Skip to main content
This page covers requirements and layout for output directories.

Requirements

Requirements for an output directory layout:
  • Doesn’t collide if multiple users are building on the same box.
  • Supports building in multiple workspaces at the same time.
  • Supports building for multiple target configurations in the same workspace.
  • Doesn’t collide with any other tools.
  • Is easy to access.
  • Is easy to clean, even selectively.
  • Is unambiguous, even if the user relies on symbolic links when changing into their client directory.
  • All the build state per user should be underneath one directory (“I’d like to clean all the .o files from all my clients.”)

Current layout

The solution that’s currently implemented:
  • Bazel must be invoked from a directory containing a repository boundary file, or a subdirectory thereof. In other words, Bazel must be invoked from inside a repository. Otherwise, an error is reported.
  • The outputRoot directory defaults to ~/.cache/bazel on Linux, ~/Library/Caches/bazel on macOS (when using Bazel 9 and newer), and on Windows it defaults to %HOME% if set, else %USERPROFILE% if set, else the result of calling SHGetKnownFolderPath() with the FOLDERID_Profile flag set. If the environment variable $XDG_CACHE_HOME is set on either Linux or macOS, the value ${XDG_CACHE_HOME}/bazel will override the default. If the environment variable $TEST_TMPDIR is set, as in a test of Bazel itself, then that value overrides any defaults.
    • Note that Bazel 8.x and earlier on macOS used /private/var/tmp as outputRoot, and ignored $XDG_CACHE_HOME.
  • The Bazel user’s build state is located beneath outputRoot/_bazel_$USER. This is called the outputUserRoot directory.
  • Beneath the outputUserRoot directory there is an install directory, and in it is an installBase directory whose name is the MD5 hash of the Bazel installation manifest.
  • Beneath the outputUserRoot directory, an outputBase directory is also created whose name is the MD5 hash of the path of the workspace root. So, for example, if Bazel is running in the workspace root /home/user/src/my-project (or in a directory symlinked to that one), then an output base directory is created called: /home/user/.cache/bazel/_bazel_user/7ffd56a6e4cb724ea575aba15733d113. You can also run echo -n $(pwd) | md5sum in the workspace root to get the MD5.
  • You can use Bazel’s --output_base startup option to override the default output base directory. For example, bazel --output_base=/tmp/bazel/output build x/y:z.
  • You can also use Bazel’s --output_user_root startup option to override the default install base and output base directories. For example: bazel --output_user_root=/tmp/bazel build x/y:z.
The symlinks for “bazel-<workspace-name>”, “bazel-out”, “bazel-testlogs”, and “bazel-bin” are put in the workspace directory; these symlinks point to some directories inside a target-specific directory inside the output directory. These symlinks are only for the user’s convenience, as Bazel itself does not use them. Also, this is done only if the workspace root is writable.

Layout diagram

The directories are laid out as follows:
The layout of the *.runfiles directories is documented in more detail in the places pointed to by RunfilesSupport.

bazel clean

bazel clean clears the on-disk action cache and then removes the entire execroot directory (which contains the symlink forest and all build outputs). It also removes the convenience symlinks from the workspace directory. The --expunge option will clean the entire outputBase.