"""
Galaxy Metadata

"""

import copy
import cPickle
import json
import logging
import os
import shutil
import sys
import tempfile
import weakref

from os.path import abspath

import galaxy.model
from galaxy.util import listify, stringify_dictionary_keys, string_as_bool
from galaxy.util.odict import odict
from galaxy.util import in_directory
from galaxy.web import form_builder
from sqlalchemy.orm import object_session

log = logging.getLogger(__name__)

STATEMENTS = "__galaxy_statements__" #this is the name of the property in a Datatype class where new metadata spec element Statements are stored

class Statement( object ):
    """
    This class inserts its target into a list in the surrounding
    class.  the data.Data class has a metaclass which executes these
    statements.  This is how we shove the metadata element spec into
    the class.
    """
    def __init__( self, target ):
        self.target = target

    def __call__( self, *args, **kwargs ):
        class_locals = sys._getframe( 1 ).f_locals #get the locals dictionary of the frame object one down in the call stack (i.e. the Datatype class calling MetadataElement)
        statements = class_locals.setdefault( STATEMENTS, [] ) #get and set '__galaxy_statments__' to an empty list if not in locals dict
        statements.append( ( self, args, kwargs ) ) #add Statement containing info to populate a MetadataElementSpec

    @classmethod
    def process( cls, element ):
        for statement, args, kwargs in getattr( element, STATEMENTS, [] ):
            statement.target( element, *args, **kwargs ) #statement.target is MetadataElementSpec, element is a Datatype class


class MetadataCollection( object ):
    """
    MetadataCollection is not a collection at all, but rather a proxy
    to the real metadata which is stored as a Dictionary. This class
    handles processing the metadata elements when they are set and
    retrieved, returning default values in cases when metadata is not set.
    """
    def __init__(self, parent ):
        self.parent = parent
        #initialize dict if needed
        if self.parent._metadata is None:
            self.parent._metadata = {}

    def get_parent( self ):
        if "_parent" in self.__dict__:
            return self.__dict__["_parent"]()
        return None

    def set_parent( self, parent ):
        self.__dict__["_parent"] = weakref.ref( parent ) # use weakref to prevent a circular reference interfering with garbage collection: hda/lda (parent) <--> MetadataCollection (self) ; needs to be hashable, so cannot use proxy.
    parent = property( get_parent, set_parent )

    @property
    def spec( self ):
        return self.parent.datatype.metadata_spec

    def __iter__( self ):
        return self.parent._metadata.__iter__()

    def get( self, key, default=None ):
        try:
            return self.__getattr__( key ) or default
        except:
            return default

    def items(self):
        return iter( [ ( k, self.get( k ) ) for k in self.spec.iterkeys() ] )

    def __str__(self):
        return dict( self.items() ).__str__()

    def __nonzero__( self ):
        return bool( self.parent._metadata )

    def __getattr__( self, name ):
        if name in self.spec:
            if name in self.parent._metadata:
                return self.spec[name].wrap( self.parent._metadata[name], object_session( self.parent ) )
            return self.spec[name].wrap( self.spec[name].default, object_session( self.parent ) )
        if name in self.parent._metadata:
            return self.parent._metadata[name]

    def __setattr__( self, name, value ):
        if name == "parent":
            return self.set_parent( value )
        else:
            if name in self.spec:
                self.parent._metadata[name] = self.spec[name].unwrap( value )
            else:
                self.parent._metadata[name] = value

    def element_is_set( self, name ):
        return bool( self.parent._metadata.get( name, False ) )

    def get_html_by_name( self, name, **kwd ):
        if name in self.spec:
            rval = self.spec[name].param.get_html( value=getattr( self, name ), context=self, **kwd )
            if rval is None:
                return self.spec[name].no_value
            return rval

    def make_dict_copy( self, to_copy ):
        """Makes a deep copy of input iterable to_copy according to self.spec"""
        rval = {}
        for key, value in to_copy.items():
            if key in self.spec:
                rval[key] = self.spec[key].param.make_copy( value, target_context=self, source_context=to_copy )
        return rval

    def from_JSON_dict( self, filename ):
        dataset = self.parent
        log.debug( 'loading metadata from file for: %s %s' % ( dataset.__class__.__name__, dataset.id ) )
        JSONified_dict = json.load( open( filename ) )
        for name, spec in self.spec.items():
            if name in JSONified_dict:
                dataset._metadata[ name ] = spec.param.from_external_value( JSONified_dict[ name ], dataset )
            elif name in dataset._metadata:
                #if the metadata value is not found in our externally set metadata but it has a value in the 'old'
                #metadata associated with our dataset, we'll delete it from our dataset's metadata dict
                del dataset._metadata[ name ]

    def to_JSON_dict( self, filename ):
        #galaxy.model.customtypes.json_encoder.encode()
        meta_dict = {}
        dataset_meta_dict = self.parent._metadata
        for name, spec in self.spec.items():
            if name in dataset_meta_dict:
                meta_dict[ name ] = spec.param.to_external_value( dataset_meta_dict[ name ] )
        json.dump( meta_dict, open( filename, 'wb+' ) )

    def __getstate__( self ):
        return None #cannot pickle a weakref item (self._parent), when data._metadata_collection is None, it will be recreated on demand


class MetadataSpecCollection( odict ):
    """
    A simple extension of dict which allows cleaner access to items
    and allows the values to be iterated over directly as if it were a
    list.  append() is also implemented for simplicity and does not
    "append".
    """
    def __init__( self, dict = None ):
        odict.__init__( self, dict = None )

    def append( self, item ):
        self[item.name] = item

    def iter( self ):
        return self.itervalues()

    def __getattr__( self, name ):
        return self.get( name )

    def __repr__( self ):
        # force elements to draw with __str__ for sphinx-apidoc
        return ', '.join([ item.__str__() for item in self.iter() ])


class MetadataParameter( object ):
    def __init__( self, spec ):
        self.spec = spec

    def get_html_field( self, value=None, context=None, other_values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}
        return form_builder.TextField( self.spec.name, value=value )

    def get_html( self, value, context=None, other_values=None, **kwd ):
        """
        The "context" is simply the metadata collection/bunch holding
        this piece of metadata. This is passed in to allow for
        metadata to validate against each other (note: this could turn
        into a huge, recursive mess if not done with care). For
        example, a column assignment should validate against the
        number of columns in the dataset.
        """
        context = context or {}
        other_values = other_values or {}

        if self.spec.get("readonly"):
            return value
        if self.spec.get("optional"):
            checked = False
            if value: checked = "true"
            checkbox = form_builder.CheckboxField( "is_" + self.spec.name, checked=checked )
            return checkbox.get_html() + self.get_html_field( value=value, context=context, other_values=other_values, **kwd ).get_html()
        else:
            return self.get_html_field( value=value, context=context, other_values=other_values, **kwd ).get_html()

    def to_string( self, value ):
        return str( value )

    def make_copy( self, value, target_context = None, source_context = None ):
        return copy.deepcopy( value )

    @classmethod
    def marshal ( cls, value ):
        """
        This method should/can be overridden to convert the incoming
        value to whatever type it is supposed to be.
        """
        return value

    def validate( self, value ):
        """
        Throw an exception if the value is invalid.
        """
        pass

    def unwrap( self, form_value ):
        """
        Turns a value into its storable form.
        """
        value = self.marshal( form_value )
        self.validate( value )
        return value

    def wrap( self, value, session ):
        """
        Turns a value into its usable form.
        """
        return value

    def from_external_value( self, value, parent ):
        """
        Turns a value read from an external dict into its value to be pushed directly into the metadata dict.
        """
        return value

    def to_external_value( self, value ):
        """
        Turns a value read from a metadata into its value to be pushed directly into the external dict.
        """
        return value


class MetadataElementSpec( object ):
    """
    Defines a metadata element and adds it to the metadata_spec (which
    is a MetadataSpecCollection) of datatype.
    """
    def __init__( self, datatype,
                  name=None, desc=None, param=MetadataParameter, default=None, no_value = None,
                  visible=True, set_in_upload = False, **kwargs ):
        self.name = name
        self.desc = desc or name
        self.default = default
        self.no_value = no_value
        self.visible = visible
        self.set_in_upload = set_in_upload
        # Catch-all, allows for extra attributes to be set
        self.__dict__.update(kwargs)
        # set up param last, as it uses values set above
        self.param = param( self )
        # add spec element to the spec
        datatype.metadata_spec.append( self )

    def get( self, name, default=None ):
        return self.__dict__.get(name, default)

    def wrap( self, value, session ):
        """
        Turns a stored value into its usable form.
        """
        return self.param.wrap( value, session )

    def unwrap( self, value ):
        """
        Turns an incoming value into its storable form.
        """
        return self.param.unwrap( value )

    def __str__( self ):
        #TODO??: assuming param is the class of this MetadataElementSpec - add the plain class name for that
        spec_dict = dict( param_class=self.param.__class__.__name__ )
        spec_dict.update( self.__dict__ )
        return ( "{name} ({param_class}): {desc}, defaults to '{default}'".format( **spec_dict ) )

# create a statement class that, when called,
#   will add a new MetadataElementSpec to a class's metadata_spec
MetadataElement = Statement( MetadataElementSpec )


"""
MetadataParameter sub-classes.
"""

class SelectParameter( MetadataParameter ):
    def __init__( self, spec ):
        MetadataParameter.__init__( self, spec )
        self.values = self.spec.get( "values" )
        self.multiple = string_as_bool( self.spec.get( "multiple" ) )

    def to_string( self, value ):
        if value in [ None, [] ]:
            return str( self.spec.no_value )
        if not isinstance( value, list ):
            value = [value]
        return ",".join( map( str, value ) )

    def get_html_field( self, value=None, context=None, other_values=None, values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}

        field = form_builder.SelectField( self.spec.name, multiple=self.multiple, display=self.spec.get("display") )
        if self.values:
            value_list = self.values
        elif values:
            value_list = values
        elif value:
            value_list = [ ( v, v ) for v in listify( value )]
        else:
            value_list = []
        for val, label in value_list:
            try:
                if ( self.multiple and val in value ) or ( not self.multiple and val == value ):
                    field.add_option( label, val, selected=True )
                else:
                    field.add_option( label, val, selected=False )
            except TypeError:
                field.add_option( val, label, selected=False )
        return field

    def get_html( self, value, context=None, other_values=None, values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}

        if self.spec.get("readonly"):
            if value in [ None, [] ]:
                return str( self.spec.no_value )
            return ", ".join( map( str, value ) )
        return MetadataParameter.get_html( self, value, context=context, other_values=other_values, values=values, **kwd )

    def wrap( self, value, session ):
        value = self.marshal( value ) #do we really need this (wasteful)? - yes because we are not sure that all existing selects have been stored previously as lists. Also this will handle the case where defaults/no_values are specified and are single non-list values.
        if self.multiple:
            return value
        elif value:
            return value[0] #single select, only return the first value
        return None

    @classmethod
    def marshal( cls, value ):
        # Store select as list, even if single item
        if value is None: return []
        if not isinstance( value, list ): return [value]
        return value


class DBKeyParameter( SelectParameter ):

    def get_html_field( self, value=None, context=None, other_values=None, values=None, **kwd):
        context = context or {}
        other_values = other_values or {}
        try:
            values = kwd['trans'].db_builds
        except KeyError:
            pass
        return super(DBKeyParameter, self).get_html_field( value, context, other_values, values, **kwd)

    def get_html( self, value=None, context=None, other_values=None, values=None, **kwd):
        context = context or {}
        other_values = other_values or {}
        try:
            values = kwd['trans'].db_builds
        except KeyError:
            pass
        return super(DBKeyParameter, self).get_html( value, context, other_values, values, **kwd)


class RangeParameter( SelectParameter ):

    def __init__( self, spec ):
        SelectParameter.__init__( self, spec )
        # The spec must be set with min and max values
        self.min = spec.get( "min" ) or 1
        self.max = spec.get( "max" ) or 1
        self.step = self.spec.get( "step" ) or 1

    def get_html_field( self, value=None, context=None, other_values=None, values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}

        if values is None:
            values = zip( range( self.min, self.max, self.step ), range( self.min, self.max, self.step ))
        return SelectParameter.get_html_field( self, value=value, context=context, other_values=other_values, values=values, **kwd )

    def get_html( self, value, context=None, other_values=None, values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}

        if values is None:
            values = zip( range( self.min, self.max, self.step ), range( self.min, self.max, self.step ))
        return SelectParameter.get_html( self, value, context=context, other_values=other_values, values=values, **kwd )

    @classmethod
    def marshal( cls, value ):
        value = SelectParameter.marshal( value )
        values = [ int(x) for x in value ]
        return values


class ColumnParameter( RangeParameter ):

    def get_html_field( self, value=None, context=None, other_values=None, values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}

        if values is None and context:
            column_range = range( 1, ( context.columns or 0 ) + 1, 1 )
            values = zip( column_range, column_range )
        return RangeParameter.get_html_field( self, value=value, context=context, other_values=other_values, values=values, **kwd )

    def get_html( self, value, context=None, other_values=None, values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}

        if values is None and context:
            column_range = range( 1, ( context.columns or 0 ) + 1, 1 )
            values = zip( column_range, column_range )
        return RangeParameter.get_html( self, value, context=context, other_values=other_values, values=values, **kwd )


class ColumnTypesParameter( MetadataParameter ):

    def to_string( self, value ):
        return ",".join( map( str, value ) )


class ListParameter( MetadataParameter ):

    def to_string( self, value ):
        return ",".join( [str(x) for x in value] )


class DictParameter( MetadataParameter ):

    def to_string( self, value ):
        return  json.dumps( value )


class PythonObjectParameter( MetadataParameter ):

    def to_string( self, value ):
        if not value:
            return self.spec._to_string( self.spec.no_value )
        return self.spec._to_string( value )

    def get_html_field( self, value=None, context=None, other_values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}
        return form_builder.TextField( self.spec.name, value=self._to_string( value ) )

    def get_html( self, value=None, context=None, other_values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}
        return str( self )

    @classmethod
    def marshal( cls, value ):
        return value


class FileParameter( MetadataParameter ):

    def to_string( self, value ):
        if not value:
            return str( self.spec.no_value )
        return value.file_name

    def get_html_field( self, value=None, context=None, other_values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}
        return form_builder.TextField( self.spec.name, value=str( value.id ) )

    def get_html( self, value=None, context=None, other_values=None, **kwd ):
        context = context or {}
        other_values = other_values or {}
        return "<div>No display available for Metadata Files</div>"

    def wrap( self, value, session ):
        if value is None:
            return None
        if isinstance( value, galaxy.model.MetadataFile ) or isinstance( value, MetadataTempFile ):
            return value
        mf = session.query( galaxy.model.MetadataFile ).get( value )
        return mf

    def make_copy( self, value, target_context, source_context ):
        value = self.wrap( value, object_session( target_context.parent ) )
        if value:
            new_value = galaxy.model.MetadataFile( dataset = target_context.parent, name = self.spec.name )
            object_session( target_context.parent ).add( new_value )
            object_session( target_context.parent ).flush()
            shutil.copy( value.file_name, new_value.file_name )
            return self.unwrap( new_value )
        return None

    @classmethod
    def marshal( cls, value ):
        if isinstance( value, galaxy.model.MetadataFile ):
            value = value.id
        return value

    def from_external_value( self, value, parent ):
        """
        Turns a value read from a external dict into its value to be pushed directly into the metadata dict.
        """
        if MetadataTempFile.is_JSONified_value( value ):
            value = MetadataTempFile.from_JSON( value )
        if isinstance( value, MetadataTempFile ):
            mf = parent.metadata.get( self.spec.name, None)
            if mf is None:
                mf = self.new_file( dataset = parent, **value.kwds )
            # Ensure the metadata file gets updated with content
            parent.dataset.object_store.update_from_file( mf, file_name=value.file_name, extra_dir='_metadata_files', extra_dir_at_root=True, alt_name=os.path.basename(mf.file_name) )
            os.unlink( value.file_name )
            value = mf.id
        return value

    def to_external_value( self, value ):
        """
        Turns a value read from a metadata into its value to be pushed directly into the external dict.
        """
        if isinstance( value, galaxy.model.MetadataFile ):
            value = value.id
        elif isinstance( value, MetadataTempFile ):
            value = MetadataTempFile.to_JSON( value )
        return value

    def new_file( self, dataset = None, **kwds ):
        if object_session( dataset ):
            mf = galaxy.model.MetadataFile( name = self.spec.name, dataset = dataset, **kwds )
            object_session( dataset ).add( mf )
            object_session( dataset ).flush() #flush to assign id
            return mf
        else:
            #we need to make a tmp file that is accessable to the head node,
            #we will be copying its contents into the MetadataFile objects filename after restoring from JSON
            #we do not include 'dataset' in the kwds passed, as from_JSON_value() will handle this for us
            return MetadataTempFile( **kwds )


#This class is used when a database file connection is not available
class MetadataTempFile( object ):
    tmp_dir = 'database/tmp' #this should be overwritten as necessary in calling scripts

    def __init__( self, **kwds ):
        self.kwds = kwds
        self._filename = None

    @property
    def file_name( self ):
        if self._filename is None:
            #we need to create a tmp file, accessable across all nodes/heads, save the name, and return it
            self._filename = abspath( tempfile.NamedTemporaryFile( dir = self.tmp_dir, prefix = "metadata_temp_file_" ).name )
            open( self._filename, 'wb+' ) #create an empty file, so it can't be reused using tempfile
        return self._filename

    def to_JSON( self ):
        return { '__class__':self.__class__.__name__, 'filename':self.file_name, 'kwds':self.kwds }

    @classmethod
    def from_JSON( cls, json_dict ):
        #need to ensure our keywords are not unicode
        rval = cls( **stringify_dictionary_keys( json_dict['kwds'] ) )
        rval._filename = json_dict['filename']
        return rval

    @classmethod
    def is_JSONified_value( cls, value ):
        return ( isinstance( value, dict ) and value.get( '__class__', None ) == cls.__name__ )

    @classmethod
    def cleanup_from_JSON_dict_filename( cls, filename ):
        try:
            for key, value in json.load( open( filename ) ).items():
                if cls.is_JSONified_value( value ):
                    value = cls.from_JSON( value )
                if isinstance( value, cls ) and os.path.exists( value.file_name ):
                    log.debug( 'Cleaning up abandoned MetadataTempFile file: %s' % value.file_name )
                    os.unlink( value.file_name )
        except Exception, e:
            log.debug( 'Failed to cleanup MetadataTempFile temp files from %s: %s' % ( filename, e ) )


#Class with methods allowing set_meta() to be called externally to the Galaxy head
class JobExternalOutputMetadataWrapper( object ):
    #this class allows access to external metadata filenames for all outputs associated with a job
    #We will use JSON as the medium of exchange of information, except for the DatasetInstance object which will use pickle (in the future this could be JSONified as well)

    def __init__( self, job ):
        self.job_id = job.id

    def get_output_filenames_by_dataset( self, dataset, sa_session ):
        if isinstance( dataset, galaxy.model.HistoryDatasetAssociation ):
            return sa_session.query( galaxy.model.JobExternalOutputMetadata ) \
                             .filter_by( job_id = self.job_id, history_dataset_association_id = dataset.id ) \
                             .first() #there should only be one or None
        elif isinstance( dataset, galaxy.model.LibraryDatasetDatasetAssociation ):
            return sa_session.query( galaxy.model.JobExternalOutputMetadata ) \
                             .filter_by( job_id = self.job_id, library_dataset_dataset_association_id = dataset.id ) \
                             .first() #there should only be one or None
        return None

    def get_dataset_metadata_key( self, dataset ):
        # Set meta can be called on library items and history items,
        # need to make different keys for them, since ids can overlap
        return "%s_%d" % ( dataset.__class__.__name__, dataset.id )

    def setup_external_metadata( self, datasets, sa_session, exec_dir=None, tmp_dir=None, dataset_files_path=None,
                                 output_fnames=None, config_root=None, config_file=None, datatypes_config=None, job_metadata=None, compute_tmp_dir=None, kwds=None ):
        kwds = kwds or {}
        if tmp_dir is None:
            tmp_dir = MetadataTempFile.tmp_dir

        # path is calculated for Galaxy, may be different on compute - rewrite
        # for the compute server.
        def metadata_path_on_compute(path):
            compute_path = path
            log.info(compute_tmp_dir)
            if compute_tmp_dir and tmp_dir and in_directory(path, tmp_dir):
                path_relative = os.path.relpath(path, tmp_dir)
                compute_path = os.path.join(compute_tmp_dir, path_relative)
            return compute_path

        #fill in metadata_files_dict and return the command with args required to set metadata
        def __metadata_files_list_to_cmd_line( metadata_files ):
            def __get_filename_override():
                if output_fnames:
                    for dataset_path in output_fnames:
                        if dataset_path.false_path and dataset_path.real_path == metadata_files.dataset.file_name:
                            return dataset_path.false_path
                return ""
            line = "%s,%s,%s,%s,%s,%s" % (
                metadata_path_on_compute(metadata_files.filename_in),
                metadata_path_on_compute(metadata_files.filename_kwds),
                metadata_path_on_compute(metadata_files.filename_out),
                metadata_path_on_compute(metadata_files.filename_results_code),
                __get_filename_override(),
                metadata_path_on_compute(metadata_files.filename_override_metadata),
            )
            log.info(line)
            return line
        if not isinstance( datasets, list ):
            datasets = [ datasets ]
        if exec_dir is None:
            exec_dir = os.path.abspath( os.getcwd() )
        if dataset_files_path is None:
            dataset_files_path = galaxy.model.Dataset.file_path
        if config_root is None:
            config_root = os.path.abspath( os.getcwd() )
        if datatypes_config is None:
            raise Exception( 'In setup_external_metadata, the received datatypes_config is None.' )
            datatypes_config = 'datatypes_conf.xml'
        metadata_files_list = []
        for dataset in datasets:
            key = self.get_dataset_metadata_key( dataset )
            #future note:
            #wonkiness in job execution causes build command line to be called more than once
            #when setting metadata externally, via 'auto-detect' button in edit attributes, etc.,
            #we don't want to overwrite (losing the ability to cleanup) our existing dataset keys and files,
            #so we will only populate the dictionary once
            metadata_files = self.get_output_filenames_by_dataset( dataset, sa_session )
            if not metadata_files:
                metadata_files = galaxy.model.JobExternalOutputMetadata( dataset = dataset)
                metadata_files.job_id = self.job_id
                #we are using tempfile to create unique filenames, tempfile always returns an absolute path
                #we will use pathnames relative to the galaxy root, to accommodate instances where the galaxy root
                #is located differently, i.e. on a cluster node with a different filesystem structure

                #file to store existing dataset
                metadata_files.filename_in = abspath( tempfile.NamedTemporaryFile( dir = tmp_dir, prefix = "metadata_in_%s_" % key ).name )

                #FIXME: HACK
                #sqlalchemy introduced 'expire_on_commit' flag for sessionmaker at version 0.5x
                #This may be causing the dataset attribute of the dataset_association object to no-longer be loaded into memory when needed for pickling.
                #For now, we'll simply 'touch' dataset_association.dataset to force it back into memory.
                dataset.dataset #force dataset_association.dataset to be loaded before pickling
                #A better fix could be setting 'expire_on_commit=False' on the session, or modifying where commits occur, or ?

                cPickle.dump( dataset, open( metadata_files.filename_in, 'wb+' ) )
                #file to store metadata results of set_meta()
                metadata_files.filename_out = abspath( tempfile.NamedTemporaryFile( dir = tmp_dir, prefix = "metadata_out_%s_" % key ).name )
                open( metadata_files.filename_out, 'wb+' ) # create the file on disk, so it cannot be reused by tempfile (unlikely, but possible)
                #file to store a 'return code' indicating the results of the set_meta() call
                #results code is like (True/False - if setting metadata was successful/failed , exception or string of reason of success/failure )
                metadata_files.filename_results_code = abspath( tempfile.NamedTemporaryFile( dir = tmp_dir, prefix = "metadata_results_%s_" % key ).name )
                json.dump( ( False, 'External set_meta() not called' ), open( metadata_files.filename_results_code, 'wb+' ) ) # create the file on disk, so it cannot be reused by tempfile (unlikely, but possible)
                #file to store kwds passed to set_meta()
                metadata_files.filename_kwds = abspath( tempfile.NamedTemporaryFile( dir = tmp_dir, prefix = "metadata_kwds_%s_" % key ).name )
                json.dump( kwds, open( metadata_files.filename_kwds, 'wb+' ), ensure_ascii=True )
                #existing metadata file parameters need to be overridden with cluster-writable file locations
                metadata_files.filename_override_metadata = abspath( tempfile.NamedTemporaryFile( dir = tmp_dir, prefix = "metadata_override_%s_" % key ).name )
                open( metadata_files.filename_override_metadata, 'wb+' ) # create the file on disk, so it cannot be reused by tempfile (unlikely, but possible)
                override_metadata = []
                for meta_key, spec_value in dataset.metadata.spec.iteritems():
                    if isinstance( spec_value.param, FileParameter ) and dataset.metadata.get( meta_key, None ) is not None:
                        metadata_temp = MetadataTempFile()
                        shutil.copy( dataset.metadata.get( meta_key, None ).file_name, metadata_temp.file_name )
                        override_metadata.append( ( meta_key, metadata_temp.to_JSON() ) )
                json.dump( override_metadata, open( metadata_files.filename_override_metadata, 'wb+' ) )
                #add to session and flush
                sa_session.add( metadata_files )
                sa_session.flush()
            metadata_files_list.append( metadata_files )
        #return command required to build
        return "%s %s %s %s %s %s %s %s" % ( os.path.join( exec_dir, 'set_metadata.sh' ), dataset_files_path, tmp_dir, config_root, config_file, datatypes_config, job_metadata, " ".join( map( __metadata_files_list_to_cmd_line, metadata_files_list ) ) )

    def external_metadata_set_successfully( self, dataset, sa_session ):
        metadata_files = self.get_output_filenames_by_dataset( dataset, sa_session )
        if not metadata_files:
            return False # this file doesn't exist
        rval, rstring = json.load( open( metadata_files.filename_results_code ) )
        if not rval:
            log.debug( 'setting metadata externally failed for %s %s: %s' % ( dataset.__class__.__name__, dataset.id, rstring ) )
        return rval

    def cleanup_external_metadata( self, sa_session ):
        log.debug( 'Cleaning up external metadata files' )
        for metadata_files in sa_session.query( galaxy.model.Job ).get( self.job_id ).external_output_metadata:
            #we need to confirm that any MetadataTempFile files were removed, if not we need to remove them
            #can occur if the job was stopped before completion, but a MetadataTempFile is used in the set_meta
            MetadataTempFile.cleanup_from_JSON_dict_filename( metadata_files.filename_out )
            dataset_key = self.get_dataset_metadata_key( metadata_files.dataset )
            for key, fname in [ ( 'filename_in', metadata_files.filename_in ), ( 'filename_out', metadata_files.filename_out ), ( 'filename_results_code', metadata_files.filename_results_code ), ( 'filename_kwds', metadata_files.filename_kwds ), ( 'filename_override_metadata', metadata_files.filename_override_metadata ) ]:
                try:
                    os.remove( fname )
                except Exception, e:
                    log.debug( 'Failed to cleanup external metadata file (%s) for %s: %s' % ( key, dataset_key, e ) )

    def set_job_runner_external_pid( self, pid, sa_session ):
        for metadata_files in sa_session.query( galaxy.model.Job ).get( self.job_id ).external_output_metadata:
            metadata_files.job_runner_external_pid = pid
            sa_session.add( metadata_files )
            sa_session.flush()