# XML-database conversion

import copy

import re

from xml.dom import Node

import dicts

import exc

import Parser

import sql

import sql_io

import sql_gen

import strings

import util

import xml_dom

import xml_func

import xpath

def name_of(node): return re.sub(r'^.*\.', r'', node.tagName)

ptr_suffix = '_id'

def is_ptr(node_name): return node_name.lower().endswith(ptr_suffix)

def ptr_type_guess(node_name):

    assert is_ptr(node_name)

    return node_name[:-len(ptr_suffix)]

def ptr_target(node):

    assert is_ptr(name_of(node))

    return xml_dom.value_node(node)

def find_by_name(node, name):

    for parent in xml_dom.NodeParentIter(node):

        if name_of(parent) == name: return parent

        else:

            for child in xml_dom.NodeElemIter(parent):

                child_name = name_of(child)

                if is_ptr(child_name):

                    target = ptr_target(child)

                    if target.tagName == name: return target

                elif child_name == name: return child

    return None

class ColRef:

    '''A reference to a table column'''

    def __init__(self, name, idx):

        self.name = name

        self.idx = idx

    def __str__(self): return self.name

input_col_prefix = xml_func.var_name_prefix

put_special_funcs = set(['_setDefault', '_simplifyPath'])

no_parent_ids_loc = object() # tells put() there is no parent_ids_loc

def put(db, node, row_ins_ct_ref=None, on_error=exc.reraise, col_defaults={},

    in_table=None, parent_ids_loc=no_parent_ids_loc, next=None):

    '''

    @param node To use an entire XML document, pass root.firstChild.

    '''

    if node == None: return None # when no rows, root.firstChild == None

    elif xml_dom.is_text_node(node): return xml_dom.value(node)

    def put_(node):

        if util.is_str(node): return node

        return put(db, node, row_ins_ct_ref, on_error, col_defaults, in_table,

            parent_ids_loc, next)

    def augment_error(e): exc.add_msg(e, 'node:\n'+strings.ustr(node))

    def on_error_(e):

        augment_error(e)

        on_error(e)

    def wrap_e(e):

        augment_error(e)

        raise xml_func.SyntaxError(e)

    is_func = xml_func.is_func(node)

    out_table = name_of(node)

    # Divide children into fields and children with fkeys to parent

    row = dicts.OnceOnlyDict()

    children = []

    try:

        for child in xml_dom.NodeElemIter(node):

            child_name = name_of(child)

            if xml_dom.is_empty(child): row[child_name] = None

            elif xml_dom.is_text(child):

                row[child_name] = strings.to_unicode(xml_dom.value(child))

            else:

                child_value = xml_dom.value_node(child)

                if ((is_func or is_ptr(child_name)

                    or xml_func.is_func(child_value))

                    and not xml_func.is_func(child)):

                    row[child_name] = child_value

                else: children.append(child)

    except dicts.KeyExistsError, e: wrap_e(e)

    # Special handling for structural XML functions

    if out_table == '_setDefault':

        # Parse args

        try: path = row.pop('path')

        except KeyError, e: wrap_e(e)

        col_defaults.update([(k, put_(v)) for k, v in row.iteritems()])

        return put_(path)

    elif out_table == '_simplifyPath':

        # Parse args

        try:

            next = row['next'] # modifies outer next var used by put_()

            path = row['path']

        except KeyError, e: wrap_e(e)

        try: next = xpath.parse(next)

        except Parser.SyntaxError, e: wrap_e(e)

        try: next = next[0].name

        except IndexError, e: wrap_e(e)

        return put_(path)

    is_literals = in_table == None

    in_tables = []

    no_empty = set()

    if not is_literals:

        in_tables.append(in_table)

        no_empty.add(in_table)

    def pkey_name(table): return sql.pkey_name(db, table, True)

    # Add fkey to parent

    if parent_ids_loc is not no_parent_ids_loc:

        if sql_gen.is_table_col(parent_ids_loc):

            no_empty.add(parent_ids_loc.table)

        parent_ptr = node.getAttribute('fkey')

        if parent_ptr == '': parent_ptr = pkey_name(name_of(node.parentNode))

        row[parent_ptr] = parent_ids_loc

    # Parse input columns

    row = row.inner # now allow keys to be overwritten

    for out_col, value in row.iteritems():

        if (not is_literals and util.is_str(value)

            and value.startswith(input_col_prefix)): # value is input column

            row[out_col] = sql_gen.Col(strings.remove_prefix(input_col_prefix,

                value), in_table)

    # Optimizations for structural XML functions

    if out_table == '_alt': # return first arg if non-NULL

        args = row.items()

        args.sort()

        out_col, value = min(args) # first arg

        if xml_dom.is_node(value): row[out_col] = value = put_(value)

        if not sql_gen.is_nullable(db, value): return value

    # Process values

    parent_ids_loc = no_parent_ids_loc # applies to this section

    for out_col, value in row.iteritems():

        # Handle forward pointers

        if xml_dom.is_node(value): row[out_col] = value = put_(value)

        # Translate values

        if isinstance(value, sql_gen.Col): # value is table column

            assert sql_gen.is_table_col(value)

            if value.table is not in_table: in_tables.append(value.table)

        else: # value is literal value

            row[out_col] = sql_gen.NamedCol(out_col, value)

    # Insert node

    try: pkeys_loc = sql_io.put_table(db, out_table, in_tables, row,

        row_ins_ct_ref, next, col_defaults, on_error_)

    except Exception, e:

        augment_error(e)

        raise

    if sql_gen.is_table_col(pkeys_loc): no_empty.add(pkeys_loc.table)

    sql.empty_temp(db, set(in_tables) - no_empty)

    # Insert children with fkeys to parent

    parent_ids_loc = pkeys_loc # applies to this section

    for child in children: put_(child)

    return pkeys_loc

def get(db, node, limit=None, start=None):

    def pkey_name(table): return sql.pkey_name(db, table)

    node = node.firstChild

    table = name_of(node)

    pkey_ = pkey_name(table)

    fields = []

    conds = {}

    for child in xml_dom.NodeElemIter(node):

        child_name = name_of(child)

        if xml_dom.is_empty(child): fields.append(child_name)

        elif xml_dom.is_text(child): conds[child_name] = xml_dom.value(child)

        else: raise Exception('Joins not supported yet')

    id_ = xml_dom.get_id(node)

    if id_ != None: conds[pkey_name(table)] = id_ # replace any existing value

    if fields == []: fields.append(pkey_)

    return sql.select(db, table, fields, conds, limit, start)

# Controls when and how put_table() will partition the input table

partition_size = 1000000 # rows; must be >= NCBI.nodes size

def put_table(db, node, in_table, in_row_ct_ref=None, row_ins_ct_ref=None,

    limit=None, start=0, on_error=exc.reraise, col_defaults={},

    partition_size=partition_size):

    '''

    @param node The XML tree that transforms the input to the output. Similar to

        put()'s node param, but with the input column name prefixed by

        input_col_prefix in place of the column value.

    @return sql_gen.Col Where the pkeys (from INSERT RETURNING) are made

        available

    '''

    in_table = sql_gen.as_Table(in_table)

    sql_io.mk_errors_table(db, in_table)

    in_table.set_srcs([in_table], overwrite=False)

    db.src = strings.ustr(in_table)

    db.autoanalyze = True # but don't do this in row-based import

    db.autoexplain = True # but don't do this in row-based import

    # Subset and partition in_table

    # OK to do even if table already the right size because it takes <1 sec.

    full_in_table = in_table

    pkeys_loc = None # used if loop is never executed

    total = 0

    while limit == None or total < limit:

        # Adjust partition size if last partition

        this_limit = partition_size

        if limit != None: this_limit = min(this_limit, limit - total)

        # Row # is interally 0-based, but 1-based to the user

        db.log_debug('********** Partition: rows '+str(start+1)+'-'

            +str(start+this_limit)+' **********', level=1.2)

        # Subset in_table

        in_table = sql_gen.Table(strings.ustr(full_in_table),

            srcs=full_in_table.srcs, is_temp=True) # prepend schema to name

        sql.copy_table_struct(db, full_in_table, in_table)

        try: sql.add_row_num(db, in_table, 'row_num')

        except sql.DatabaseErrors: pass # already has pkey

        cur = sql.insert_select(db, in_table, None, sql.mk_select(db,

            full_in_table, limit=this_limit, start=start))

        this_ct = cur.rowcount

        total += this_ct

        start += this_ct # advance start to fetch next set

        if this_ct == 0: break # in_table size is multiple of partition_size

        # Import data

        pkeys_loc = put(db, node, row_ins_ct_ref, on_error, col_defaults,

            in_table)

        if in_row_ct_ref != None: in_row_ct_ref[0] += this_ct

        sql.empty_temp(db, in_table)

        if this_ct < partition_size: break # partial partition = last

        # Work around PostgreSQL's temp table disk space leak

        db.reconnect()

    return pkeys_loc