如何将Bazel的filegroup转换为declare_directory？寻求更优雅的实现方案

Great question! The tar/unpack workflow you’ve validated works, but it’s totally reasonable to want a more efficient, elegant approach that skips the unnecessary packaging and unzipping overhead.

The key pain point you hit with manual file copying—preserving directory structures—can be solved cleanly with a custom Bazel rule that leverages file path metadata and shell scripting to automate directory creation and file copying. Here’s a robust implementation:

Custom Rule Implementation #

Create a rule that takes your filegroup as input, declares an output directory, and copies files while retaining their original workspace-relative paths:

def _filegroup_to_directory_impl(ctx):
    # Declare the output directory
    out_dir = ctx.actions.declare_directory(ctx.attr.name)
    input_files = ctx.files.srcs

    # Shell script to handle directory creation and file copying
    copy_script = """
# Create the root output directory
mkdir -p "$1"

# Iterate over all input files
for src_file in "${@:2}"; do
    # Extract the workspace-relative path (strip leading ./ if present)
    rel_path="${src_file#./}"
    # Build the full target path in the output directory
    target_path="$1/$rel_path"
    
    # Automatically create parent directories for the target file
    mkdir -p "$(dirname "$target_path")"
    
    # Copy the file to its target location
    cp "$src_file" "$target_path"
done
"""

    # Execute the script with the output directory and input files as arguments
    ctx.actions.run_shell(
        inputs=input_files,
        outputs=[out_dir],
        arguments=[out_dir.path] + [f.path for f in input_files],
        command=copy_script,
    )

    # Return the output directory as the rule's result
    return [DefaultInfo(files=depset([out_dir]))]

# Define the rule for use in your BUILD files
filegroup_to_directory = rule(
    implementation=_filegroup_to_directory_impl,
    attrs={
        "srcs": attr.label_list(allow_files=True, mandatory=True),
    },
)

How This Works #

• Directory Structure Preservation: Each input File in Bazel has a path attribute that includes its workspace-relative location. The script extracts this path, creates any necessary parent directories (via mkdir -p "$(dirname "$target_path")"), then copies the file to the matching location in the output directory.
• No Tar Overhead: This skips the pkg_tar and untar steps entirely, reducing build time and avoiding unnecessary intermediate artifacts.
• Flexibility: You can extend this script easily—for example, add cp -a to preserve file permissions/symlinks if needed, or modify path handling if you want to adjust how directories are mapped.

Usage in BUILD Files #

filegroup(
    name = "my_files",
    srcs = glob(["src/**/*", "docs/*.md"]),
)

# Convert the filegroup to a declare_directory
filegroup_to_directory(
    name = "my_files_dir",
    srcs = [":my_files"],
)

This approach is far cleaner than manual per-file copying and avoids the inefficiency of tar/unpack cycles. It’s lightweight, maintainable, and keeps your build pipeline straightforward.

内容的提问来源于stack exchange，提问作者Yanjun Zhu