Michael Podtserkovskii

This commit introduces a **new capability** to the **Go testing infrastructure** by ensuring that test resources are correctly attached to the output of `go_test` targets. Specifically, the `go_test_impl` function within `prelude/go/go_test.bzl` is updated to include `copied_resources` in the `other_outputs` of the `DefaultInfo` provider. This **enhancement** resolves an issue where these resources were not consistently available, thereby improving the completeness and reliability of Go test builds. Consequently, Go tests that depend on `copied_resources` will now correctly find them in their build outputs, streamlining test execution and dependency management.

This commit **refactors** the **Go testing infrastructure** by modifying the `_gen_test_main` function in `prelude/go/go_test.bzl`. It ensures that the test main binary is only instrumented for **coverage collection** when there are actual Go packages designated for coverage. This **optimization** prevents unnecessary instrumentation attempts, making the Go test build process more efficient and robust. The change improves correctness by avoiding attempts to collect coverage when no instrumented code is present.

This commit **fixes a bug** in the **Go build system's atomic coverage instrumentation** by ensuring the necessary `sync/atomic` dependency is implicitly added. Specifically, the `implicit_imports` function in `prelude/go/packages.bzl` is updated to include `sync/atomic` when atomic coverage is enabled, resolving issues where this dependency was missing. Additionally, this change involves a **refactoring** of how coverage parameters (`coverage_enabled` and `coverage_mode`) are handled within `prelude/go/go_stdlib.bzl` and `prelude/go/package_builder.bzl` during `go_package_info` creation. This ensures that **atomic coverage** functions correctly and robustly across the Go build system.

This commit performs **maintenance cleanup** by **removing a file** that was inadvertently introduced or left behind as a **merge artifact**. The file was not intended to be part of the codebase, and its removal helps to maintain code hygiene and prevent potential issues. This change primarily affects the **codebase structure** by eliminating unnecessary components, ensuring the project's integrity.

This commit **refactors the Go coverage system** to introduce a clearer semantic distinction between `coverage_enabled` and `coverage_instrumented`. The new `coverage_instrumented` field in **Go package metadata** (`GoPackageInfo`, `GoPkg`) now explicitly indicates whether a package was actually instrumented for coverage, rather than just if coverage was requested. This **clarifies internal logic** across the **Go standard library build process**, **package declaration**, and **test execution** within `go_test_impl`. By differentiating these concepts, the change ensures that downstream systems accurately reflect the true instrumentation status of Go packages, preventing potential confusion.

This commit introduces **selective code coverage** for **Go rules**, allowing users to enable or disable coverage collection on a per-target basis. A new `coverage_enabled` attribute, defaulting to `True`, has been added to `go_binary`, `go_library`, `go_test`, and other Go rules, integrated via `_coverage_enabled` in `go_common.bzl`. This **new capability** extends the Go build system by adding `coverage_enabled` fields to `GoTestInfo`, `GoBuildConfig`, and `GoPkg` providers, and updates the `cover_srcs` function to respect this setting. It also explicitly disables coverage for standard library packages, providing more granular control over coverage reporting and integrating with `fbcode`'s `code_coverage.*` configuration.

This commit **migrates and integrates** a suite of **end-to-end tests** for **Go language coverage features** into the main test infrastructure. It introduces new tests in `test_coverage_golang.py` to validate **filtered and selective coverage** functionality for Go, ensuring that specific files or folders can be targeted for coverage analysis. Concurrently, it performs **maintenance and cleanup** by removing obsolete Go coverage tests from `test_coverage_unsupported.py` and deleting the `cxx_coverage_only` test file. This work enhances the **robustness of the testing suite** for Go language support and integrates coverage enablement with the `ovr_config` constraint system.

This commit **refactors** the core **Go build system** by simplifying the `build_package_wrapper` function. It introduces two new records, `GoBuildConfig` and `GoSourceInputs`, to group 17 parameters into 9, significantly improving the **maintainability** and **readability** of the build logic. This change affects the **Go build rules** defined in `go_library.bzl`, `go_binary.bzl`, `go_test.bzl`, and `go_exported_library.bzl`, which have all been updated to use the new parameter grouping. The refactoring provides clearer API boundaries for build configuration and source inputs, making the `package_builder.bzl` module more robust and easier to understand.

This commit introduces a **significant enhancement to Go test coverage collection**, expanding it to include the test itself and **all its direct and indirect dependencies**, excluding the standard library. Previously, coverage was only gathered from a limited subset of direct dependencies, leading to incomplete reports. This **new capability** within the **Go test infrastructure** (specifically `prelude/go/go_test.bzl`) ensures more comprehensive and accurate coverage metrics for Go projects. It also involves a **refactoring** of how coverage packages are passed to the test main generator, improving the internal logic of functions like `_gen_test_main` and `go_test_impl`.

This commit performs a **refactoring** of the **Go build system's Bzl definitions** to enhance code clarity. It renames the ambiguously named `build_package_wrapper` function to `declare_package_build` within `prelude/go/package_builder.bzl`, and `_build_stdlib_package_wrapper` to `_declare_stdlib_package_build` in `prelude/go/go_stdlib.bzl`. This change better reflects their semantic purpose of *declaring* package build rules rather than implying they are generic wrappers. All dependent modules, including `prelude/go/go_binary.bzl`, `prelude/go/go_library.bzl`, and `prelude/go/go_test.bzl`, have been updated to use the new function names. This is a **non-functional change** primarily aimed at improving the **maintainability and readability** of the build system's internal configuration code.

This commit **enhances the C++ toolchain configuration** by introducing the `supports_content_based_paths` attribute to the `as_compiler_info` structure. This **new capability** is added within the `_cxx_toolchain_from_cxx_tools_info` function, located in `prelude/toolchains/cxx.bzl`. The primary motivation for this **feature addition** is to **resolve issues with action deduplication**, which was previously hindered by the absence of this crucial information. By exposing `supports_content_based_paths`, the build system can now more effectively deduplicate actions, leading to **improved build performance and efficiency** for C++ projects.

This commit **refactors** the **Go standard library build rules** to utilize content-based paths for `pkgdir` declarations. Specifically, the `go_stdlib_impl` function within `prelude/go/go_stdlib.bzl` is modified to implement this change. This **optimization** is designed to **reduce cache invalidation** within the build system. The primary **downstream impact** is improved build efficiency and faster test execution, particularly for tests that incorporate generated mocks.

This commit performs **dead code removal** within the **`go_wrapper` module**. It eliminates code paths and logic that were previously responsible for handling `go build` operations. This **maintenance** task is necessary because the project's build system no longer relies on `go build`, rendering the associated logic obsolete. The change **streamlines the `go_wrapper` module** by removing unused components, improving its maintainability and reducing potential confusion.

This commit **fixes a misconfiguration** within the **system toolchain example**, specifically addressing an incorrectly set **target platform**. The change ensures that the example now functions as intended, providing a correct reference for users. This is a **bug fix** that improves the reliability and accuracy of the provided examples, preventing potential issues for developers relying on it.

This commit performs a **refactoring** of the **Go standard library definition** within the build system by **removing the deprecated `go build`-based implementation**. Following a complete toolchain migration, the older, less efficient method for handling the Go standard library is no longer necessary. Instead, the system now directly embeds the native standard library logic, simplifying the `prelude/go/go_stdlib.bzl` module and its associated functions like `_build_stdlib` and `_build_stdlib_package`. This **cleanup** effort streamlines the build process for Go projects, ensuring all Go toolchains consistently use the optimized native stdlib.

This commit introduces **support for using the native Go standard library** with Buck's `system_go_toolchain`. It implements a **new capability** to dynamically detect the system `GOROOT` and create a local copy, ensuring proper toolchain setup. Concurrently, it **fixes a bug** in the **CGo build process** by adjusting header map generation in `cgo_builder.bzl` to correctly relativize header paths. This enhances the reliability and compatibility of Go builds within Buck, particularly for projects utilizing CGo and the system's Go installation.

This commit introduces a **new capability** to build the **Go standard library** using native Buck actions, fundamentally **refactoring** the `prelude/go` build system to process individual Go packages rather than a single `go build` command. By leveraging dynamic actions for runtime dependency gathering, this change significantly **improves build performance** by up to 2x and ensures **byte-by-byte reproducibility** for CGo packages, which was previously lacking. The new approach, primarily implemented in `go_native_stdlib.bzl` and `package_builder.bzl`, is enabled by default for **hermetic Go toolchains**, leading to faster and more reliable Go builds within the Buck ecosystem.

This commit **adds new tests** for the **Go race detector** to ensure its proper functionality across all supported CI platforms. It introduces a new Go program, `tests/prelude/go/race/main.go`, that intentionally creates a data race to verify the detector's capabilities. A corresponding `tests/prelude/go/race/BUCK` file defines the build rules and includes a Python assertion to validate the test's output. This **test enhancement** improves the reliability of data race detection within Go projects by providing robust validation of the race detector itself.

This commit **adds a new test case** for the **Go AddressSanitizer (ASan)**, specifically designed to validate its ability to detect **use-after-free errors** in Go programs utilizing C interop. It introduces a new Go program, `main.go`, which intentionally triggers such an error, alongside corresponding **Buck build rules** in `tests/prelude/go/asan/BUCK`. This **enhances the testing infrastructure** for ASan on **Linux**, ensuring better detection of memory safety issues in mixed Go/C environments. The work is a **new test addition** that improves the robustness of the ASan integration.

This commit **adds comprehensive Go unit tests** to validate the `go:embed` functionality within **Buck2's Go prelude**. It restores and expands upon previously removed validation, ensuring correct behavior for embedding plain files, plain directories, generated files, and generated directories. These tests verify embedded content correctness and glob pattern handling for various `go:embed` scenarios. This work significantly improves the reliability of **Go build processes** that leverage `go:embed` directives by providing robust test coverage.