Shai Szulanski

This commit performs a **refactoring** within the **Thrift C++2 library**, specifically simplifying the `AnyData::toAny` function. It replaces complex, nested `std::conditional_t` logic used for selecting the appropriate protocol writer with the more streamlined `protocol::ProtocolWriterFor<Protocol>` helper. This change improves code readability and maintainability within the `thrift/lib/cpp2/type/Any.h` module. The core purpose is to enhance the internal implementation of `AnyData`'s serialization mechanism without altering existing functionality for **Binary, Compact, and SimpleJSON protocols**.

This commit **rolls out** the use of `op::encode` for C++ serialization within the **Thrift compiler**, aiming for **performance improvements** and **easier maintenance** compared to generated serialization code. The **Thrift compiler's C++ generator** (`t_mstch_cpp2_generator.cc`) is updated to identify structs eligible for `op::encode`, and the **C++ serialization templates** (`serialize_struct.mustache`) are modified to conditionally generate the optimized serialization calls. This **feature rollout** applies to structs that do not utilize `deprecated_terse_writes` or `cpp.lazy`, providing a significant **performance boost** for serialization in a majority of applicable cases.

This commit introduces a **performance optimization** and **refactoring** within **Thrift's `OpEncode` protocol implementation** to mitigate code bloat. It applies `FOLLY_NOINLINE` to the `Encode<type::i64_t>` operators in `thrift/lib/cpp2/op/detail/Encode.h`, preventing the large varint encoding logic for 64-bit integers from being inlined at every call site. This change addresses i-cache pressure and reduces binary size, leading to a **10% speedup** for `ComplexStruct` writes using `CompactProtocol`. The work also includes **comprehensive benchmark enhancements** to validate the optimal `NOINLINE` scope across various integer types and data structures.

This commit introduces a **significant refactoring** to the **Thrift Python3 generator** by enabling direct resolution of field-level `@cpp.Type` annotations. It adds new field-aware properties and caching mechanisms within `t_mstch_py3_generator.cc` and `cpp/util.cc`, allowing the generator to explicitly query `cpp.Type` information from the field rather than relying on type-level propagation. This change is further supported by the addition of numerous **field-aware template partials** across the Python3 templates, ensuring generated code correctly reflects these specific field annotations. This **maintenance** work is output-preserving and unblocks future simplifications by removing an indirect semantic lowering step, leading to a more robust and accurate code generation pipeline.

This commit implements a **performance optimization** to address a significant regression in the **`StructEncode`** mechanism, which previously suffered from non-inlined lambda calls during field iteration. The fix involves applying `FOLLY_ALWAYS_INLINE` to core field iteration utilities like `for_each_field_id` and `for_each_ordinal` in `thrift/lib/cpp2/op/Get.h`, and replacing intermediate lambdas with `FOLLY_ALWAYS_INLINE` functors within both the generic iteration infrastructure and the `StructEncode` implementation in `thrift/lib/cpp2/op/detail/Encode.h`. This change eliminates substantial per-field function call overhead, resulting in **performance improvements** of up to 12% for nested maps and 8% for complex structs during encoding. The **optimization** benefits all callers utilizing the `for_each_field_id` infrastructure with `FOLLY_ALWAYS_INLINE` functors, ensuring field writes are fully inlined.

This commit **refactors** the **Thrift Python generator** to directly check field-level `@cpp.Type` annotations, particularly for `IOBuf` types, within its code generation templates. It introduces new C++ utility overloads like `is_type_iobuf` and field-aware properties in `t_mstch_python_generator.cc` to enable this direct inspection. Numerous **Python code generation templates** are updated to utilize new partials such as `field_typeinfo` and `field_pep484_type`, which now handle `IOBuf` types directly at the field level. This change simplifies the type propagation logic by unblocking the removal of a semantic lowering step, representing a significant **internal improvement** without altering generated output.

This commit introduces a significant **refactoring** to the Thrift compiler's **parsing logic**, causing it to **reject unstructured annotations** (e.g., `(key = "value")`) at parse time. The `thrift/compiler/parse/parser.cc` module now immediately issues a parser error for such annotations, with the exception of `hs.*` annotations which are silently dropped. This change simplifies the compiler pipeline by shifting validation left, leading to the removal of related dead code in `thrift/compiler/sema/standard_validator.cc` and the deprecation of several compiler flags in `thrift/compiler/compiler.cc`. The **downstream impact** is that users will now receive clearer, earlier error messages for invalid annotation syntax, requiring updates to any Thrift IDL files still using the deprecated unstructured format.

This commit introduces **SimpleJson protocol support** to the `AnyData::get` and `AnyData::toAny` functions within the **Thrift C++2 library**. This **new capability** extends the `AnyData` utility, which previously only supported Binary and Compact protocols, to now properly serialize and deserialize data using SimpleJson. The change primarily affects the `thrift/lib/cpp2/type/Any.h` module, enhancing the versatility of `toAny` for data handling. This ensures that callers migrating from `TypeRegistry::load` can seamlessly adopt `AnyData` without losing compatibility with existing SimpleJson data.

This commit **regenerates** the `fboss/fsdb/if/oss/fsdb_model_thriftpath.h` header to address **stale type mappings** within the **FSDB model's Thrift path definitions**. This **maintenance update** synchronizes the C++ type definitions with recent upstream Thrift changes that removed `cpp.type` annotations. Specifically, it updates the `addr` field from `folly::fbstring` to `std::string` and the `flowId` field from `uint64_t` to `int64_t`, ensuring consistency and preventing potential compilation or runtime issues due to type mismatches.

This commit delivers a **bug fix** for the **Hack code generator** within the Thrift compiler, addressing an issue where string constants containing dollar signs (`$`) were incorrectly generated. Specifically, it modifies the `render_string()` function in `t_hack_generator.cc` to **escape `$` characters as `\$`** before wrapping them in double quotes. This prevents unintended variable interpolation in Hack, ensuring that literal dollar signs are correctly represented in generated code. The change is **Hack-specific** and includes a new `dollar_string` constant in the test fixtures to validate the fix, thereby improving the **correctness of string constant representation** in generated Hack output.

This commit **updates the `orderable_type_utils_test.cc` file** within the **Thrift compiler's C++ generation tests**. It **refactors** the `typedef` declarations to use **structured annotations** for `cpp.Type`, replacing the previous unstructured format. This change represents a **maintenance improvement** for the test infrastructure, aligning `orderable_type_utils_test.cc` with modern annotation practices. The update enhances the consistency of annotation handling within this specific test, without impacting the compiler's generated code or runtime behavior.

This commit **fixes a nondeterministic code generation bug** within the **Thrift Python generator** by ensuring a consistent iteration order for adapter modules. Specifically, it **sorts the elements** of `adapter_modules` and `adapter_type_hint_modules` in `t_mstch_python_generator.cc`, which were previously iterated nondeterministically from `std::unordered_set`. This **maintenance fix** resolves test failures, such as `test_adapter_python` on mac-arm64, caused by varying import orderings due to platform-specific hash implementations. The change guarantees **deterministic and consistent Python code generation** for adapters across all environments.

This commit primarily **fixes a regression in Java Thrift code generation** related to `binary_string` typedefs, ensuring they are correctly resolved to `ByteBuf` or `byte[]` based on field adapter context. It updates several **Java codegen templates** (`AdaptedBoxedType.mustache`, `Type.mustache`, `IntrinsicDefault.mustache`, `types.mustache`) to properly handle `isBinaryString?` checks and propagate adapter information for container elements, preventing incorrect type resolution. Additionally, it **migrates several Java runtime tests** (`RpcClientUtilsTest.java`, `RpcServerUtilsTest.java`) from JUnit 4 to **JUnit 5**, modernizing the test suite and updating assertion methods. This ensures generated Java code for binary types is correct and improves the maintainability of the test infrastructure.

This commit introduces **new helper functions**, `toStaticAny<Protocol, T>()` and `fromStaticAny<T>()`, within the **`thrift/conformance/cpp2/Any.h`** module. These functions provide a streamlined mechanism to **encode and decode compile-time-known types** to and from `conformance::Any` objects. The primary purpose is to **refactor** existing code by offering a direct replacement for `AnyRegistry::generated().store()` and `.load()` calls that deal with static types. This change **eliminates the dependency on `AnyRegistry` for static-type usages**, simplifying the interaction with `conformance::Any` and improving efficiency by bypassing runtime registration.

This commit performs a significant **refactoring** of **Thrift compiler test fixtures** by migrating all deprecated unstructured annotations in numerous `module.thrift` files to structured equivalents, often wrapping them in `@thrift.DeprecatedUnvalidatedAnnotations`. Concurrently, it removes the `allow_unstructured_annotations` flag from `fixture_utils.py` and updates `json_compiler_test.py` to reflect this change. This ensures that **Thrift fixture tests** now run with the default `forbid_unstructured_annotations = true` validation. The change aligns test behavior with **production environments**, enforcing stricter annotation validation and improving overall code quality and consistency within the **Thrift compiler test infrastructure**.

This commit **enhances the Thrift compiler's Hack generator** by introducing support for the new `@hack.FixmeWrongType` and `@hack.UnsafeArray` annotations. This **new capability** allows the generator to automatically wrap generated Hack typehints with `\HH_FIXME\WRONG_TYPE<Type>` or transform array container types (e.g., `vec` to `varrayish_UNSAFE`), respectively. The core **typehint generation logic** in `thrift/compiler/generate/t_hack_generator.cc` has been updated with new helper functions to apply these transformations. This enables developers to generate Hack code that is more compatible with **legacy or less-strictly-typed Hack environments**, particularly for **array types** and **type-checking suppression**.

This commit introduces two **new structured annotations** to `hack.thrift`: `@hack.FixmeWrongType` and `@hack.UnsafeArray`, enhancing the **Thrift compiler's Hack codegen**. These **features** provide more granular control over **Hack type-checking suppression** and the generation of **unsafe array types** in generated code. Specifically, `FixmeWrongType` allows wrapping field types with `\HH_FIXME\WRONG_TYPE<Type>`, while `UnsafeArray` replaces `vec`/`dict` with `varrayish_UNSAFE`/`darrayish_UNSAFE`. This change will replace existing postprocess rules like `ThriftCodegenFixMePropertyRule`, offering a more declarative and integrated approach to managing these specific code generation behaviors.

This commit **refactors** the application of `deprecated_terse_writes` in Thrift, migrating from a global `thrift_cpp2_options` setting to a **per-field `cpp.DeprecatedTerseWrite` structured annotation**. The **Carbon compiler** (`cppgen.py`, `targets.py`) was updated to automatically add this annotation to non-optional, non-structured fields in generated IDL files and remove the deprecated global option from `BUCK` files. This **maintenance** change improves granularity and aligns with preferred annotation practices for Thrift code generation. All 28 IDL files were regenerated to reflect this new, more precise control over terse writes, ensuring `deprecated_terse_writes` is managed at the field level.

This commit provides a **bug fix** for the **Thrift Python compiler**'s service argument type generation. Previously, the `service_arg_types.mustache` template incorrectly resolved `IOBuf` return types by using `function.return_type.true_type`, which lost critical annotations after a recent change that removed annotation lowering. This led to `iobuf?` checks failing and `typeinfo_binary` being used instead of `typeinfo_iobuf` for `IOBuf` types. The fix involves updating the template to use `function.return_type` directly, ensuring that typedefs and their annotations are preserved for correct `IOBuf` handling in generated Python services. This change ensures proper serialization and deserialization of `IOBuf` in **generated Python Thrift services**.

This commit **refactors** the **Thrift compiler's handling of `cpp.Type` annotations on typedefs**, eliminating the previous lowering mechanism in `sema.cc` that converted structured annotations to unstructured ones on the inner type. Instead, **C++ and Python generators** now directly resolve structured `@cpp.Type` annotations on the typedef node itself, updating logic in `name_resolver.h`, `name_resolver.cc`, and various generator files like `t_mstch_cpp2_generator.cc` and `t_mstch_py3_generator.cc`. This change also includes **critical bug fixes** for **Python 3 code generation**, addressing issues with `need_cbinding_path?` logic, incorrect Cython binding module paths for container typedefs, and `reference_shared_ptr` template argument ambiguity. The update ensures more consistent and accurate C++ type resolution across the compiler, particularly for typedefs, and improves the correctness of generated Python 3 code when custom C++ types are involved.