toolchain: allow to configure build_tar target (#1937)

python/cpython#18080 hasn't been back-ported to python3.8, which happens to be
the default interpreter version in a lot of popular base docker images.

This change in python's tarfile implementation affects the behavior of
https://github.com/bazelbuild/rules_docker/blob/bdea23404dd256c7ae2f8e02d346428044cc7d91/container/archive.py#L144
used by a popular container_image rule. This means that images built by bazel
in environments with different python3 versions will result in different image
digests.

Until the hermeticity of python toolchains is guaranteed, it may be useful to
have a workaround for users who can provide their own build_tar tool, akin to
https://github.com/kubernetes/repo-infra/blob/72a6f5d05659f7d255b51f152e0725adfe970718/tools/build_tar/buildtar.go

Author: dataoleg

README.md

@@ -125,6 +125,8 @@ load("@io_bazel_rules_docker//toolchains/docker:toolchain.bzl",
 )
 docker_toolchain_configure(
   name = "docker_config",
+  # OPTIONAL: Bazel target for the build_tar tool, must be compatible with build_tar.py
+  build_tar_target="<enter absolute path (i.e., must start with repo name @...//:...) to an executable build_tar target>",
   # OPTIONAL: Path to a directory which has a custom docker client config.json.
   # See https://docs.docker.com/engine/reference/commandline/cli/#configuration-files
   # for more details.

container/layer.bzl

@@ -102,7 +102,10 @@ def build_layer(
     """
     toolchain_info = ctx.toolchains["@io_bazel_rules_docker//toolchains/docker:toolchain_type"].info
     layer = output_layer
-    build_layer_exec = ctx.executable.build_layer
+    if toolchain_info.build_tar_target:
+        build_layer_exec = toolchain_info.build_tar_target.files_to_run.executable
+    else:
+        build_layer_exec = ctx.executable.build_layer
     args = ctx.actions.args()
     args.add(layer, format = "--output=%s")
     args.add(directory, format = "--directory=%s")

testing/default_toolchain/WORKSPACE

@@ -26,6 +26,10 @@ load(
     "local_tool",
 )
 
+local_tool(
+    name = "build_tar",
+)
+
 local_tool(
     name = "xz",
 )
@@ -37,6 +41,7 @@ load(
 
 docker_toolchain_configure(
     name = "docker_config",
+    build_tar_target = "@build_tar//:build_tar",
     xz_target = "@xz//:xz",
 )
 

toolchains/docker/BUILD.tpl

@@ -21,6 +21,7 @@ load("@io_bazel_rules_docker//toolchains/docker:toolchain.bzl", "docker_toolchai
 docker_toolchain(
     name = "toolchain",
     client_config = "%{DOCKER_CONFIG}",
+    %{BUILD_TAR_ATTR}
     %{GZIP_ATTR}
     %{TOOL_ATTR}
     docker_flags = ["%{DOCKER_FLAGS}"],

toolchains/docker/toolchain.bzl

@@ -18,6 +18,7 @@ This module defines docker toolchain rules
 DockerToolchainInfo = provider(
     doc = "Docker toolchain rule parameters",
     fields = {
+        "build_tar_target": "Optional Bazel target for the build_tar tool",
         "client_config": "A custom directory for the docker client " +
                          "config.json. If DOCKER_CONFIG is not specified, " +
                          "the value of the DOCKER_CONFIG environment variable " +
@@ -41,6 +42,7 @@ DockerToolchainInfo = provider(
 def _docker_toolchain_impl(ctx):
     toolchain_info = platform_common.ToolchainInfo(
         info = DockerToolchainInfo(
+            build_tar_target = ctx.attr.build_tar_target,
             docker_flags = ctx.attr.docker_flags,
             client_config = ctx.attr.client_config,
             gzip_path = ctx.attr.gzip_path,
@@ -58,6 +60,12 @@ def _docker_toolchain_impl(ctx):
 docker_toolchain = rule(
     implementation = _docker_toolchain_impl,
     attrs = {
+        "build_tar_target": attr.label(
+            allow_files = True,
+            doc = "Bazel target for the build_tar tool.",
+            cfg = "host",
+            executable = True,
+        ),
         "client_config": attr.string(
             default = "",
             doc = "A custom directory for the docker client config.json. If " +
@@ -136,13 +144,18 @@ def _toolchain_configure_impl(repository_ctx):
     docker_flags = []
     docker_flags += repository_ctx.attr.docker_flags
 
+    build_tar_attr = ""
+    if repository_ctx.attr.build_tar_target:
+        build_tar_attr = "build_tar_target = \"%s\"," % repository_ctx.attr.build_tar_target
+
     # If client_config is not set we need to pass an empty string to the
     # template.
     client_config = repository_ctx.attr.client_config or ""
     repository_ctx.template(
         "BUILD",
         Label("@io_bazel_rules_docker//toolchains/docker:BUILD.tpl"),
         {
+            "%{BUILD_TAR_ATTR}": "%s" % build_tar_attr,
             "%{DOCKER_CONFIG}": "%s" % client_config,
             "%{DOCKER_FLAGS}": "%s" % "\", \"".join(docker_flags),
             "%{TOOL_ATTR}": "%s" % tool_attr,
@@ -167,6 +180,13 @@ def _toolchain_configure_impl(repository_ctx):
 # Repository rule to generate a docker_toolchain target
 toolchain_configure = repository_rule(
     attrs = {
+        "build_tar_target": attr.label(
+            executable = True,
+            cfg = "host",
+            allow_files = True,
+            mandatory = False,
+            doc = "The bazel target for the build_tar tool.",
+        ),
         "client_config": attr.string(
             mandatory = False,
             doc = "A custom directory for the docker client " +