# coding=utf-8

from io import StringIO

import datetime

import gzip

import json
import logging
import random

import tarfile

import time

import urllib.request as urllib_request

import requests

import re  # regex

from csv import writer as csv_writer, DictWriter as csv_dict_writer

from math import sqrt, isnan

from os import environ, remove as removefile
from os.path import isfile, join, abspath, dirname

# (i) Just ignore the requirements warning, we already require `dateutils`.

from dateutil.relativedelta import relativedelta

from dateutil import parser as dateparser

from flask import Flask

from flask import request

from flask import url_for, send_from_directory, abort as abort_flask

from jinja2 import Environment, FileSystemLoader
from jinja2.utils import markupsafe

from yaml import load as yaml_load, dump

from speasy import amda
import pandas as pd

try:
    from yaml import CLoader as Loader
except ImportError:
    from yaml import Loader

from netCDF4 import Dataset, date2num

# PATH RELATIVITY #############################################################

THIS_DIRECTORY = dirname(abspath(__file__))


def get_path(relative_path):

    """Get an absolute path from the relative path to this script directory."""

    return abspath(join(THIS_DIRECTORY, relative_path))


# COLLECT GLOBAL INFORMATION FROM SOURCES #####################################

# VERSION
with open(get_path('../VERSION'), 'r') as version_file:
    version = version_file.read().strip()

# CONFIG

with open(get_path('../config.yml'), 'r', encoding='utf8') as config_file:

    config = yaml_load(config_file.read(), Loader=Loader)

FILE_DATE_FMT = "%Y-%m-%dT%H:%M:%S"

MOMENT_DATE_FMT = "%Y-%m-%dT%H:%M:%SZ"

CME_DATE_FMT = "%Y-%m-%dT%H:%MZ"

# Are we on the SSA instance for ESA?
SSA = environ.get('SSA') == 'true'

DEBUG = environ.get('DEBUG') in ['true', 'True', 1, '1']

# LOGGING #####################################################################

LOG_FILE = get_path('run.log')

hp_logger = logging.getLogger("HelioPropa")

hp_logger.setLevel(logging.DEBUG)

fileHandler = logging.FileHandler(LOG_FILE)
fileHandler.setFormatter(logging.Formatter(
    "%(asctime)s -  %(levelname)s - %(message)s"

))

# set file messages depending on env var setting
if DEBUG:

    fileHandler.setLevel(logging.DEBUG)

else:

    fileHandler.setLevel(logging.ERROR)
hp_logger.addHandler(fileHandler)


def init_console_logger():
    consoleHandler = logging.StreamHandler()
    if DEBUG:
        consoleHandler.setLevel(logging.DEBUG)
    else:
        consoleHandler.setLevel(logging.ERROR)
    formatter = logging.Formatter('CONSOLE: %(levelname)s - %(message)s')
    consoleHandler.setFormatter(formatter)
    # Add handlers to the logger
    hp_logger.addHandler(consoleHandler)
    return hp_logger

# HARDCODED CONFIGURATION #####################################################

ASTRONOMICAL_UNIT_IN_KM = 1.496e8

# Absolute path to the installed CDF library from https://cdf.gsfc.nasa.gov/
CDF_LIB = '/usr/local/lib/libcdf'

# Absolute path to the data cache directory
CACHE_DIR = get_path('../cache')

# These two configs are not in the YAML config because adding a new parameter

# will not work as-is, you'd have to edit some netcdf-related code.

# The slugs of the available parameters in the generated CSV files.
# The order matters. If you change this you also need to change the
# innermost loop of `get_data_for_target`.
# The javascript knows the targets' properties under these names.

# PROPERTIES = ('time', 'vrad', 'vtan', 'vtot', 'btan', 'brad', 'temp', 'pdyn', 'dens',
#               'atse', 'xhee', 'yhee')

PROPERTIES = ('time', 'vrad', 'vtan', 'vtot', 'btan', 'temp', 'pdyn', 'dens',

              'atse', 'xhee', 'yhee')

# The parameters that the users can handle.
# The slug MUST be one of the properties above.
PARAMETERS = {
    'pdyn': {
        'slug': 'pdyn',
        'name': 'Dyn. Pressure',
        'title': 'The dynamic pressure.',
        'units': 'nPa',
        'active': True,
        'position': 10,
    },
    'vtot': {
        'slug': 'vtot',
        'name': 'Velocity',
        'title': 'The velocity of the particles.',
        'units': 'km/s',
        'active': False,
        'position': 20,
    },
    'btan': {
        'slug': 'btan',
        'name': 'B Tangential',
        'title': 'B Tangential.',
        'units': 'nT',
        'active': False,
        'position': 30,
    },

    # 'brad': {
    #     'slug': 'brad',
    #     'name': 'B Radial',
    #     'title': 'B Radial.',
    #     'units': 'nT',
    #     'active': False,
    #     'position': 35,
    # },

    'temp': {
        'slug': 'temp',
        'name': 'Temperature',

        'title': 'The temperature.',
        'units': 'eV',

        'active': False,
        'position': 40,
    },
    'dens': {
        'slug': 'dens',
        'name': 'Density',
        'title': 'The density N.',

        'units': 'cm^-3',

        'active': False,
        'position': 50,
    },

    'atse': {
        'slug': 'atse',

        'name': 'Angle T-S-E',
        'title': 'Angle Target-Sun-Earth.',
        'units': 'deg',
        'active': False,
        'position': 60,
    },
}

# SETUP ENVIRONMENT ###########################################################

environ['SPACEPY'] = CACHE_DIR
environ['CDF_LIB'] = CDF_LIB

# SETUP FLASK ENGINE ##########################################################

app = Flask(__name__, root_path=THIS_DIRECTORY)

app.debug = DEBUG

if app.debug:

    hp_logger.info("Starting Flask app IN DEBUG MODE...")

else:

    hp_logger.info("Starting Flask app...")

def handle_error(e):

    hp_logger.error(e)

    return str(e)  # wish we could use the default error renderer here

app.register_error_handler(Exception, handle_error)

# SETUP JINJA2 TEMPLATE ENGINE ################################################

def static_global(filename):
    return url_for('static', filename=filename)


def shuffle_filter(seq):
    """
    This shuffles the sequence it is applied to.

    Jinja2 _should_ provide this.

    """
    try:
        result = list(seq)
        random.shuffle(result)
        return result
    except:
        return seq


def markdown_filter(value, nl2br=False, p=True):
    """

    Converts markdown into html.

    nl2br: set to True to replace line breaks with <br> tags
    p: set to False to remove the enclosing <p></p> tags
    """
    from markdown import markdown
    from markdown.extensions.nl2br import Nl2BrExtension
    from markdown.extensions.abbr import AbbrExtension
    extensions = [AbbrExtension()]
    if nl2br is True:
        extensions.append(Nl2BrExtension())
    markdowned = markdown(value, output_format='html5', extensions=extensions)
    if p is False:
        markdowned = markdowned.replace(r"<p>", "").replace(r"</p>", "")
    return markdowned

_js_escapes = {

    '\\': '\\u005C',
    '\'': '\\u0027',
    '"': '\\u0022',
    '>': '\\u003E',
    '<': '\\u003C',
    '&': '\\u0026',
    '=': '\\u003D',
    '-': '\\u002D',
    ';': '\\u003B',
    u'\u2028': '\\u2028',
    u'\u2029': '\\u2029'

}
# Escape every ASCII character with a value less than 32.

_js_escapes.update(('%c' % z, '\\u%04X' % z) for z in range(32))

def escapejs_filter(value):
    escaped = []
    for letter in value:
        if letter in _js_escapes:
            escaped.append(_js_escapes[letter])
        else:
            escaped.append(letter)

    return markupsafe.Markup("".join(escaped))

tpl_engine = Environment(loader=FileSystemLoader([get_path('view')]),
                         trim_blocks=True,
                         lstrip_blocks=True)

tpl_engine.globals.update(
    url_for=url_for,
    static=static_global,
)

tpl_engine.filters['markdown'] = markdown_filter
tpl_engine.filters['md'] = markdown_filter
tpl_engine.filters['shuffle'] = shuffle_filter

tpl_engine.filters['escapejs'] = escapejs_filter

tpl_global_vars = {
    'request': request,
    'version': version,
    'config': config,
    'now': datetime.datetime.now(),

    'is_esa': SSA,

}


# HELPERS #####################################################################

def abort(code, message):

    hp_logger.error("Abort: " + message)

    abort_flask(code, message)

def render_view(view, context=None):
    """
    A simple helper to render [view] template with [context] vars.
    It automatically adds the global template vars defined above, too.
    It returns a string, usually the HTML contents to display.
    """
    context = {} if context is None else context
    return tpl_engine.get_template(view).render(

        dict(list(tpl_global_vars.items()) + list(context.items()))

    )


# def render_page(page, title="My Page", context=None):
#     """
#     A simple helper to render the md_page.html template with [context] vars &
#     the additional contents of `page/[page].md` in the `md_page` variable.
#     It automagically adds the global template vars defined above, too.
#     It returns a string, usually the HTML contents to display.
#     """
#     if context is None:
#         context = {}
#     context['title'] = title
#     context['md_page'] = ''
#     with file(get_path('page/%s.md' % page)) as f:
#         context['md_page'] = f.read()
#     return tpl_engine.get_template('md_page.html').render(
#         dict(tpl_global_vars.items() + context.items())
#     )

def is_list_in_list(needle, haystack):
    for n in needle:
        if n not in haystack:
            return False
    return True

def round_time(dt=None, round_to=60):
    """
    Round a datetime object to any time laps in seconds
    dt : datetime.datetime object, default now.
    roundTo : Closest number of seconds to round to, default 1 minute.
    """
    if dt is None:
        dt = datetime.datetime.now()
    seconds = (dt.replace(tzinfo=None) - dt.min).seconds
    rounding = (seconds + round_to / 2) // round_to * round_to

    return dt + datetime.timedelta(0, rounding - seconds, -dt.microsecond)

def datetime_from_list(time_list):

    """

    Datetimes in retrieved CDFs are stored as lists of numbers,

    with DayOfYear starting at 0. We want it starting at 1 because it's what
    vendor parsers use, both in python and javascript.

    """

    try:
        time_list = [str(i, 'UTF8') for i in time_list]
    except Exception as e:

        hp_logger.error(e)

    # Day Of Year starts at 0, but for our datetime parser it starts at 1
    doy = '{:03d}'.format(int(''.join(time_list[4:7])) + 1)
    return datetime.datetime.strptime(

        "%s%s%s" % (''.join(time_list[0:4]), doy, ''.join(time_list[7:-1])),

        "%Y%j%H%M%S%f"
    )

# Override these using the model configuration in config.yml
default_nc_keys = {
    'hee': 'HEE',
    'vtot': 'V',
    'magn': 'B',
    'temp': 'T',
    'dens': 'N',
    'pdyn': 'P_dyn',
    'atse': 'Delta_angle',
}

def _sta_sto(_cnf, _sta, _sto):
    if 'started_at' in _cnf:
        _s0 = datetime.datetime.strptime(_cnf['started_at'], FILE_DATE_FMT)
        _s0 = max(_s0, _sta)
    else:
        _s0 = _sta
    if 'stopped_at' in _cnf:
        _s1 = datetime.datetime.strptime(_cnf['stopped_at'], FILE_DATE_FMT)
        _s1 = min(_s1, _sto)
    else:
        _s1 = _sto
    return _s0, _s1


def _read_var(_nc, _keys, _key, mandatory=False):
    try:

        return _nc.variables[_keys[_key]][:]

    except KeyError:
        pass
    if mandatory:
        raise Exception("No variable '%s' found in NetCDF." % _keys[_key])
    return [None] * len(_nc.variables['Time'])  # slow -- use numpy!

def get_local_filename(url):
    """
    Build the local cache filename for the distant file
    :param url: string
    :return: string
    """
    from slugify import slugify
    n = len('http://')
    if url.startswith('https'):
        n += 1
    s = url[n:]
    return slugify(s)

def get_target_config(slug):

    for s in config['targets']:  # dumb

        if s['slug'] == slug:
            return s

    raise Exception("No target found in configuration for '%s'." % slug)

def check_target_config(slug):
    get_target_config(slug)

def get_active_targets():
    all_targets = config['targets']
    return [t for t in all_targets if not ('locked' in t and t['locked'])]

def validate_tap_target_config(target):
    tc = get_target_config(target)
    if 'tap' not in tc:
        raise Exception("No `tap` configuration for target `%s`." % target)
    if 'target_name' not in tc['tap']:
        raise Exception("No `target_name` in the `tap` configuration for target `%s`." % target)
    return tc


# Using pyvo would be best.
# def retrieve_auroral_emissions_vopy(target_name):
#     api_url = "http://voparis-tap.obspm.fr/__system__/tap/run/tap/sync"
#     import pyvo as vo
#     service = vo.dal.TAPService(api_url)
#     # … can't figure out how to install pyvo and spacepy alongside (forking?)


def retrieve_auroral_emissions(target_name, d_started_at=None, d_stopped_at=None):

    """
    Work In Progress.

    :param target_name: You should probably not let users define this value,

                        as our sanitizing for ADQL may not be 100% safe.
                        Use values from YAML configuration, instead.
                        Below is a list of the ids we found to be existing.
                        > SELECT DISTINCT target_name FROM apis.epn_core

                        - Jupiter
                        - Saturn

                        - Mars
                        - MERCURY

                        - Titan
                        - Io
                        - VENUS
                        - Ganymede
                        - Uranus
                        - Callisto
                        - Europa

                        
                        https://pdssbn.astro.umd.edu/holdings/nh-p-rex-2-pluto-v1.0/document/codmac_level_definitions.pdf
                        processing_level == 3
                        
                        API doc and sandbox:
                        http://voparis-tap-planeto.obspm.fr/__system__/tap/run/info

    :return: 
    """

    # Try out the form
    # http://voparis-tap-planeto.obspm.fr/__system__/adql/query/form

    api_url = "http://voparis-tap.obspm.fr/__system__/tap/run/tap/sync"

    if d_started_at is None:
        d_started_at = datetime.datetime.now()
        t_started_at = time.mktime(d_started_at.timetuple()) - 3600 * 24 * 365 * 2
        # t_started_at = 1
    else:
        t_started_at = time.mktime(d_started_at.timetuple())

    if d_stopped_at is None:
        d_stopped_at = datetime.datetime.now()

    t_stopped_at = time.mktime(d_stopped_at.timetuple())

    def timestamp_to_jday(timestamp):

        return timestamp / 86400.0 + 2440587.5

    def jday_to_timestamp(jday):
        return (jday - 2440587.5) * 86400.0

    def jday_to_datetime(jday):
        return datetime.datetime.utcfromtimestamp(jday_to_timestamp(jday))

    # SELECT DISTINCT dataproduct_type FROM apis.epn_core
    # > im sp sc

    query = """
SELECT
  time_min,
  time_max,
  thumbnail_url,
  external_link
FROM apis.epn_core
WHERE target_name='{target_name}'
AND   dataproduct_type='im'

AND   time_min >= {jday_start}
AND   time_min <= {jday_stop}

ORDER BY time_min, granule_gid
""".format(
        target_name=target_name.replace("'", "\\'"),

        jday_start=timestamp_to_jday(t_started_at),
        jday_stop=timestamp_to_jday(t_stopped_at)

    )

    # AND   processing_level = 4

    #     query = """
    # SELECT DISTINCT target_name FROM apis.epn_core
    # """

    try:
        response = requests.post(api_url, {
            'REQUEST': 'doQuery',

            'LANG': 'ADQL',
            'QUERY': query,

            'TIMEOUT': '30',

            'FORMAT': 'VOTable/td'

        })

        response_xml = response.text

        import xml.etree.ElementTree as ET
        root = ET.fromstring(response_xml)
        namespaces = {'vo': 'http://www.ivoa.net/xml/VOTable/v1.3'}
        rows_xpath = "./vo:RESOURCE/vo:TABLE/vo:DATA/vo:TABLEDATA/vo:TR"
        rows = []
        for row in root.findall(rows_xpath, namespaces):
            rows.append({

                'time_min': jday_to_datetime(float(row[0].text)),
                'time_max': jday_to_datetime(float(row[1].text)),

                'thumbnail_url': row[2].text,
                'external_link': row[3].text,
            })

        return rows
    except Exception as e:
        print("Failed to retrieve auroral emissions :")
        print(e)

    # print(query)

def retrieve_amda_netcdf(orbiter, what, started_at, stopped_at):

    """

    Handles remote querying AMDA's API for URLs, and then downloading,
    extracting and caching the netCDF files.

    :param orbiter: key of the source in the YAML config
    :param what: either 'model' or 'orbit', a key in the config of the source
    :param started_at:
    :param stopped_at:
    :return: a list of local file paths to netCDF (.nc) files
    """

    url = config['amda'].format(
        dataSet=what,
        startTime=started_at.isoformat(),
        stopTime=stopped_at.isoformat()
    )

    hp_logger.info("Fetching remote gzip files list at '%s'." % url)

    retries = 0
    success = False

    errors = []

    remote_gzip_files = []
    while not success and retries < 3:
        try:

            response = urllib_request.urlopen(url)

            remote_gzip_files = json.loads(response.read())
            if not remote_gzip_files:
                raise Exception("Failed to fetch data at '%s'." % url)
            if remote_gzip_files == 'NODATASET':

                raise Exception("API says there's no dataset at '%s'." % url)
            if remote_gzip_files == 'ERROR':
                raise Exception("API returned an error at '%s'." % url)

            if remote_gzip_files == ['OUTOFTIME']:  # it happens

                return []
                # raise Exception("API says it's out of time at '%s'." % url)

            success = True
        except Exception as e:

            hp_logger.warning("Failed (%d/3) '%s' : %s" % (retries + 1, url, e.message))

            remote_gzip_files = []
            errors.append(e)

        finally:
            retries += 1
    if not remote_gzip_files:

        hp_logger.error("Failed to retrieve data from AMDA.")
        hp_logger.error("Failed to fetch gzip files list for %s at '%s' : %s" %
                        (orbiter, url, errors))

        abort(400, "Failed to fetch gzip files list for %s at '%s' : %s" %

              (orbiter, url, errors))

    else:
        remote_gzip_files = list(set(remote_gzip_files))

    hp_logger.debug("Fetched remote gzip files list : %s." % str(remote_gzip_files))

    local_gzip_files = []
    for remote_gzip_file in remote_gzip_files:

        # hotfixes to remove when fixed upstream @Myriam
        if remote_gzip_file in ['OUTOFTIME', 'ERROR']:
            continue  # sometimes half the response is okay, the other not

        if remote_gzip_file.endswith('/.gz'):

            continue  # this is just a plain bug

        remote_gzip_file = remote_gzip_file.replace('cdpp1', 'cdpp', 1)

        ################################################

        local_gzip_file = join(CACHE_DIR, get_local_filename(remote_gzip_file))

        local_gzip_files.append(local_gzip_file)
        if not isfile(local_gzip_file):

            hp_logger.debug("Retrieving '%s'..." % local_gzip_file)

            urllib_request.urlretrieve(remote_gzip_file, local_gzip_file)

            hp_logger.debug("Retrieved '%s'." % local_gzip_file)

        else:

            hp_logger.debug("Found '%s' in the cache." % local_gzip_file)

    local_netc_files = []
    for local_gzip_file in local_gzip_files:
        local_netc_file = local_gzip_file[0:-3]

        hp_logger.debug("Unzipping '%s'..." % local_gzip_file)

        success = True
        try:

            with gzip.open(local_gzip_file) as f:

                file_content = f.read()
                with open(local_netc_file, 'w+b') as g:
                    g.write(file_content)
        except Exception as e:
            success = False

            hp_logger.error("Cannot process gz file '%s' from '%s' : %s" %
                            (local_gzip_file, url, e))

            # Sometimes, the downloaded gz is corrupted, and CRC checks fail.
            # We want to delete the local gz file and try again next time.
            removefile(local_gzip_file)

        if success:

            local_netc_files.append(local_netc_file)

            hp_logger.debug("Unzipped '%s'." % local_gzip_file)

    return list(set(local_netc_files))  # remove possible dupes

# class DataParser:
#     """
#     Default data parser
#     A wip to try to handle code exeptions sanely.
#     """
#
#     # Override these using the model configuration
#     default_nc_keys = {
#         'hee': 'HEE',
#         'vtot': 'V',
#         'magn': 'B',
#         'temp': 'T',
#         'dens': 'N',
#         'pdyn': 'P_dyn',
#         'atse': 'Delta_angle',
#     }
#
#     def __init__(self, target, model):
#         self.target = target
#         self.model = model
#         pass
#
#     def _read_var(self, nc, _keys, _key, mandatory=False):
#         try:
#             return nc.variables[_keys[_key]]
#         except KeyError:
#             pass
#         if mandatory:
#             raise Exception("No variable '%s' found in NetCDF." % _keys[_key])
#         return [None] * len(nc.variables['Time'])  # slow -- use numpy?
#
#     def parse(self, cdf_handle):
#         nc_keys = self.default_nc_keys.copy()
#
#         times = cdf_handle.variables['Time']  # YYYY DOY HH MM SS .ms
#         data_v = self._read_var(cdf_handle, nc_keys, 'vtot')
#         data_b = self._read_var(cdf_handle, nc_keys, 'magn')
#         data_t = self._read_var(cdf_handle, nc_keys, 'temp')
#         data_n = self._read_var(cdf_handle, nc_keys, 'dens')
#         data_p = self._read_var(cdf_handle, nc_keys, 'pdyn')
#         data_a = self._read_var(cdf_handle, nc_keys, 'atse')
#
#         return zip()

def get_data_for_target(target_config, input_slug,
                        started_at, stopped_at):

    """
    :return: dict whose keys are datetime as str, values tuples of data
    """

    hp_logger.debug("Grabbing data for '%s'..." % target_config['slug'])

    try:

        models = target_config['models'][input_slug]

    except Exception as e:
        abort(500, "Invalid model configuration for '%s' : %s"

              % (target_config['slug'], str(e)))
    try:

        orbits = target_config['orbit']['models']

    except KeyError as e:
        orbits = []
        # abort(500, "Invalid orbit configuration for '%s' : %s"
        #       % (target_config['slug'], str(e)))

    precision = "%Y-%m-%dT%H"  # model and orbits times are only equal-ish

    orbit_data = {}  # keys are datetime as str, values arrays of XY

    for orbit in orbits:

        s0, s1 = _sta_sto(orbit, started_at, stopped_at)

        nc_keys = default_nc_keys.copy()
        if 'parameters' in orbit:
            nc_keys.update(orbit['parameters'])

        orbit_files = retrieve_amda_netcdf(
            target_config['slug'], orbit['slug'], s0, s1
        )
        for orbit_file in orbit_files:

            hp_logger.debug("%s: opening orbit NETCDF4 '%s'..." %
                            (target_config['name'], orbit_file))

            cdf_handle = Dataset(orbit_file, "r", format="NETCDF4")
            times = cdf_handle.variables['Time']  # YYYY DOY HH MM SS .ms

            data_hee = _read_var(cdf_handle, nc_keys, 'hee', mandatory=True)

            hp_logger.debug("%s: aggregating data from '%s'..." %
                            (target_config['name'], orbit_file))

            for ltime, datum_hee in zip(times, data_hee):

                try:

                    dtime = datetime_from_list(ltime)
                except Exception:

                    hp_logger.error("Failed to parse time from get__data_for_target %s." % ltime)

                    raise

                # Keep only what's in the interval

                if s0 <= dtime <= s1:

                    dkey = round_time(dtime, 60 * 60).strftime(precision)

                    orbit_data[dkey] = datum_hee
            cdf_handle.close()

    all_data = {}  # keys are datetime as str, values tuples of data

    for model in models:
        s0, s1 = _sta_sto(model, started_at, stopped_at)
        model_files = retrieve_amda_netcdf(
            target_config['slug'], model['slug'], s0, s1
        )

        nc_keys = default_nc_keys.copy()
        if 'parameters' in model:
            nc_keys.update(model['parameters'])

        if len(model_files) == 0:

            hp_logger.warning("No model data for '%s' '%s'."
                              % (target_config['slug'], model['slug']))

        for model_file in model_files:

            hp_logger.debug("%s: opening model NETCDF4 '%s'..." %
                            (target_config['name'], model_file))

            cdf_handle = Dataset(model_file, "r", format="NETCDF4")

            # log.debug(cdf_handle.variables.keys())

            times = cdf_handle.variables['Time']  # YYYY DOY HH MM SS .ms

            data_v = _read_var(cdf_handle, nc_keys, 'vtot')
            data_b = _read_var(cdf_handle, nc_keys, 'magn')
            data_t = _read_var(cdf_handle, nc_keys, 'temp')
            data_n = _read_var(cdf_handle, nc_keys, 'dens')
            data_p = _read_var(cdf_handle, nc_keys, 'pdyn')
            data_a = _read_var(cdf_handle, nc_keys, 'atse')

            # Usually:
            # Time, StartTime, StopTime, V, B, N, T, Delta_angle, P_dyn
            # Earth:
            # Time, BartelsNumber, ImfID, SwID, ImfPoints,
            # SwPoints, B_M_av, B_Vec_av, B_Theta_av,
            # B_Phi_av, B, T, N, V, Vlat, Vlon,
            # Alpha, RamP, E, Beta, Ma, Kp, R, DST,
            # AE, Flux, Flag, F10_Index, StartTime, StopTime

            # Don't ignore, but instead, compute a mean ?
            # Look for discretisation ? temporal drizzling ? 1D drizzling ?
            # The easy linear mean feels a bit rough too. Running mean too.
            # FIXME
            ignored_count = 0

            hp_logger.debug("%s: aggregating data from '%s'..." %
                            (target_config['name'], model_file))

            for ltime, datum_v, datum_b, datum_t, datum_n, datum_p, datum_a \
                    in zip(times, data_v, data_b, data_t, data_n, data_p, data_a):

                try:

                    dtime = datetime_from_list(ltime)
                except Exception:

                    hp_logger.error("Failed to parse time from %s." % ltime)

                    raise

                if not (s0 <= dtime <= s1):
                    continue  # Cull what's out of the interval

                droundtime = round_time(dtime, 60 * 60)

                dkey = droundtime.strftime(precision)

                x_hee = None
                y_hee = None
                if dkey in orbit_data:
                    x_hee = orbit_data[dkey][0] / ASTRONOMICAL_UNIT_IN_KM
                    y_hee = orbit_data[dkey][1] / ASTRONOMICAL_UNIT_IN_KM

                # First exception: V may be a vector instead of a scalar
                if hasattr(datum_v, '__len__'):
                    vrad = datum_v[0]
                    vtan = datum_v[1]
                    vtot = sqrt(vrad * vrad + vtan * vtan)
                else:  # eg: Earth
                    vrad = None
                    vtan = None
                    vtot = datum_v

                # Second exception: Earth is always at (1, 0)
                if target_config['slug'] == 'earth':
                    x_hee = 1
                    y_hee = 0

                # Third exception: B is a Vector3 or a Vector2 for Earth
                if target_config['slug'] == 'earth':
                    if model['slug'] == 'omni_hour_all':  # Vector3
                        datum_b = datum_b[0]
                    # if model['slug'] == 'ace_swepam_real':  # Vector2
                    #     datum_b = datum_b[0]

                    if model['slug'] == 'ace_swepam_real_1h':

                        datum_p = datum_n * vtot * vtot * 1.6726e-6

                    if vtot is None or isnan(vtot):
                        continue

                # Fourth exception: Earth temp is in K, not eV
                # @nandre: à vérifier ! Where is la constante de Boltzmann ?
                if target_config['slug'] == 'earth' and datum_t:
                    datum_t = datum_t / 11605.0

                # Keep adding exceptions here until you can't or become mad

                # Ignore bad data

                # if numpy.isnan(datum_t):
                #     continue

                # Crude/Bad drizzling condition : first found datum in the hour
                # gets to set the data. Improve this.
                if dkey not in all_data:
                    # /!. The Set value MUST be in the same order as PROPERTIES
                    all_data[dkey] = (
                        dtime.strftime("%Y-%m-%dT%H:%M:%S+00:00"),
                        vrad, vtan, vtot,
                        datum_b, datum_t, datum_p, datum_n, datum_a,
                        x_hee, y_hee
                    )
                else:
                    ignored_count += 1

            # Improve this loop so as to remove this stinky debug log
            if ignored_count > 0:

                hp_logger.debug("    Ignored %d datum(s) during ~\"drizzling\"."
                                % ignored_count)

            cdf_handle.close()

    return all_data

def generate_csv_contents(target_slug, input_slug, started_at, stopped_at):

    target_config = get_target_config(target_slug)

    hp_logger.debug("Crunching CSV contents for '%s'..." % target_config['name'])

    si = StringIO()

    cw = csv_writer(si)
    cw.writerow(PROPERTIES)

    all_data = get_data_for_target(
        target_config=target_config, input_slug=input_slug,
        started_at=started_at, stopped_at=stopped_at
    )

    hp_logger.debug("Writing and sorting CSV for '%s'..." % target_config['slug'])

    for dkey in sorted(all_data):
        cw.writerow(all_data[dkey])

    hp_logger.info("Generated CSV contents for '%s'." % target_config['slug'])

    return si.getvalue()

def generate_csv_contents_spz(target_slug, input_slug, started_at, stopped_at):

    global_now = datetime.datetime.now()

    hp_logger.debug(

        f"Generating csv content with Speasy for {target_slug}/{input_slug} between {started_at} {stopped_at}")

    target_config = get_target_config(target_slug)

    plasma_params = {}
    for plasma_struct in target_config['models'][input_slug]:
        plasma_params = {**plasma_params, **plasma_struct['parameters']}
    orbit_params = {}
    for orbit_struct in target_config['orbit']['models']:
        orbit_params = {**orbit_params, **orbit_struct['parameters']}
    parameters_dict = {**plasma_params, **orbit_params}

    hp_logger.info(f"Aggregating dataframes speazy parameters for '{input_slug}' to '{target_slug}'")

    list_df = []

    for _name, _id in parameters_dict.items():

        hp_logger.debug(f"Getting parameter id '{_id}' for '{_name}' ...")
        param_now = datetime.datetime.now()

        amda_dataset = amda.get_data(_id, started_at, stopped_at)
        if amda_dataset is None:

            hp_logger.debug(f" ... No data")

            continue
        _df = amda_dataset.to_dataframe()

        param_then = datetime.datetime.now()
        hp_logger.debug(f" ... Got {len(_df)} in {param_then - param_now}")

        if _name == 'xy_v':
            _df = _df.rename(columns={_df.columns[0]: 'vrad', _df.columns[1]: 'vtan'})
        elif _name == 'xy_hee':
            _df = _df.drop(_df.columns[2], axis=1)
            _df = _df.rename(columns={_df.columns[0]: 'xhee', _df.columns[1]: 'yhee'})

        elif _name == 'xy_b':
            _df = _df.drop(_df.columns[2], axis=1)
            _df = _df.rename(columns={_df.columns[0]: 'brad', _df.columns[1]: 'btan'})

        else:
            _df = _df.rename(columns={_df.columns[0]: _name})

        # _df = _df[~_df.index.duplicated()]

        # resample to frequency, for later concatenation
        _df = _df.resample('1H').mean()

        # _df = _df.loc[_df.first_valid_index():_df.last_valid_index()]

        list_df.append(_df)

    from math import sqrt
    final_df = pd.concat(list_df, axis=1)

    # build vtot param  if doesnt exist
    if 'vtot' not in final_df.columns \
            and 'vtan' in final_df.columns \
            and 'vrad' in final_df.columns:

        final_df['vtot'] = final_df.apply(lambda x: sqrt(x.vtan * x.vtan + x.vrad * x.vrad), axis=1)

    # Hardcoded earth coordinates
    if target_slug == 'earth':
        final_df['xhee'] = 1
        final_df['yhee'] = 0

    cols_ordered = ['vrad', 'vtan', 'vtot', 'btan', 'brad', 'temp', 'pdyn', 'dens', 'atse', 'xhee', 'yhee']

    cols_ordered = [_c for _c in cols_ordered if _c in final_df.columns]

    final_df = final_df[cols_ordered]
    final_df.index.name = 'time'

    global_then = datetime.datetime.now()
    hp_logger.debug(f"Took {global_then - global_now} to generate")

    return final_df.to_csv(date_format='%Y-%m-%dT%H:%M:%S+00:00', float_format="%.5f",
                           header=True,
                           sep=",")

def generate_csv_file_if_needed(target_slug, input_slug,
                                started_at, stopped_at):

    filename = "%s_%s_%s_%s.csv" % (target_slug, input_slug,
                                    started_at.strftime(FILE_DATE_FMT),
                                    stopped_at.strftime(FILE_DATE_FMT))

    local_csv_file = join(CACHE_DIR, filename)

    skip_generation = False

    # It needs to have more than one line to not be empty (headers)

    if isfile(local_csv_file):

        with open(local_csv_file) as f:

            skip_generation = len(f.readlines()) > 0

    if skip_generation:
        hp_logger.debug(f"{local_csv_file} already exists")
        return None
    else:
        hp_logger.debug(f"{local_csv_file} doesnt exist or is empty")

    # temporary switch while migrating each target to spz

    if target_slug in ['rosetta', 'juno', 'p67']:

        csv_generator = generate_csv_contents

    else:
        csv_generator = generate_csv_contents_spz

    hp_logger.info("Generating CSV '%s'..." % local_csv_file)
    try:
        with open(local_csv_file, mode="w+") as f:
            f.write(csv_generator(
                target_slug=target_slug, input_slug=input_slug,
                started_at=started_at, stopped_at=stopped_at
            ))
        hp_logger.info("Generation of '%s' done." % filename)
    except Exception as e:
        from sys import exc_info
        from traceback import extract_tb
        exc_type, exc_value, exc_traceback = exc_info()
        hp_logger.error(e)
        for trace in extract_tb(exc_traceback):
            hp_logger.error(trace)
        if isfile(local_csv_file):
            hp_logger.warning("Removing failed CSV '%s'..." % local_csv_file)
            removefile(local_csv_file)
        abort(500, "Failed creating CSV '%s' : %s" % (filename, e))

def remove_all_files(in_directory):
    """
    Will throw if something horrible happens.
    Does not remove recursively (could be done with os.walk if needed).
    Does not remove directories either.
    :param in_directory: absolute path to directory
    :return:
    """
    import os

    if not os.path.isdir(in_directory):
        raise ValueError("No directory to clean at '%s'.")

    removed_files = []
    for file_name in os.listdir(in_directory):
        file_path = os.path.join(in_directory, file_name)
        if os.path.isfile(file_path):
            os.remove(file_path)
            removed_files.append(file_path)

    return removed_files

def remove_files_created_before(date, in_directory):
    """

    Will throw if something horrible happens.
    Does not remove recursively (could be done with os.walk if needed).
    Does not remove directories either.

    :param date: datetime object

    :param in_directory: absolute path to directory

    :return:
    """
    import os
    import time

    secs = time.mktime(date.timetuple())

    if not os.path.isdir(in_directory):
        raise ValueError("No directory to clean at '%s'.")

    removed_files = []
    for file_name in os.listdir(in_directory):
        file_path = os.path.join(in_directory, file_name)

        if os.path.isfile(file_path):
            t = os.stat(file_path)
            if t.st_ctime < secs:
                os.remove(file_path)
                removed_files.append(file_path)

    return removed_files

def get_input_slug_from_query(inp=None):
    if inp is None:

        input_slug = request.args.get('input_slug', config['defaults']['input_slug'])

    else:
        input_slug = inp

    if input_slug not in [i['slug'] for i in config['inputs']]:

        input_slug = config['defaults']['input_slug']  # be tolerant instead of yelling loudly

    return input_slug

def get_default_sta_sto(sta=None, sto=None):

    # TODO : maybe get from configuration
    relative_after = config['defaults']['relative_after']
    relative_before = config['defaults']['relative_before']
    before = relativedelta(months=relative_before)
    after = relativedelta(months=relative_after)

    if sta is None and sto is None:
        today = datetime.datetime.now().replace(hour=0, minute=0, second=0)
        sta_date = today - before
        sto_date = today + after

    if sta is None and sto is not None:
        sto_date = dateparser.isoparse(sto)
        sta_date = sto_date - before

    if sta is not None and sto is None:
        sta_date = dateparser.isoparse(sta)
        sto_date = sta_date + after

    return sta_date, sto_date


def get_interval_from_query(_req=None):

    """

    Get the interval from the query as flask.request, or from defaults.

    Returns ISO date strings.

    """

    sta_default, sto_default  = get_default_sta_sto()

    # convert to string
    started_at_default = sta_default.strftime(FILE_DATE_FMT)
    stopped_at_default = sto_default.strftime(FILE_DATE_FMT)

    if _req is not None and 'Request' == _req.__class__.__name__:
        started_at_str = _req.args.get('started_at', started_at_default)
        stopped_at_str = _req.args.get('stopped_at', stopped_at_default)

    # convert back to proper format
    started_at_date = dateparser.isoparse(started_at_str)
    stopped_at_date = dateparser.isoparse(stopped_at_str)
    started_at = started_at_date.strftime(FILE_DATE_FMT)
    stopped_at = stopped_at_date.strftime(FILE_DATE_FMT)

    return started_at, stopped_at

def get_catalog_layers(input_slug, target_slug, started_at, stopped_at):

    """

    In the JSON file we have "columns" and "data".
    Of course, each JSON file has its own columns, with different conventions.

    
    :param input_slug: 
    :param target_slug: 
    :return: 
    """

    import json

    def _get_index_of_key(_data, _key):

        try:
            index = _data['columns'].index(_key)
        except ValueError:

            hp_logger.error("Key %s not found in columns of %s" % (_key, _data))

            raise
        return index

    try:
        started_at = datetime.datetime.strptime(started_at, FILE_DATE_FMT)
    except:
        abort(400, "Invalid started_at parameter : '%s'." % started_at)
    try:
        stopped_at = datetime.datetime.strptime(stopped_at, FILE_DATE_FMT)
    except:
        abort(400, "Invalid stopped_at parameter : '%s'." % stopped_at)

    catalog_layers = {}

    for config_layer in config['layers']['catalogs']:

        if 'data' not in config_layer:
            continue

        catalog_layers[config_layer['slug']] = []
        for cl_datum in config_layer['data']:
            if input_slug not in cl_datum:
                continue
            if target_slug not in cl_datum[input_slug]:
                continue

            if cl_datum[input_slug][target_slug] is None:
                # We used ~ in the config, there are no constraints
                constraints = []
            else:
                constraints = cl_datum[input_slug][target_slug]['constraints']

            with open(get_path("../data/catalog/%s" % cl_datum['file'])) as f:
                json_data = json.load(f)

                if 'start' not in cl_datum:

                    hp_logger.error("Invalid configuration: 'start' is missing.")

                    continue  # skip this

                if 'format' not in cl_datum:

                    cl_datum['format'] = CME_DATE_FMT
                    # log.error("Invalid configuration: 'format' is missing.")
                    # continue  # skip this

                start_index = _get_index_of_key(json_data, cl_datum['start'])
                if 'stop' not in cl_datum:
                    stop_index = start_index
                else:
                    stop_index = _get_index_of_key(json_data, cl_datum['stop'])

                for json_datum in json_data['data']:

                    validates_any_constraint = False

                    if 0 == len(constraints):
                        validates_any_constraint = True

                    for constraint in constraints:
                        validates_constraint = True

                        for key, possible_values in iter(constraint.items()):

                            actual_value = json_datum[_get_index_of_key(
                                json_data, key
                            )]
                            if actual_value not in possible_values:
                                validates_constraint = False
                                break
                        if validates_constraint:
                            validates_any_constraint = True
                            break
                    if not validates_any_constraint:

                        continue

                    start_time = json_datum[start_index]
                    stop_time = json_datum[stop_index]

                    start_time = datetime.datetime.strptime(
                        start_time, cl_datum['format']
                    )
                    stop_time = datetime.datetime.strptime(
                        stop_time, cl_datum['format']
                    )

                    if start_time < started_at:
                        continue

                    catalog_layers[config_layer['slug']].append({

                        'start': start_time.strftime(MOMENT_DATE_FMT),

                        'stop': stop_time.strftime(MOMENT_DATE_FMT),

                    })

    return catalog_layers

def get_hit_counter():
    hit_count_path = get_path("../VISITS")

    if isfile(hit_count_path):
        hit_count = int(open(hit_count_path).read())
    else:
        hit_count = 1

    return hit_count

def increment_hit_counter():
    hit_count_path = get_path("../VISITS")

    if isfile(hit_count_path):
        hit_count = int(open(hit_count_path).read())
        hit_count += 1
    else:
        hit_count = 1

    hit_counter_file = open(hit_count_path, 'w')
    hit_counter_file.write(str(hit_count))
    hit_counter_file.close()

    return hit_count

def update_spacepy():
    """

    Importing pycdf will fail if the toolbox is not up to date.

    """

    try:

        hp_logger.info("Updating spacepy's toolbox…")

        import spacepy.toolbox

        spacepy.toolbox.update()
    except Exception as e:

        hp_logger.error("Failed to update spacepy : %s." % e)

tpl_global_vars['visits'] = get_hit_counter()

# ROUTING #####################################################################

@app.route('/favicon.ico')

def favicon():  # we want it served from the root, not from static/

    return send_from_directory(
        join(app.root_path, 'static', 'img'),
        'favicon.ico', mimetype='image/vnd.microsoft.icon'
    )


@app.route("/")
@app.route("/home.html")
@app.route("/index.html")
def home():

    increment_hit_counter()

    parameters = PARAMETERS.values()

    parameters = sorted(parameters, key=lambda x: x['position'])

    input_slug = get_input_slug_from_query()

    targets = [t for t in config['targets'] if not t['locked']]

    started_at, stopped_at = get_interval_from_query(request)

    for i, target in enumerate(targets):
        targets[i]['catalog_layers'] = get_catalog_layers(

            input_slug, target['slug'], started_at, stopped_at

        )

    return render_view('home.html.jinja2', {

        # 'targets': config['targets'],
        'targets': targets,

        'parameters': parameters,

        'input_slug': input_slug,

        'started_at': started_at,
        'stopped_at': stopped_at,

        'planets': [s for s in config['targets'] if s['type'] == 'planet'],

        'probes': [s for s in config['targets'] if s['type'] == 'probe'],
        'comets': [s for s in config['targets'] if s['type'] == 'comet'],
        'visits': get_hit_counter(),

    })

@app.route("/about.html")
def about():

    import uuid

    increment_hit_counter()
    return render_view('about.html.jinja2', {

        'authors_emails': [a['mail'] for a in config['authors']],
        'uuid4': str(uuid.uuid4())[0:3],
        'visits': get_hit_counter(),

    })

@app.route("/help.html")
def help():
    return render_view('help.html.jinja2', {
        'visits': get_hit_counter(),
    })

@app.route("/<target>_<inp>_<started_at>_<stopped_at>.csv")
def download_target_csv(target, inp, started_at, stopped_at):

    """
    Grab data and orbit data for the specified `target`,
    rearrange it and return it as a CSV file.
    `started_at` and `stopped_at` should be UTC.

    `inp` is the input slug, omni or sa or sb.

    """

    check_target_config(target)

    try:

        started_at = datetime.datetime.strptime(started_at, FILE_DATE_FMT)

    except:
        abort(400, "Invalid started_at parameter : '%s'." % started_at)
    try:

        stopped_at = datetime.datetime.strptime(stopped_at, FILE_DATE_FMT)

    except:
        abort(400, "Invalid stopped_at parameter : '%s'." % stopped_at)

    input_slug = get_input_slug_from_query(inp=inp)

    filename = "%s_%s_%s_%s.csv" % (target, input_slug,
                                    started_at.strftime(FILE_DATE_FMT),
                                    stopped_at.strftime(FILE_DATE_FMT))

    local_csv_file = join(CACHE_DIR, filename)

    generate_csv_file_if_needed(
        target_slug=target, input_slug=input_slug,
        started_at=started_at, stopped_at=stopped_at
    )

    if not isfile(local_csv_file):
        abort(500, "Could not cache CSV file at '%s'." % local_csv_file)

    return send_from_directory(CACHE_DIR, filename)

@app.route("/<targets>_<inp>_<started_at>_<stopped_at>.tar.gz")
def download_targets_tarball(targets, inp, started_at, stopped_at):

    """

    Grab data and orbit data for each of the specified `targets`,
    in their own CSV file, and make a tarball of them.
    `started_at` and `stopped_at` should be UTC strings.

    Note: we do not use this route anymore, but let's keep it shelved for now.

    targets: string list of targets' slugs, separated by `-`.

    """

    separator = '-'

    targets = targets.split(separator)

    targets.sorty()

    targets_configs = []
    for target in targets:

        if not target:
            abort(400, "Invalid targets format : `%s`." % targets)

        targets_configs.append(get_target_config(target))

    if 0 == len(targets_configs):

        abort(400, "No valid targets specified. What are you doing?")

    try:

        started_at = datetime.datetime.strptime(started_at, FILE_DATE_FMT)

    except:
        abort(400, "Invalid started_at parameter : '%s'." % started_at)
    try:

        stopped_at = datetime.datetime.strptime(stopped_at, FILE_DATE_FMT)

    except:
        abort(400, "Invalid stopped_at parameter : '%s'." % stopped_at)

    sta = started_at.strftime(FILE_DATE_FMT)
    sto = stopped_at.strftime(FILE_DATE_FMT)

    input_slug = get_input_slug_from_query(inp=inp)

    gzip_filename = "%s_%s_%s_%s.tar.gz" % (
        separator.join(targets), input_slug, sta, sto
    )

    local_gzip_file = join(CACHE_DIR, gzip_filename)

    if not isfile(local_gzip_file):

        hp_logger.debug("Creating the CSV files for the tarball...")

        for target_config in targets_configs:

            filename = "%s_%s_%s_%s.csv" % (
                target_config['slug'], input_slug, sta, sto
            )

            local_csv_file = join(CACHE_DIR, filename)

            if not isfile(local_csv_file):
                with open(local_csv_file, mode="w+") as f:

                    f.write(generate_csv_contents(
                        target_slug=target_config['slug'],
                        started_at=started_at,
                        stopped_at=stopped_at,
                        input_slug=input_slug
                    ))

        hp_logger.debug("Creating the tarball '%s'..." % local_gzip_file)

        with tarfile.open(local_gzip_file, "w:gz") as tar:
            for target_config in targets_configs:

                filename = "%s_%s_%s_%s.csv" % (
                    target_config['slug'], input_slug, sta, sto
                )

                local_csv_file = join(CACHE_DIR, filename)

                tar.add(local_csv_file, arcname=filename)

    if not isfile(local_gzip_file):

        abort(500, "No tarball to serve. Looked at '%s'." % local_gzip_file)

    return send_from_directory(CACHE_DIR, gzip_filename)

@app.route("/<targets>_<inp>_<params>_<started_at>_<stopped_at>.nc")
def download_targets_netcdf(targets, inp, params, started_at, stopped_at):

    """

    NOTE : This is not used anymore.
    

    Grab data and orbit data for the specified `target`,

    rearrange it and return it as a NetCDF file.

    `started_at` and `stopped_at` are expected to be UTC.

    targets: string list of targets' slugs, separated by `-`.
    params: string list of targets' parameters, separated by `-`.
    """

    separator = '-'  # /!\ this char should never be in target's slugs

    targets = targets.split(separator)

    targets.sorty()

    targets_configs = []
    for target in targets:
        if not target:
            abort(400, "Invalid targets format : `%s`." % targets)
        targets_configs.append(get_target_config(target))
    if 0 == len(targets_configs):
        abort(400, "No valid targets specified. What are you doing?")

    params = params.split(separator)

    params.sorty()

    if 0 == len(params):
        abort(400, "No valid parameters specified. What are you doing?")
    if not is_list_in_list(params, PARAMETERS.keys()):
        abort(400, "Some parameters are not recognized in '%s'." % str(params))

    date_fmt = FILE_DATE_FMT

    try:
        started_at = datetime.datetime.strptime(started_at, date_fmt)
    except:
        abort(400, "Invalid started_at parameter : '%s'." % started_at)
    try:
        stopped_at = datetime.datetime.strptime(stopped_at, date_fmt)
    except:
        abort(400, "Invalid stopped_at parameter : '%s'." % stopped_at)
    sta = started_at.strftime(date_fmt)
    sto = stopped_at.strftime(date_fmt)

    input_slug = get_input_slug_from_query(inp=inp)

    nc_filename = "%s_%s_%s_%s_%s.nc" % (
        separator.join(targets), separator.join(params), input_slug, sta, sto
    )

    nc_path = join(CACHE_DIR, nc_filename)

    if not isfile(nc_path):

        hp_logger.debug("Creating the NetCDF file '%s'..." % nc_filename)

        nc_handle = Dataset(nc_path, "w", format="NETCDF4")
        try:

            nc_handle.description = "Model and orbit data for targets"  # todo

            nc_handle.history = "Created " + time.ctime(time.time())

            nc_handle.source = "Heliopropa (CDDP)"

            available_params = list(PROPERTIES)
            for target in targets_configs:
                target_slug = target['slug']

                hp_logger.debug("Adding group '%s' to the NetCDF..." % target_slug)

                nc_group = nc_handle.createGroup(target_slug)

                data = get_data_for_target(
                    target_config=target, input_slug=input_slug,
                    started_at=started_at, stopped_at=stopped_at
                )

                dkeys = sorted(data)

                dimension = 'dim_' + target_slug

                nc_handle.createDimension(dimension, len(dkeys))

                # TIME #

                nc_time = nc_group.createVariable('time', 'i8', (dimension,))

                nc_time.units = "hours since 1970-01-01 00:00:00"
                nc_time.calendar = "standard"
                times = []
                for dkey in dkeys:
                    time_as_string = data[dkey][0][:-6]  # remove +00:00 tail
                    date = datetime.datetime.strptime(time_as_string, date_fmt)
                    times.append(date2num(
                        date, units=nc_time.units, calendar=nc_time.calendar
                    ))
                nc_time[:] = times

                # SELECTED PARAMETERS #

                nc_vars = []
                indices = []
                for param in params:
                    indices.append(available_params.index(param))

                    nc_var = nc_group.createVariable(param, 'f8', (dimension,))

                    nc_var.units = PARAMETERS[param]['units']

                    nc_vars.append(nc_var)
                for i, nc_var in enumerate(nc_vars):
                    index = indices[i]
                    values = []
                    for dkey in dkeys:
                        dval = data[dkey]
                        values.append(dval[index])
                    nc_var[:] = values

                # ORBIT #

                nc_x = nc_group.createVariable('xhee', 'f8', (dimension,))

                nc_x.units = 'Au'

                nc_y = nc_group.createVariable('yhee', 'f8', (dimension,))

                nc_y.units = 'Au'
                values_x = []
                values_y = []

                index_x = available_params.index('xhee')
                index_y = available_params.index('yhee')

                for dkey in dkeys:
                    dval = data[dkey]
                    values_x.append(dval[index_x])
                    values_y.append(dval[index_y])
                nc_x[:] = values_x
                nc_y[:] = values_y

            hp_logger.debug("Writing NetCDF '%s'..." % nc_filename)

        except Exception:

            hp_logger.error("Failed to generate NetCDF '%s'." % nc_filename)

            raise

        finally:
            nc_handle.close()

    if not isfile(nc_path):
        abort(500, "No NetCDF to serve. Looked at '%s'." % nc_path)

    return send_from_directory(CACHE_DIR, nc_filename)

@app.route("/<targets>_<inp>_<started_at>_<stopped_at>.cdf")
def download_targets_cdf(targets, inp, started_at, stopped_at):

    """
    Grab data and orbit data for the specified `target`,
    rearrange it and return it as a CDF file.
    `started_at` and `stopped_at` are expected to be UTC.

    targets: string list of targets' slugs, separated by `-`.
    params: string list of targets' parameters, separated by `-`.
    """
    separator = '-'  # /!\ this char should never be in target's slugs
    targets = targets.split(separator)

    targets.sorty()

    targets_configs = []
    for target in targets:
        if not target:
            abort(400, "Invalid targets format : `%s`." % targets)
        targets_configs.append(get_target_config(target))
    if 0 == len(targets_configs):
        abort(400, "No valid targets specified. What are you doing?")

    params = PARAMETERS.keys()

    try:
        started_at = datetime.datetime.strptime(started_at, FILE_DATE_FMT)
    except:
        abort(400, "Invalid started_at parameter : '%s'." % started_at)
    try:
        stopped_at = datetime.datetime.strptime(stopped_at, FILE_DATE_FMT)
    except:
        abort(400, "Invalid stopped_at parameter : '%s'." % stopped_at)
    sta = started_at.strftime(FILE_DATE_FMT)
    sto = stopped_at.strftime(FILE_DATE_FMT)

    input_slug = get_input_slug_from_query(inp=inp)

    cdf_filename = "%s_%s_%s_%s.cdf" % (
        separator.join(targets), input_slug, sta, sto
    )

    cdf_path = join(CACHE_DIR, cdf_filename)

    if not isfile(cdf_path):

        hp_logger.debug("Creating the CDF file '%s'..." % cdf_filename)

        try:

            from spacepy import pycdf

        except ImportError:
            # If spacepy's toolbox is not up-to-date, importing will fail.
            # So, let's update and try again !
            update_spacepy()
            try:
                from spacepy import pycdf
            except ImportError as e:

                hp_logger.error("Failed to import pycdf from spacepy : %s" % e)

                raise

        except Exception as e:

            hp_logger.error("Failed to import pycdf from spacepy : %s" % e)

            raise

        try:

            cdf_handle = pycdf.CDF(cdf_path, masterpath='')

            targets_names = ', '.join([t['name'] for t in targets_configs])
            description = "Model and orbit data for %s." % targets_names

            cdf_handle.attrs['Description'] = description
            cdf_handle.attrs['Author'] = "Heliopropa.irap.omp.eu (CDPP)"
            cdf_handle.attrs['Created'] = str(time.ctime(time.time()))

            # fixme: Try changing the name from 00 to something relevant
            cdf_handle.attrs['Title'] = "Heliopropa - %s" % targets_names
            #######

            available_params = list(PROPERTIES)
            for target in targets_configs:
                target_slug = target['slug']

                data = get_data_for_target(
                    target_config=target, input_slug=input_slug,
                    started_at=started_at, stopped_at=stopped_at
                )

                dkeys = sorted(data)

                values = []

                for dkey in dkeys:

                    time_str = data[dkey][0][:-6]  # remove +00:00 tail
                    date = datetime.datetime.strptime(time_str, FILE_DATE_FMT)
                    values.append(date)
                kt = "%s_time" % target_slug

                cdf_handle.new(kt, type=pycdf.const.CDF_EPOCH)

                cdf_handle[kt] = values

                # cdf_handle[kt].attrs['FIELDNAM'] = "Time since 0 A.D"

                for param in params:
                    k = "%s_%s" % (target_slug, param)

                    values = []
                    i = available_params.index(param)

                    has_nones = False

                    for dkey in dkeys:

                        value = data[dkey][i]
                        if value is None:
                            has_nones = True
                        values.append(value)
                    if has_nones:
                        # PyCDF hates it when there are Nones.
                        # Since we don't know what value to set instead,
                        # let's skip the param altogether.
                        continue

                    cdf_handle[k] = values

                    attrs = cdf_handle[k].attrs
                    attrs['UNITS'] = PARAMETERS[param]['units']
                    attrs['LABLAXIS'] = PARAMETERS[param]['name']
                    attrs['FIELDNAM'] = PARAMETERS[param]['title']
                    if values:
                        attrs['VALIDMIN'] = min(values)
                        attrs['VALIDMAX'] = max(values)

                kx = "%s_xhee" % target_slug
                ky = "%s_yhee" % target_slug

                values_xhee = []
                values_yhee = []
                index_x = available_params.index('xhee')
                index_y = available_params.index('yhee')
                for dkey in dkeys:

                    value_xhee = data[dkey][index_x]
                    value_yhee = data[dkey][index_y]
                    # We've got some `None`s cropping up in the data sometimes.

                    # PyCDF does not digest Nones at all.

                    # While they solve this upstream, let's make an ugly fix!
                    if (value_xhee is not None) and (value_yhee is not None):
                        values_xhee.append(value_xhee)
                        values_yhee.append(value_yhee)
                    else:

                        values_xhee.append(0)
                        values_yhee.append(0)

                        hp_logger.warning("Orbit data for %s has NaNs." % target_slug)