feat: use full minecraft version to get pack_format

[edayot]

This PR fix #471 by creating a database of minecraft version in a resources folder (like what mecha does for command trees) and using it to get the pack_format.

Disadvantages :

• This does add a new thing to update beet on a new minecraft release.
• This transform the two old classvar to properties, they act the same with an object of that class, but breaks few thing that were using the class itself. fixed

This PR comes with a PR for mecha with new command tree.

[rx-dev]

mostly approved, makes sense to me

[rx-dev]

not sure if the logic for the file should be here, unsure where to put it ..

[edayot]

yes, i didn't want to read the json every time the pack format is accessed

[rx-dev]

i guess this file is eager instead of lazy. which is unlike most other files in beet. unsure how i feel about that

[edayot]

i can change it to be lazy, like creating a cached function that return a object with all pack format registry, it would look like this :

class RegistryHolder(BaseModel):
    pack_format_registry: list[PackFormatRegistry]
    data_pack_format_registry: dict[Tuple[int, ...], int]
    resource_pack_format_registry: dict[Tuple[int, ...], int]

@cache
def get_pack_format_registry() -> RegistryHolder:
    """Get the pack format registry."""
    data = json.loads(
        files("beet.resources").joinpath(f"pack_format_registry.json").read_text()
    )
    pack_format_registry: list[PackFormatRegistry] = []
    for item in data:
        pack_format_registry.append(PackFormatRegistry.model_validate(item))


    data_pack_format_registry = {
        split_version(x.id): x.data_pack_version
        for x in pack_format_registry
        if x.type == "release"
    }
    resource_pack_format_registry = {
        split_version(x.id): x.resource_pack_version
        for x in pack_format_registry
        if x.type == "release"
    }
    return RegistryHolder(
        pack_format_registry=pack_format_registry,
        data_pack_format_registry=data_pack_format_registry,
        resource_pack_format_registry=resource_pack_format_registry,
    )

EDIT : copilot suggested this :

"""
Pack format registry resource from https://raw.githubusercontent.com/misode/mcmeta/refs/heads/summary/versions/data.json
"""

__all__ = [
    "PackFormatRegistry",
    "pack_format_registry",
    "data_pack_format_registry",
    "resource_pack_format_registry",
]
from importlib.resources import files
import json
from functools import lru_cache
from pydantic import BaseModel
from beet.core.utils import split_version


class PackFormatRegistry(BaseModel):
    id: str
    name: str
    release_target: str | None
    type: str
    stable: bool
    data_version: int
    protocol_version: int
    data_pack_version: int
    resource_pack_version: int
    build_time: str
    release_time: str
    sha1: str


@lru_cache(maxsize=1)
def pack_format_registry() -> list[PackFormatRegistry]:
    """Get the pack format registry, loaded lazily."""
    data = json.loads(
        files("beet.resources").joinpath("pack_format_registry.json").read_text()
    )
    return [PackFormatRegistry.model_validate(item) for item in data]


@lru_cache(maxsize=1)
def data_pack_format_registry() -> dict[tuple[int, ...], int]:
    """Get the data pack format registry, loaded lazily."""
    return {
        split_version(x.id): x.data_pack_version
        for x in pack_format_registry()
        if x.type == "release"
    }


@lru_cache(maxsize=1)
def resource_pack_format_registry() -> dict[tuple[int, ...], int]:
    """Get the resource pack format registry, loaded lazily."""
    return {
        split_version(x.id): x.resource_pack_version
        for x in pack_format_registry()
        if x.type == "release"
    }

[rx-dev]

i also thinks it makes sense to pull the file remotely and cache it maybe? we do that for vanilla but that's also an optional thing versus this is something that people will use often.

thanks for providing both implementations

[edayot]

But pulling it directly from misode/mcmeta makes it impossible to use beet offline, i like the idea to use misode/mcmeta file but we still needs to add a local registry as a backup strategy

[misode]

I don't personally like the idea of relying on misode/mcmeta directly during runtime.

[misode]

Beet technically supported down to 1.13 before. I'm not sure if we care about this regression?

[edayot]

I wasn't taking into account this, beet technically support MC 1.6 pack_format 1. Might be needed to hardcode even more values.

[misode]

This is a pretty large file and has a lot of unused data. Do we want to try to generate this file (in github actions?) and preprocess it to only keep the relevant fields?

[edayot]

Might be a good idea, i was taking this file because it was very convenient.

I was sorting informations i needed in python, but realistically we don't need :

• snapshots (as element in the list)
• release_target
• type
• stable
• build_time
• release_time
• sha1

[misode]

Is it necessary (or even a good idea) to support snapshots? Mecha and other beet features might not necessarily support these versions. Maybe we should just hardcode a list of pack formats for all the release versions?

[edayot]

As discuss in others comments, the plan wasn't to support snapshot.

[misode]

Was this tested? Does this work without any other changes? I am guessing the map created here will have tuples with mixed length 2 and 3? Before they were always tuples of length 2.

[edayot]

The return type of split_version was always Tuple[int, ...], so any tuple of any length. Also i tested this version against mecha, bolt and my repos and found no breaking changes.

[misode]

This illustrates that this PR is a breaking change. Specifying minecraft: 1.18 in the config now specifies exactly 1.18, not the last 1.18.x. I know we discussed this a couple months ago in a google docs, but we'll need to be very careful merging this if people rely on this config.

[edayot]

I don't think it's a big deal personally but i account for the issue.

Here's the ggdoc : https://docs.google.com/document/d/1m0niC2I80gMS_yRjQFP9PglhxztAs0b8O5y7XshXp-k/edit?tab=t.0#heading=h.qkfryuqiybl2

[edayot]

Your last comment reminded me of the existence of this ggdoc, it would be a good idea to implement this in another PR.

[edayot]

Might be a good idea, i was taking this file because it was very convenient.

I was sorting informations i needed in python, but realistically we don't need :

• snapshots (as element in the list)
• release_target
• type
• stable
• build_time
• release_time
• sha1

[edayot]

As discuss in others comments, the plan wasn't to support snapshot.

[edayot]

I wasn't taking into account this, beet technically support MC 1.6 pack_format 1. Might be needed to hardcode even more values.

[edayot]

The return type of split_version was always Tuple[int, ...], so any tuple of any length. Also i tested this version against mecha, bolt and my repos and found no breaking changes.

[edayot]

I don't think it's a big deal personally but i account for the issue.

Here's the ggdoc : https://docs.google.com/document/d/1m0niC2I80gMS_yRjQFP9PglhxztAs0b8O5y7XshXp-k/edit?tab=t.0#heading=h.qkfryuqiybl2