dulwich/dulwich/gc.py

"""Git garbage collection implementation."""

import logging
import os
import time
from collections import deque
from collections.abc import Callable
from dataclasses import dataclass, field
from typing import TYPE_CHECKING, Optional

from dulwich.object_store import (
    BaseObjectStore,
    DiskObjectStore,
)
from dulwich.objects import Commit, ObjectID, Tag, Tree
from dulwich.refs import RefsContainer

if TYPE_CHECKING:
    from .config import Config
    from .repo import BaseRepo, Repo


DEFAULT_GC_AUTO = 6700
DEFAULT_GC_AUTO_PACK_LIMIT = 50


@dataclass
class GCStats:
    """Statistics from garbage collection."""

    pruned_objects: set[bytes] = field(default_factory=set)
    bytes_freed: int = 0
    packs_before: int = 0
    packs_after: int = 0
    loose_objects_before: int = 0
    loose_objects_after: int = 0


def find_reachable_objects(
    object_store: BaseObjectStore,
    refs_container: RefsContainer,
    include_reflogs: bool = True,
    progress: Callable[[str], None] | None = None,
) -> set[bytes]:
    """Find all reachable objects in the repository.

    Args:
        object_store: Object store to search
        refs_container: Reference container
        include_reflogs: Whether to include reflog entries
        progress: Optional progress callback

    Returns:
        Set of reachable object SHAs
    """
    reachable = set()
    pending: deque[ObjectID] = deque()

    # Start with all refs
    for ref in refs_container.allkeys():
        try:
            sha = refs_container[ref]  # This follows symbolic refs
            if sha and sha not in reachable:
                pending.append(sha)
                reachable.add(sha)
        except KeyError:
            # Broken ref
            if progress:
                progress(f"Warning: Broken ref {ref.decode('utf-8', 'replace')}")
            continue

    # TODO: Add reflog support when reflog functionality is available

    # Walk all reachable objects
    while pending:
        sha = pending.popleft()

        if progress:
            progress(f"Checking object {sha.decode('ascii', 'replace')}")

        try:
            obj = object_store[sha]
        except KeyError:
            continue

        # Add referenced objects
        if isinstance(obj, Commit):
            # Tree
            if obj.tree not in reachable:
                pending.append(obj.tree)
                reachable.add(obj.tree)
            # Parents
            for parent in obj.parents:
                if parent not in reachable:
                    pending.append(parent)
                    reachable.add(parent)
        elif isinstance(obj, Tree):
            # Tree entries
            for entry in obj.items():
                assert entry.sha is not None
                if entry.sha not in reachable:
                    pending.append(entry.sha)
                    reachable.add(entry.sha)
        elif isinstance(obj, Tag):
            # Tagged object
            if obj.object[1] not in reachable:
                pending.append(obj.object[1])
                reachable.add(obj.object[1])

    return reachable


def find_unreachable_objects(
    object_store: BaseObjectStore,
    refs_container: RefsContainer,
    include_reflogs: bool = True,
    progress: Callable[[str], None] | None = None,
) -> set[bytes]:
    """Find all unreachable objects in the repository.

    Args:
        object_store: Object store to search
        refs_container: Reference container
        include_reflogs: Whether to include reflog entries
        progress: Optional progress callback

    Returns:
        Set of unreachable object SHAs
    """
    reachable = find_reachable_objects(
        object_store, refs_container, include_reflogs, progress
    )

    unreachable = set()
    for sha in object_store:
        if sha not in reachable:
            unreachable.add(sha)

    return unreachable


def prune_unreachable_objects(
    object_store: DiskObjectStore,
    refs_container: RefsContainer,
    grace_period: int | None = None,
    dry_run: bool = False,
    progress: Callable[[str], None] | None = None,
) -> tuple[set[bytes], int]:
    """Remove unreachable objects from the repository.

    Args:
        object_store: Object store to prune
        refs_container: Reference container
        grace_period: Grace period in seconds (objects newer than this are kept)
        dry_run: If True, only report what would be deleted
        progress: Optional progress callback

    Returns:
        Tuple of (set of pruned object SHAs, total bytes freed)
    """
    unreachable = find_unreachable_objects(
        object_store, refs_container, progress=progress
    )

    pruned = set()
    bytes_freed = 0

    for sha in unreachable:
        try:
            obj = object_store[sha]

            # Check grace period
            if grace_period is not None:
                try:
                    mtime = object_store.get_object_mtime(sha)
                    age = time.time() - mtime
                    if age < grace_period:
                        if progress:
                            progress(
                                f"Keeping {sha.decode('ascii', 'replace')} (age: {age:.0f}s < grace period: {grace_period}s)"
                            )
                        continue
                except KeyError:
                    # Object not found, skip it
                    continue

            if progress:
                progress(f"Pruning {sha.decode('ascii', 'replace')}")

            # Calculate size before attempting deletion
            obj_size = len(obj.as_raw_string())

            if not dry_run:
                object_store.delete_loose_object(sha)

            # Only count as pruned if we get here (deletion succeeded or dry run)
            pruned.add(sha)
            bytes_freed += obj_size

        except KeyError:
            # Object already gone
            pass
        except OSError as e:
            # File system errors during deletion
            if progress:
                progress(f"Error pruning {sha.decode('ascii', 'replace')}: {e}")
    return pruned, bytes_freed


def garbage_collect(
    repo: "Repo",
    auto: bool = False,
    aggressive: bool = False,
    prune: bool = True,
    grace_period: int | None = 1209600,  # 2 weeks default
    dry_run: bool = False,
    progress: Callable[[str], None] | None = None,
) -> GCStats:
    """Run garbage collection on a repository.

    Args:
        repo: Repository to garbage collect
        auto: Whether this is an automatic gc
        aggressive: Whether to use aggressive settings
        prune: Whether to prune unreachable objects
        grace_period: Grace period for pruning in seconds
        dry_run: If True, only report what would be done
        progress: Optional progress callback

    Returns:
        GCStats object with garbage collection statistics
    """
    stats = GCStats()

    object_store = repo.object_store
    refs_container = repo.refs

    # Count initial state
    stats.packs_before = len(list(object_store.packs))
    stats.loose_objects_before = object_store.count_loose_objects()

    # Find unreachable objects to exclude from repacking
    unreachable_to_prune = set()
    if prune:
        if progress:
            progress("Finding unreachable objects")
        unreachable = find_unreachable_objects(
            object_store, refs_container, progress=progress
        )

        # Apply grace period check
        for sha in unreachable:
            try:
                if grace_period is not None:
                    try:
                        mtime = object_store.get_object_mtime(sha)
                        age = time.time() - mtime
                        if age < grace_period:
                            if progress:
                                progress(
                                    f"Keeping {sha.decode('ascii', 'replace')} (age: {age:.0f}s < grace period: {grace_period}s)"
                                )
                            continue
                    except KeyError:
                        # Object not found, skip it
                        continue

                unreachable_to_prune.add(sha)
                obj = object_store[sha]
                stats.bytes_freed += len(obj.as_raw_string())
            except KeyError:
                pass

        stats.pruned_objects = unreachable_to_prune

    # Pack refs
    if progress:
        progress("Packing references")
    if not dry_run:
        repo.refs.pack_refs()

    # Delete loose unreachable objects
    if prune and not dry_run:
        for sha in unreachable_to_prune:
            if object_store.contains_loose(sha):
                try:
                    object_store.delete_loose_object(sha)
                except OSError:
                    pass

    # Repack everything, excluding unreachable objects
    # This handles both loose object packing and pack consolidation
    if progress:
        progress("Repacking repository")
    if not dry_run:
        if prune and unreachable_to_prune:
            # Repack excluding unreachable objects
            object_store.repack(exclude=unreachable_to_prune, progress=progress)
        else:
            # Normal repack
            object_store.repack(progress=progress)

    # Prune orphaned temporary files
    if progress:
        progress("Pruning temporary files")
    if not dry_run:
        object_store.prune(grace_period=grace_period)

    # Count final state
    stats.packs_after = len(list(object_store.packs))
    stats.loose_objects_after = object_store.count_loose_objects()

    return stats


def should_run_gc(repo: "BaseRepo", config: Optional["Config"] = None) -> bool:
    """Check if automatic garbage collection should run.

    Args:
        repo: Repository to check
        config: Configuration to use (defaults to repo config)

    Returns:
        True if GC should run, False otherwise
    """
    # Check environment variable first
    if os.environ.get("GIT_AUTO_GC") == "0":
        return False

    # Check programmatic disable flag
    if getattr(repo, "_autogc_disabled", False):
        return False

    if config is None:
        config = repo.get_config()

    # Check if auto GC is disabled
    try:
        gc_auto = config.get(b"gc", b"auto")
        gc_auto_value = int(gc_auto)
    except KeyError:
        gc_auto_value = DEFAULT_GC_AUTO

    if gc_auto_value == 0:
        # Auto GC is disabled
        return False

    # Check loose object count
    object_store = repo.object_store
    if not isinstance(object_store, DiskObjectStore):
        # Can't count loose objects on non-disk stores
        return False

    loose_count = object_store.count_loose_objects()
    if loose_count >= gc_auto_value:
        return True

    # Check pack file count
    try:
        gc_auto_pack_limit = config.get(b"gc", b"autoPackLimit")
        pack_limit = int(gc_auto_pack_limit)
    except KeyError:
        pack_limit = DEFAULT_GC_AUTO_PACK_LIMIT

    if pack_limit > 0:
        pack_count = object_store.count_pack_files()
        if pack_count >= pack_limit:
            return True

    return False


def maybe_auto_gc(
    repo: "Repo",
    config: Optional["Config"] = None,
    progress: Callable[[str], None] | None = None,
) -> bool:
    """Run automatic garbage collection if needed.

    Args:
        repo: Repository to potentially GC
        config: Configuration to use (defaults to repo config)
        progress: Optional progress reporting callback

    Returns:
        True if GC was run, False otherwise
    """
    if not should_run_gc(repo, config):
        return False

    # Check for gc.log file - only for disk-based repos
    if not hasattr(repo, "controldir"):
        # For non-disk repos, just run GC without gc.log handling
        garbage_collect(repo, auto=True, progress=progress)
        return True

    gc_log_path = os.path.join(repo.controldir(), "gc.log")
    if os.path.exists(gc_log_path):
        # Check gc.logExpiry
        if config is None:
            config = repo.get_config()
        try:
            log_expiry = config.get(b"gc", b"logExpiry")
        except KeyError:
            # Default to 1 day
            expiry_seconds = 86400
        else:
            # Parse time value (simplified - just support days for now)
            if log_expiry.endswith((b".days", b".day")):
                days = int(log_expiry.split(b".")[0])
                expiry_seconds = days * 86400
            else:
                # Default to 1 day
                expiry_seconds = 86400

        stat_info = os.stat(gc_log_path)
        if time.time() - stat_info.st_mtime < expiry_seconds:
            # gc.log exists and is not expired - skip GC
            with open(gc_log_path, "rb") as f:
                logging.info(
                    "gc.log content: %s", f.read().decode("utf-8", errors="replace")
                )
            return False

    # TODO: Support gc.autoDetach to run in background
    # For now, run in foreground

    try:
        # Run GC with auto=True flag
        garbage_collect(repo, auto=True, progress=progress)

        # Remove gc.log on successful completion
        if os.path.exists(gc_log_path):
            try:
                os.unlink(gc_log_path)
            except FileNotFoundError:
                pass

        return True
    except OSError as e:
        # Write error to gc.log
        with open(gc_log_path, "wb") as f:
            f.write(f"Auto GC failed: {e}\n".encode())
        # Don't propagate the error - auto GC failures shouldn't break operations
        return False