Skip to content

Latest commit

 

History

History
152 lines (126 loc) · 7.87 KB

customizable_phase.md

File metadata and controls

152 lines (126 loc) · 7.87 KB

Customizable Phase

Contents

Overview

Phases increase configurability. Rule implementations are defined as a list of phases. Each phase defines a specific step, which helps breaking up implementation into smaller and more readable groups. Some phases are independent from others, which means the order doesn't matter. However, some phases depend on outputs of previous phases, in this case, we should make sure it meets all the prerequisites before executing phases.

The biggest benefit of phases is that it is customizable. If default phase A is not doing what you expect, you may switch it with your self-defined phase A. One use case is to write your own compilation phase with your favorite Scala compiler. You may also extend the default phase list for more functionalities. One use case is to check the Scala format.

Who needs customizable phase

Customizable phase is an advanced feature for people who want the rules to do more. If you are an experienced Bazel rules developer, we make this powerful API public for you to do custom work without impacting other consumers. If you have no experience on writing Bazel rules, we are happy to help but be aware it may be frustrating at first.

If you don't need to customize your rules and just need the default setup to work correctly, then just load the following file for default rules:

load("@io_bazel_rules_scala//scala:scala.bzl")

Otherwise read on:

As a consumer

You need to load the following 2 files:

load("@io_bazel_rules_scala//scala:advanced_usage/providers.bzl", "ScalaRulePhase")
load("@io_bazel_rules_scala//scala:advanced_usage/scala.bzl", "make_scala_binary")

ScalaRulePhase is a phase provider to pass in custom phases. Rules with make_ prefix, like make_scala_binary, are customizable rules. make_<RULE_NAME>s take a dictionary as input. It currently supports appending attrs and outputs to default rules, as well as modifying the phase list.

For example:

ext_add_custom_phase = {
    "attrs": {
        "custom_content": attr.string(
            default = "This is custom content",
        ),
    },
    "outputs": {
        "custom_output": "%{name}.custom-output",
    },
    "phase_providers": [
        "//custom/phase:phase_custom_write_extra_file",
    ],
}

custom_scala_binary = make_scala_binary(ext_add_custom_phase)

make_<RULE_NAME>s append attrs and outputs to the default rule definitions. All items in attrs can be accessed by ctx.attr, and all items in outputs can be accessed by ctx.outputs. phase_providers takes a list of targets which define how you want to modify phase list.

def _add_custom_phase_singleton_implementation(ctx):
    return [
        ScalaRulePhase(
            custom_phases = [
                ("last", "", "custom_write_extra_file", phase_custom_write_extra_file),
            ],
        ),
    ]

add_custom_phase_singleton = rule(
    implementation = _add_custom_phase_singleton_implementation,
)

add_custom_phase_singleton is a rule solely to pass in custom phases using ScalaRulePhase. The custom_phases field in ScalaRulePhase takes a list of tuples. Each tuple has 4 elements:

(relation, peer_name, phase_name, phase_function)
  • relation: the position to add a new phase
  • peer_name: the existing phase to compare the position with
  • phase_name: the name of the new phase, also used to access phase information
  • phase_function: the function of the new phase

There are 5 possible relations:

  • ^ or first
  • $ or last
  • - or before
  • + or after
  • = or replace

The symbols and words are interchangable. If first or last is used, it puts your custom phase at the beginning or the end of the phase list, peer_name is not needed.

Then you have to call the rule in a BUILD

add_custom_phase_singleton(
    name = "phase_custom_write_extra_file",
    visibility = ["//visibility:public"],
)

You may now see phase_providers in ext_add_custom_phase is pointing to this target.

The last step is to write the function of the phase. For example:

def phase_custom_write_extra_file(ctx, p):
    ctx.actions.write(
        output = ctx.outputs.custom_output,
        content = ctx.attr.custom_content,
    )

Every phase has 2 arguments, ctx and p. ctx gives you access to the fields defined in rules. p is the global provider, which contains information from initial state as well as all the previous phases. You may access the information from previous phases by p.<PHASE_NAME>.<FIELD_NAME>. For example, if the previous phase, said phase_jar with phase name jar, returns a struct

def phase_jar(ctx, p):
    # Some works to get the jars
    return struct(
        class_jar = class_jar,
        ijar = ijar,
    )

You are able to access information like p.jar.class_jar in phase_custom_write_extra_file. You can provide the information for later phases in the same way, then they can access it by p.custom_write_extra_file.<FIELD_NAME>.

You should be able to define the files above entirely in your own workspace without making change to the bazelbuild/rules_scala. If you believe your custom phase will be valuable to the community, please refer to As a contributor. Pull requests are welcome.

As a contributor

Besides the basics in As a consumer, the followings help you understand how phases are setup if you plan to contribute to bazelbuild/rules_scala.

These are the relevant files

  • scala/private/phases/api.bzl: the API of executing and modifying the phase list
  • scala/private/phases/phases.bzl: re-expose phases for convenience
  • scala/private/phases/phase_<PHASE_NAME>.bzl: all the phase definitions

Currently phase architecture is used by 7 rules:

  • scala_library
  • scala_macro_library
  • scala_library_for_plugin_bootstrapping
  • scala_binary
  • scala_test
  • scala_junit_test
  • scala_repl

If you need to expose providers to downstream targets you need to return a dict of providers (provider-name to provider instance) from your phase under the external_providers attribute.

If you need to override a provider returned by a previous phase you can adjust your phase to be after it and return the same key from your phase and it will override it. Note you probably have a good reason to override since you're meddling with the public return value of a different phase.

In each of the rule implementations, it calls run_phases and returns the accumulated values of the external_providers dict declared by the phases.

To make a new phase, you have to define a new phase_<PHASE_NAME>.bzl in scala/private/phases/. Function definition should have 2 arguments, ctx and p. You may expose the information for later phases by returning a struct. In some phases, there are multiple phase functions since different rules may take slightly different input arguemnts. You may want to re-expose the phase definition in scala/private/phases/phases.bzl, so it's more convenient to access in rule files.

In the rule implementations, put your new phase in builtin_customizable_phases list. The phases are executed sequentially, the order matters if the new phase depends on previous phases.

Phase naming convention

Files in scala/private/phases/

  • phase_<PHASE_NAME>.bzl: phase definition file

Function names in phase_<PHASE_NAME>.bzl

  • phase_<PHASE_NAME>_<RULE_NAME>: function with custom inputs of specific rule
  • phase_<PHASE_NAME>_common: function without custom inputs
  • _phase_<PHASE_NAME>_default: private function that takes _args for custom inputs
  • _phase_<PHASE_NAME>: private function with the actual logic

See phase_compile.bzl for example.