"""Supply backup functionality for gludb. Note that we mainly provide a way for
users to create a small script that backs up their data to S3. We are assuming
that the S3 bucket used will be configured for archival (either deletion or
archival to Glacier)
"""

import sys
import os
import pkgutil
import tarfile

from importlib import import_module
from inspect import getmembers, isclass, getmro
from tempfile import NamedTemporaryFile

from boto.s3.connection import S3Connection, OrdinaryCallingFormat
from boto.s3.key import Key

from .utils import now_field
from .config import get_mapping
from .data import Storable
from .simple import DBObject, Field

# TODO: need the back to check for previous backups and only run if enough time
#       has elapsed. Best way is probably: store json for backup after
#       success. Then we read it back to get date/time of last backup


def is_backup_class(cls):
    """Return true if given class supports back up. Currently this means a
    gludb.data.Storable-derived class that has a mapping as defined in
    gludb.config"""
    return True if (
        isclass(cls) and
        issubclass(cls, Storable) and
        get_mapping(cls, no_mapping_ok=True)
    ) else False


def backup_name(cls):
    """Return a usable name for the data stored for a class"""
    return cls.__name__ + '--' + cls.get_table_name()

# write_line is used below for Python 2&3 compatibility. We provide strip_line
# as a convenience for people reading backups (since it includes implicit
# encoding)
if sys.version_info >= (3, 0):
    def write_line(file_obj, line):
        file_obj.write(bytes(str(line) + '\n', 'UTF-8'))

    def strip_line(line):
        return str(line, 'UTF-8').strip()
else:
    def write_line(file_obj, line):
        file_obj.write(str(line))
        file_obj.write('\n')

    def strip_line(line):
        return str(line).strip()

# Maintain only one copy of the doc string for functions that have different
# versions for Python 2 and 3
write_line.__doc__ = """Write the given line to the given file-like
object with a terminating linefeed character. UTF-8 encoding is assumed"""

strip_line.__doc__ = """Return a stripped (trimmed) version of the line read
back from a file assuming that the line was originally written with the
write_line function UTF-8 encoding is assumed"""


# Turns out the library qualname doesn't handle annotated classes and we need
# Python 2 for gcd docs .... so we'll force-annotated Backup below
class Backup(object):
    """This is the main interface to gludb.backup functionality. When creating
    the instance must specify a bucket_name for backups. You may optionally
    specify a name, AWS access ID, and AWS secret key. Note that if you don't
    specify AWS credentials the environment variables AWS_ACCESS_KEY_ID and
    AWS_SECRET_ACCESS_KEY will checked. If those aren't defined, then empty
    strings will be used"""

    name = Field('backup')
    timestamp = Field(now_field)
    aws_access_key = Field('')
    aws_secret_key = Field('')
    bucketname = Field('')
    backup_log = Field(list)

    def setup(self, *args, **kwrds):
        # We don't ever store this name-to-class mapping
        self.classes = dict()

        # If they didn't manually specify AWS creds when we were created,
        # then we read them from the AWS-documented environ variables that
        # boto would use anyway
        if not self.aws_access_key:
            self.aws_access_key = os.environ.get('AWS_ACCESS_KEY_ID', '')
        if not self.aws_secret_key:
            self.aws_secret_key = os.environ.get('AWS_SECRET_ACCESS_KEY', '')

    def add_package(
        self,
        pkg_name,
        recurse=True,
        include_bases=True,
        parent_pkg=None
    ):
        """Add all classes to the backup in the specified package (including
        all modules and all sub-packages) for which is_backup_class returns
        True. Note that self.add_class is used, so base classes will added as
        well.

        Parameters:
        * pkg_name - a string representing the package name. It may be
          relative _if_ parent_pkg is supplied as well
        * recurse - (default value of True) if False, sub-packages will _not_
          be examined
        * include_bases - (default value of True) is passed directly to
          add_class for every class added
        * parent_pkg - a string representing the parent package of the relative
          package specified in pkg_name. Note that you should specify
          parent_pkg _only_ if pkg_name should be interpreted as relative

        An an example of both relative and absolute package imports, these
        are equivalent:

        ````
        backup.add_package('toppackage.subpackage')
        backup.add_package('subpackage', parent_pkg='toppackage')
        ````
        """
        if parent_pkg:
            pkg = import_module('.' + pkg_name, parent_pkg)
        else:
            pkg = import_module(pkg_name)

        for module_loader, name, ispkg in pkgutil.walk_packages(pkg.__path__):
            if not ispkg:
                # Module
                mod = import_module('.' + name, pkg_name)
                for name, member in getmembers(mod):
                    if is_backup_class(member):
                        self.add_class(member, include_bases=include_bases)
            elif recurse:
                # Package and we're supposed to recurse
                self.add_package(
                    pkg_name + '.' + name,
                    recurse=True,
                    include_bases=include_bases,
                    parent_pkg=parent_pkg
                )

    def add_class(self, cls, include_bases=True):
        """Add the specified class (which should be a class object, _not_ a
        string). By default all base classes for which is_backup_class returns
        True will also be added. `include_bases=False` may be spcified to
        suppress this behavior. The total number of classes added is returned.
        Note that if is_backup_class does not return True for the class object
        passed in, 0 will be returned. If you specify include_bases=False, then
        the maximum value that can be returned is 1."""
        if not is_backup_class(cls):
            return 0

        added = 0

        cls_name = backup_name(cls)
        if cls_name not in self.classes:
            self.classes[cls_name] = cls
            self.log("Added class for backup: %s", cls_name)
            added = 1

        if include_bases:
            for candidate_cls in getmro(cls):
                if is_backup_class(cls):
                    # Note that we don't keep recursing on base classes
                    added += self.add_class(candidate_cls, include_bases=False)

        return added

    def log(self, entry, *args):
        """Append the string supplied to the log (a list of strings). If
        additional arguments are supplied, then first string is assumed to be
        a format string and the other args are used for string interpolation.
        For instance `backup.log("%d + %d == %d", 1, 1, 2)` would result in the
        string `'1 + 1 == 2'` being logged"""
        if args:
            entry = entry % args
        self.backup_log.append(entry)

    def run_backup(self):
        """The actual backup is performed. The data for all added classes is
        extracted and written to a file per class where each line (terminated
        by a line feed character) is the JSON representing a single object.
        Those files are all archived in a single gzip'ed tarball which is
        stored in the AWS S3 bucket specified when the current instance of
        Backup was created"""
        self.log("Starting backup at %s", now_field())
        self.log("Backup config object created at %s", self.timestamp)

        # Make sure we're good to go
        for fld in ['aws_access_key', 'aws_secret_key', 'bucketname']:
            val = getattr(self, fld, None)
            if not val:
                self.log("Backup cannot start: %s is a required field", fld)
                raise ValueError(self.backup_log[-1])

        # Start the compressed tarball our data is stored in
        backup_file = NamedTemporaryFile(suffix=".tar.gz")
        backup_tarfile = tarfile.open(fileobj=backup_file, mode='w:gz')

        for cls_name, cls in self.classes.items():
            self.log("Backing up %s", cls_name)

            rec_count = 0

            with NamedTemporaryFile() as record_file:
                for rec in cls.find_all():
                    write_line(record_file, rec.to_data())
                    rec_count += 1

                record_file.flush()
                backup_tarfile.add(record_file.name, arcname=cls_name+'.json')

            self.log("%s => %d records backed up", cls_name, rec_count)

        # Finalize archive
        backup_tarfile.close()
        backup_file.flush()
        backup_size = os.stat(backup_file.name)[6]

        # Figure out key name for archived file
        key_name = ('Backup_' + now_field() + '.tar.gz').replace(':', '_')

        # upload archive to s3
        if os.environ.get('DEBUG', False) or os.environ.get('travis', False):
            # Local or CI - connect to our mock s3 service
            conn = S3Connection(
                '', '',
                is_secure=False, port=8888, host='localhost',
                calling_format=OrdinaryCallingFormat()
            )
        else:
            conn = S3Connection(self.aws_access_key, self.aws_secret_key)

        bucket = conn.get_bucket(self.bucketname)
        key = Key(bucket)
        key.key = key_name

        self.log(
            "Sending %s [size=%d bytes] with key name %s",
            backup_file.name,
            backup_size,
            key_name
        )

        # TODO: should probably look into a multi-part upload for larger backup
        key.set_contents_from_filename(backup_file.name)
        self.log("Sent %s", backup_file.name)

        # All done
        backup_file.close()
        self.log("Backup completed")

        # return the bucket name and key name for the completed backup
        return self.bucketname, key_name

# TODO: return to an annotation when we can use Python3 for our doc generation
Backup = DBObject(table_name='BackupHistory')(Backup)