#!/usr/bin/env python3
import argparse
import sys
import json
from typing import Collection

import pywikiapi
import mwparserfromhell
import requests
import luadata
import tomli
from SPARQLWrapper import SPARQLWrapper, JSON

USER_AGENT = 'osmwiki-overpass-import'
OSMWIKI_ENDPOINT = 'https://wiki.openstreetmap.org/w/api.php'
OVERPASS_ENDPOINT = 'https://lz4.overpass-api.de/api/interpreter'
WIKIDATA_ENDPOINT = 'https://query.wikidata.org/sparql'

osmwiki = pywikiapi.Site(OSMWIKI_ENDPOINT)

parser = argparse.ArgumentParser()
parser.add_argument('--update', action='store_true')
args = parser.parse_args()


def find_template_calls(template_name):
    for page in osmwiki.query_pages(
        generator='embeddedin',
        geititle='Template:' + template_name,
        prop='revisions',
        rvprop='content',
        rvslots='main',
    ):
        doc = mwparserfromhell.parse(page['revisions'][0]['slots']['main']['content'])
        for template in doc.filter_templates():
            if template.name.matches(template_name):
                yield page, template


# step 1: find relation ids by looking for template calls

relation_ids = {}

for page, tpl in find_template_calls('list relations'):
    if not tpl.has(1):
        continue
    for arg in tpl.get(1).split():
        if arg.isdigit():
            relation_ids[int(arg)] = relation_ids.get(arg, False) or tpl.has('length')
        else:
            # TODO: only if verbose output is enabled
            print(
                f'[warning] found unexpected argument "{arg}" in {page["title"]}',
                file=sys.stderr,
            )

# step 2: query overpass turbo about relations

sess = requests.session()
sess.headers['user-agent'] = USER_AGENT

relations = {}

overpass_info = None


def get_length(route, relations):
    """
    The length operator of Overpass QL does not support superrelations
    (relations containing relations) and just returns 0 for those,
    so we need to implement our own recursion.
    """
    if route['members']:
        # If there are any members of type relation we recurse,
        # ensuring that we still report a sensible result when
        # ways are mistakenly added directly to superrelations.

        # TODO: gracefully handle KeyError exception
        return sum(get_length(relations[m], relations) for m in route['members'])
    else:
        return float(route['tags']['length'])


def query_lengths_from_wikidata(wikidata_ids):
    query = """SELECT ?id ?length WHERE {
      VALUES ?id { %s }
      ?id p:P2043/psn:P2043/wikibase:quantityAmount ?length  .
    }""" % ' '.join(
        f'wd:' + x for x in wikidata_ids
    )

    sparql = SPARQLWrapper(WIKIDATA_ENDPOINT, agent=USER_AGENT)
    sparql.setQuery(query)
    sparql.setReturnFormat(JSON)
    return {
        res['id']['value'].rsplit('/', maxsplit=1)[1]: float(res['length']['value'])
        for res in sparql.query().convert()["results"]["bindings"]
    }


def query_overpass(relation_ids: Collection[int], with_length: bool):
    # TODO: split query if it gets too large

    query = '[out:json]; ('
    for rel_id in relation_ids:
        query += f'relation({rel_id});'
        if with_length:
            query += 'rel(r);'  # 1. recursion
            query += 'rel(r);'  # 2. recursion
            query += 'rel(r);'  # 3. recursion
    query += ');'
    if with_length:
        query += 'convert result ::=::, ::id=id(), length=length(), member_ids=per_member(ref()), member_types=per_member(mtype());'
    query += 'out body;'

    res = sess.get(OVERPASS_ENDPOINT, params=dict(data=query))
    res.raise_for_status()

    KEYS = ('name', 'name:en', 'wikidata', 'wikipedia', 'website')

    res = res.json()

    all_relations = {rel['id']: rel for rel in res['elements']}

    if with_length:
        for rel in all_relations.values():
            member_ids = [int(x) for x in rel['tags']['member_ids'].split(';')]
            member_types = rel['tags']['member_types'].split(';')
            rel['members'] = [
                m_id
                for m_id, m_type in zip(member_ids, member_types)
                if m_type == 'relation'
            ]

        def wikidata_ids():
            for rel_id in relation_ids:
                wikidata_id = all_relations[rel_id]['tags'].get('wikidata')
                if wikidata_id:
                    yield wikidata_id

        wikidata_lengths = query_lengths_from_wikidata(wikidata_ids())

    for rel_id in relation_ids:
        rel = all_relations.get(rel_id)
        if not rel:
            # TODO: report 404 in generated data
            print(
                f'[warning] could not find relation {rel_id}',
                file=sys.stderr,
            )
            continue
        data = dict(tags={k: v for k, v in rel['tags'].items() if k in KEYS})

        if with_length:
            data['mapped_length'] = round(get_length(rel, all_relations) / 1000, 1)
            wikidata_id = rel['tags'].get('wikidata')
            if wikidata_id in wikidata_lengths:
                data['wikidata_length'] = round(wikidata_lengths[wikidata_id] / 1000, 1)

        relations[rel_id] = data

    global overpass_info
    overpass_info = res['osm3s']


query_overpass([rel_id for rel_id, length in relation_ids.items() if not length], False)
query_overpass([rel_id for rel_id, length in relation_ids.items() if length], True)

# step 3: serialize data as Lua

text = f'''\
-- This page is automatically generated by a Python script using the Overpass API.
-- {overpass_info['copyright']}
-- The wikidata_length data is queried from www.wikidata.org and available under Creative Commons CC0 License.

return ''' + luadata.serialize(
    dict(relations=relations)
)

if args.update:
    with open('credentials.toml', 'rb') as f:
        creds = tomli.load(f)
    osmwiki.login(creds['username'], creds['password'])
    csrf_token = osmwiki('query', meta='tokens')['query']['tokens']['csrftoken']
    osmwiki(
        'edit',
        title='Module:Report/data',
        text=text,
        token=csrf_token,
        summary=f"update (osm_base = {overpass_info['timestamp_osm_base']})",
    )
else:
    print(text)