported 31/31 result.

(Written by OpenAI Codex)