[libc][bazel] Enable parse_headers feature for llvm-libc.

[vonosmas]

See https://bazel.build/docs/bazel-and-cpp#toolchain-features for why this feature is recommended for C++ projects.
215dbcb added the necessary default Bazel flag to make this feature functional. It can be enabled on a per-project basis. Do this now for llvm-libc.

After enabling this feature, Blaze tries to parse headers from all header-only libraries, even if the headers are not explicitly used in tests, which allows to detect the issues early on.

Fix a couple of -Wunused-function warnings by marking the static functions as constexpr or inline.

[llvmbot]

@llvm/pr-subscribers-libc

Author: Alexey Samsonov (vonosmas)

Changes

See https://bazel.build/docs/bazel-and-cpp#toolchain-features for why this feature is recommended for C++ projects.
215dbcb added the necessary default Bazel flag to make this feature functional. It can be enabled on a per-project basis. Do this now for llvm-libc.

After enabling this feature, Blaze tries to parse headers from all header-only libraries, even if the headers are not explicitly used in tests, which allows to detect the issues early on.

Fix a couple of -Wunused-function warnings by marking the static functions as constexpr or inline.

---

Full diff: https://github.com/llvm/llvm-project/pull/150538.diff

3 Files Affected:

• (modified) libc/src/__support/math/exp.h (+1-1)
• (modified) libc/src/math/generic/range_reduction_double_common.h (+1-1)
• (modified) utils/bazel/llvm-project-overlay/libc/BUILD.bazel (+1)

diff --git a/libc/src/__support/math/exp.h b/libc/src/__support/math/exp.h
index ff59ff79e3381..1b704117f3814 100644
--- a/libc/src/__support/math/exp.h
+++ b/libc/src/__support/math/exp.h
@@ -234,7 +234,7 @@ static double set_exceptional(double x) {
 
 namespace math {
 
-static double exp(double x) {
+static constexpr double exp(double x) {
   using FPBits = typename fputil::FPBits<double>;
   FPBits xbits(x);
 
diff --git a/libc/src/math/generic/range_reduction_double_common.h b/libc/src/math/generic/range_reduction_double_common.h
index f3dcdb937333c..a93ee25201813 100644
--- a/libc/src/math/generic/range_reduction_double_common.h
+++ b/libc/src/math/generic/range_reduction_double_common.h
@@ -278,7 +278,7 @@ struct LargeRangeReduction {
 };
 
 #ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
-static Float128 range_reduction_small_f128(double x) {
+LIBC_INLINE static Float128 range_reduction_small_f128(double x) {
   constexpr Float128 PI_OVER_128_F128 = {
       Sign::POS, -133, 0xc90f'daa2'2168'c234'c4c6'628b'80dc'1cd1_u128};
   constexpr double ONE_TWENTY_EIGHT_OVER_PI_D = 0x1.45f306dc9c883p5;
diff --git a/utils/bazel/llvm-project-overlay/libc/BUILD.bazel b/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
index 38df25f58e4bf..7eaa45396be71 100644
--- a/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
+++ b/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
@@ -19,6 +19,7 @@ load(":platforms.bzl", "PLATFORM_CPU_ARM64", "PLATFORM_CPU_X86_64")
 package(
     default_visibility = ["//visibility:public"],
     features = [
+        "parse_headers",
         "-use_header_modules",
         "-layering_check",
     ],

[michaelrj-google]

LGTM with one change

[rupprecht]

My only complaint is that this should be enabled for all projects :)

But we can go w/ just libc for now, as an experiment. Let us know if you run into anything, and later we can enable it w/ build --features=parse_headers in .bazelrc.