workinf_Blender_Wasm/my-tools/wbv2_local_preparation.py

"""Discover explicit local recipes and admit their output without model calls."""

import ast
import hashlib
import json
import re
import time
from pathlib import Path


def recipe_paths(root):
    root = Path(root).resolve()
    directory = root / "my-tools"
    if directory.is_symlink():
        return []
    return sorted(p for p in directory.glob("prepare_wbv2_*workflow.py")
                  if p.is_file() and not p.is_symlink())[:128]


def _declarations(path):
    if path.stat().st_size > 128 * 1024:
        return None
    tree = ast.parse(path.read_text(encoding="utf-8"))
    values = {}
    for node in tree.body:
        if isinstance(node, ast.Assign) and len(node.targets) == 1:
            target = node.targets[0]
            if isinstance(target, ast.Name) and target.id in {"JOB", "PREPARATION_INPUTS"}:
                values[target.id] = ast.literal_eval(node.value)
    job = values.get("JOB")
    inputs = values.get("PREPARATION_INPUTS", [])
    if not isinstance(job, str) or not re.fullmatch(r"[A-Z0-9][A-Z0-9_-]{0,127}", job):
        return None
    if not isinstance(inputs, list) or len(inputs) > 128 or not all(isinstance(p, str) for p in inputs):
        return None
    return job, inputs


def _local(root, relative):
    path = Path(relative)
    if path.is_absolute() or ".." in path.parts:
        raise ValueError("preparation input escapes repository")
    current = root
    for part in path.parts:
        current /= part
        if current.is_symlink():
            raise ValueError("preparation input is a symlink")
    return current


def _fingerprint(root, context, paths):
    files = {}
    for relative in sorted(set(paths)):
        path = _local(root, relative)
        files[relative] = hashlib.sha256(path.read_bytes()).hexdigest() if path.is_file() else None
    authority = {key: context[key] for key in ("task", "parentSha256")}
    return hashlib.sha256(json.dumps([authority, files], sort_keys=True).encode()).hexdigest()


def recipe_input_paths(root, jobs):
    wanted = {job["id"] for job in jobs}
    paths = set()
    for script in recipe_paths(root):
        paths.add(str(script.relative_to(root)))
        try:
            declaration = _declarations(script)
            if declaration and declaration[0] in wanted:
                for relative in declaration[1]:
                    _local(root, relative)
                    paths.add(relative)
        except (OSError, SyntaxError, ValueError, TypeError):
            continue
    return paths


def _failed_origin_owned(root, origin, prior, context, identifier, script):
    owned = prior.get("generatedOrigin")
    if (prior.get("state") != "blocked" or prior.get("job") != identifier
            or prior.get("script") != script or prior.get("task") != context["task"]
            or prior.get("parentSha256") != context["parentSha256"]
            or not isinstance(owned, dict)):
        return False
    try:
        relative = str(origin.relative_to(root))
        return (owned.get("path") == relative
                and _local(root, relative).is_file()
                and hashlib.sha256(origin.read_bytes()).hexdigest() == owned.get("sha256"))
    except (OSError, ValueError):
        return False


def prepare_local_contracts(runner, context, jobs):
    from wbv2_autopilot import queue_context, read_json, write_json
    from wbv2_task_dispatch import prepared_contract

    recipes = {}
    for script in recipe_paths(runner.root):
        try:
            declaration = _declarations(script)
        except (OSError, SyntaxError, ValueError, TypeError):
            continue
        if declaration:
            job, inputs = declaration
            recipes.setdefault(job, []).append((script, inputs))
    prepared = []
    for job in jobs:
        identifier = job["id"]
        choices = recipes.get(identifier, [])
        if not re.fullmatch(r"[A-Z0-9][A-Z0-9_-]{0,127}", identifier) or len(choices) != 1:
            continue
        # An existing origin belongs to the developer, including one being edited.
        # Admission errors must not cause its generator to overwrite those edits.
        origin = _local(runner.root, f"my-tools/local-workflows/{identifier}.json")
        if runner.local_contract_path(job):
            continue
        script, declared = choices[0]
        relative = str(script.relative_to(runner.root))
        inputs = [relative, context["parentManifest"], *job.get("ownedPaths", []), *declared]
        if len(inputs) > 256:
            continue
        try:
            fingerprint = _fingerprint(runner.root, context, inputs)
        except (OSError, ValueError):
            continue
        record_path = runner.directory / "preparation-attempts" / f"{identifier}.json"
        prior = read_json(record_path, {})
        if prior.get("fingerprint") == fingerprint:
            continue
        if runner.stop_requested or (runner.directory / "pause.request").exists():
            break
        if queue_context(runner.root) != context:
            raise InterruptedError("queue changed before local preparation")
        if origin.exists():
            if not _failed_origin_owned(runner.root, origin, prior, context, identifier, relative):
                continue
            origin.unlink()
        started = time.monotonic()
        record = {"schemaVersion": 1, "job": identifier, "script": relative,
                  "fingerprint": fingerprint, "state": "running"}
        record.update(task=context["task"], parentSha256=context["parentSha256"])
        write_json(record_path, record)
        proc = runner.run_process(["python3", relative], runner.root,
                                  runner.directory / "preparation" / f"{identifier}.log",
                                  min(120, runner.policy["checkTimeoutSeconds"]),
                                  phase="local_workflow_prepare")
        record.update(seconds=time.monotonic() - started, exitCode=proc["exitCode"])
        if queue_context(runner.root) != context:
            record["state"] = "authority_changed"
            write_json(record_path, record)
            raise InterruptedError("queue changed during local preparation")
        ok = proc["exitCode"] == 0 and not any(proc.get(key) for key in ("interrupted", "timedOut", "logLimit"))
        contract, findings = prepared_contract(runner.root, context, job) if ok else (None, [])
        if not contract:
            try:
                output = str(origin.relative_to(runner.root))
                if _local(runner.root, output).is_file():
                    record["generatedOrigin"] = {
                        "path": output,
                        "sha256": hashlib.sha256(origin.read_bytes()).hexdigest(),
                    }
            except (OSError, ValueError):
                pass
        record.update(state="prepared" if contract else "blocked", findings=findings)
        write_json(record_path, record)
        runner.event("local.prepared" if contract else "local.prepare_failed", job=identifier,
                     script=relative, seconds=record["seconds"], exitCode=proc["exitCode"])
        if contract:
            prepared.append(identifier)
        if proc.get("interrupted"):
            break
    return prepared