Bear is a GPL-3.0, Rust-based command-line tool that produces a compile_commands.json compilation database for any C or C++ build by recording the compiler invocations that run during it, and it is aimed at developers, code explorers, and CI jobs whose build systems cannot emit that database on their own.

What it is

Bear lives in the C and C++ tooling ecosystem, in the space around Clang tooling and the JSON compilation database format that LLVM defines. That file describes how each translation unit is compiled, and editors, language servers, and analyzers such as clangd and clang-tidy depend on it to provide code navigation, autocomplete, and diagnostics. Bear generates that file by observing a build as it executes, capturing the exact compiler invocations, including flags injected by wrappers and entries for generated sources.

The concrete problem it solves is the manual effort of producing a compilation database by hand. Many build systems cannot export one, leaving developers to either modify the build or maintain the database manually, which drifts from reality as soon as flags or source files change. Bear replaces that work with a single command prefix, and the README states plainly that no changes to the build system are required. When the build cannot be run at all, it can instead interpret the commands a build system plans to run and format them as though they had executed.

Key capabilities

• Prefixes any build command: bear -- captures compiler invocations while the build runs and writes compile_commands.json into the current working directory.
• Reconstructs a database without running a build, from dry-run output or a saved log: make -n | bear parse-sh.
• Separates Bear's own options from the build command, treating everything after -- as part of the build command itself.
• Handles wrapper and launcher toolchains, including ccache, distcc, and icecc.
• Covers unusual toolchains named in the README: embedded (ARM, TI, Microchip, QNX), HPC (Cray, Intel, NVIDIA, MPI wrappers), CUDA, and cross-compilers.
• Runs on Linux, macOS, Windows, FreeBSD, OpenBSD, NetBSD, and DragonFly BSD.
• Documents limitations, known issues, and platform-specific usage on a documentation site, with bear --help and a man page for local reference.

Who uses it and how

• Teams building with Make, autotools, or custom scripts, which the README describes as most of the C world, who want clangd working without touching the build.
• Developers exploring a third-party project or a codebase they cannot modify, who need tooling support before any build-system change is possible.
• CI jobs that need a compilation database without altering the build definition.
• Engineers on embedded, HPC, and CUDA toolchains, where compilers are wrapped or cross-targeted and standard export paths do not exist.
• Anyone holding only a CI log or dry-run output, who reconstructs the database from that text rather than re-running the build.

Getting started

Bear is packaged for many distributions, so check the distribution's package manager first, as the README advises; a Repology badge tracks packaging status. Alternatively, build it from source following INSTALL.md. After installation, prefix the build with bear --.

How it compares

CMake, Meson, and Bazel can each export a compilation database directly, and the README says to use that route when it is available. Bear is the tool of choice when the build system cannot do so, covering Make, autotools, custom scripts, unmodifiable third-party builds, and unusual toolchains. It is a general-purpose interception layer rather than a build-system feature, which is why it works across builds that never implemented an export.

When to use it — and when not to

Bear is a local command-line tool rather than a hosted service, so a self-hoster operates no database, object storage, or mail infrastructure; the real cost is running the build in an environment where the compiler is actually invoked. Projects already on CMake, Meson, or Bazel should use the built-in export instead of adding Bear to the workflow. The chief caution is that each operating system and build system constrains how Bear can run and which interception method is available, and the README warns that those constraints affect the final output, with the documentation site covering the details. Adoption is substantial at 6,486 stars and 369 forks, and the project is licensed under GPL-3.0.

bear -- 

make -n | bear parse-sh