How to Cache and Share Operations

This guide shows you how to use anonymous targets in BXL to cache and share operations across different commands, improving bxl performance and reducing peak memory usage.

When to use caching

Use anonymous target caching if you need to:

1. Cache results of expensive Starlark computations
2. Share cached work across different parts of your bxl
3. Share cached work across different bxl commands
4. Reduce peak memory usage

Basic caching with anonymous targets

1. Define bxl anon target rule

my_anon = bxl.anon_rule(
    impl = _my_anon_impl,
    attrs = {
        "foo": attrs.int(),
        "bar": attrs.str(),
        ...
    },
)

You can find the supported attributes at the "Attributes" sections here

2. Define the anon target impl

def _my_anon_impl(bxl_ctx: bxl.Context, attrs: struct) -> list[Provider]
    # Your implementation here
    return [DefaultInfo(...), ...]

3. Create and resolve the anonymous target in your BXL script

You can use actions.anon_target to create one anon target or actions.anon_targets to create several anon targets.

def _bxl_main_impl(bxl_ctx: bxl.Context):
    ...
    actions = ctx.bxl_actions().actions

    # Create anonymous target
     promise = actions.anon_target(
        my_anon,
        attrs = {
            "foo": 42,
            "bar": "hello world",
        }
    ).promise

    # Resolve the anon target result
    result = ctx.resolve(actions, promise)

    # Use the anon target result
    ...

Now you have a anon target and the the output of this anon target will be cached using a cache key composed of its attributes, target platform, and any bxl script modifiers.

Examples

Examples can be found at 'tests/core/bxl/test_anon_bxl_data/anon_bxl.bxl' of buck2 repro folder.

Run such command at tests/core/bxl/test_anon_bxl_data to run the example bxl script

buck2 bxl anon_bxl.bxl:eval_anon_bxl

Further Reading

You can refer Anonymous Targets to learn more about it.