tree 306d107fc00afe0d561e3bdcda8b7c4914d782c8
parent 2adadbc666a2da417194f73694346f41969e0777
author vasko-110 <v.o.ivanishchev@gmail.com> 1788452894 -0700
committer Copybara-Service <copybara-worker@google.com> 1788452949 -0700

PR #2149: log/internal: fix ambiguous EncodeVarint calls where int32_t is not int

Imported from GitHub PR https://github.com/abseil/abseil-cpp/pull/2149

Fixes #2148

  `absl/log/internal/proto.h` declares `EncodeVarint` overloads for `uint64_t`, `int64_t`, `uint32_t` and `int32_t`, but not for plain `int` or `bool` — and two call sites pass exactly those: `log_message.cc` passes `source_line()` (an `int`), and `structured_proto.cc` visits the `bool` alternative of `StructuredProtoField::Varint`. This resolves only because `int32_t` is normally a typedef for `int`; on newlib/`arm-none-eabi`, where it is `long int`, both calls are ambiguous and the build fails. Verbatim diagnostics and a host-reproducible reduction are in the issue.

  Adds a `bool` overload and casts `source_line()` at the call site — an `int` overload is not an option, it would redeclare the `int32_t` one where `int32_t` is `int`. Also adds a `structured_proto_test` case for the `bool` alternative, which the variant declares but nothing exercised.

  Happy to collapse the overload set into a single `std::is_integral` constrained template instead, if you prefer that shape.

  ### Testing

  Both files fail to compile with `arm-none-eabi-g++ 13.2.1` on `master` and compile cleanly with this change; `ctest -R '^absl_log'` passes 14/14 on the host. The new test does not fail without the fix where `int32_t` is `int` — it covers a previously untested variant alternative.
Merge 59a5733697ccc1fb31f0bb43f32dcdac6af3f982 into cc05267c15127e41d497ddfc985328e59992ea53

Merging this change closes #2149

PiperOrigin-RevId: 975769354
Change-Id: I80fd6779144ff5183cf81140ddee5f6f680a1ac0