path: root/store.py

''' Implements a store of disutils PKG-INFO entries, keyed off name, version.
'''
import sys, os, re, time, hashlib, random, types, math, stat, errno
import logging, string, datetime, calendar, binascii, urllib2, cgi
import posixpath
from collections import defaultdict
import cPickle as pickle
try:
    import psycopg2
except ImportError:
    pass
try:
    import sqlite3
    sqlite3_cursor = sqlite3.Cursor
except ImportError:
    sqlite3_cursor = type(None)
from defusedxml import ElementTree
import trove, openid2rp
from mini_pkg_resources import safe_name
from description_utils import processDescription
# csrf modules
import hmac
from base64 import b64encode
import openid.store.sqlstore
import oauth
import requests
import urlparse
import time
from functools import wraps

import tasks

# we import both the old and new (PEP 386) methods of handling versions since
# some version strings are not compatible with the new method and we can fall
# back on the old version
from distutils.version import LooseVersion
from verlib import NormalizedVersion, suggest_normalized_version

def enumerate(sequence):
    return [(i, sequence[i]) for i in range(len(sequence))]


chars = string.ascii_letters + string.digits

dist_file_types = [
    ('sdist',            'Source'),
    ('bdist_dumb',       '"dumb" binary'),
    ('bdist_rpm',        'RPM'),
    ('bdist_wininst',    'MS Windows installer'),
    ('bdist_msi',        'MS Windows MSI installer'),
    ('bdist_egg',        'Python Egg'),
    ('bdist_dmg',        'OS X Disk Image'),
    ('bdist_wheel',      'Python Wheel'),
]
dist_file_types_d = dict(dist_file_types)

# This could have been done with Postgres ENUMs, however
# a) they are not extensible, and
# b) they are not supported in other databases
class dependency:
    requires = 1
    provides = 2
    obsoletes = 3
    requires_dist = 4
    provides_dist = 5
    obsoletes_dist = 6
    requires_external = 7
    project_url = 8
    by_val = {}
for k,v in dependency.__dict__.items():
    if not isinstance(v, int):
        continue
    dependency.by_val[v] = k

keep_conn = False
connection = None
keep_trove = True


def normalize_package_name(n):
    "Return lower-cased version of safe_name of n."
    return safe_name(n).lower()

class ResultRow:
    '''Turn a tuple of row values into something that may be looked up by
    both column index and name.

    Also, convert any unicode values coming out of the database into UTF-8
    encoded 8-bit strings.
    '''
    def __init__(self, cols, info=None):
        self.cols = cols
        self.cols_d = {}
        for i, col in enumerate(cols):
            self.cols_d[col] = i
        self.info = info
    def __getitem__(self, item):
        if isinstance(item, int):
            value = self.info[item]
        else:
            n = self.cols_d[item]
            value = self.info[n]
        return self.decode(value)
    def __nonzero__(self):
        return bool(self.info)
    def items(self):
        return [(col, self.decode(self.info[i]))
                for i, col in enumerate(self.cols)]
    def as_dict(self):
        d = {}
        for i, col in enumerate(self.cols):
            d[col] = self.decode(self.info[i])
        return d
    def keys(self):
        return self.cols
    def values(self):
        return map(self.decode, self.info)

    def decode(self, value):
        if value is None:
            return value
        if isinstance(value, str):
            # decode strings stored as utf-8 into unicode
            return value.decode('utf-8')
        return value

def utf8getter(n):
    def utf8get(fields):
        if fields[n] is None: return fields[n]
        return fields[n].decode('utf-8', 'replace')

    return utf8get

def itemgetter(n):
    return lambda fields:fields[n]

def FastResultRow(cols):
    """Create a ResultRow-like class that has all fields already preparsed.
    Non-UTF-8-String columns must be suffixed with !."""
    getters = {}
    _keys = []
    for i, col in enumerate(cols.split()):
        if col[-1] == '!':
            col = col[:-1]
            getter = itemgetter(i)
        else:
            getter = utf8getter(i)
        _keys.append(col)
        getters[i] = getters[col] = getter
    class _FastResultRow:
        _getters = getters
        cols = _keys

        def __init__(self, cols, info):
            self.info = info

        def __getitem__(self, index):
            try:
                return self._getters[index](self.info)
            except KeyError:
                if isinstance(index, int):
                    raise IndexError, 'row index out of range'
                raise

        def __len__(self):
            return len(self.info)

        def __nonzero__(self):
            return bool(self.info)

        def as_dict(self):
            res = {}
            for key in self.cols:
                res[key] = self[key]
            return res

        def keys(self):
            return self.cols

        def values(self):
            res = [None] * len(self.info)
            for i in xrange(len(self.info)):
                res[i] = self[i]
            return res

        def items(self):
            res = [None] * len(self.info)
            for i, col in enumerate(self.cols):
                res[i] = (col, self[col])
            return res

    return _FastResultRow

def Result(cols, sequence, type=ResultRow):
    return [type(cols, item) for item in iter(sequence)]

def safe_execute(cursor, sql, params=None):
    """Tries to safely execute the given sql

    This will try to encode the incoming parameters into UTF-8 (where
    possible).

    """
    # Fast path to no param queries
    if params is None:
        return cursor.execute(sql)

    if isinstance(cursor, sqlite3_cursor):
        sql = sql.replace('%s', "?")

    # Encode every incoming param to UTF-8 if it's a string
    safe_params = []
    for param in params:
        if isinstance(param, unicode):
            safe_params.append(param.encode("UTF-8", "replace"))
        else:
            safe_params.append(param)
    return cursor.execute(sql, safe_params)

def binary(cursor, bytes):
    if isinstance(cursor, sqlite3_cursor):
        # XXX is this correct?
        return bytes
    return psycopg2.Binary(bytes)

class StorageError(Exception):
    pass

class Store:
    ''' Store info about packages, and allow query and retrieval.

        XXX update schema info ...
            Packages are unique by (name, version).
    '''
    def __init__(self, config, queue=None):
        self.config = config
        self.username = None
        self.userip = None
        self._conn = None
        self._cursor = None
        self._trove = None
        if self.config.database_driver == 'sqlite3':
            self.true, self.false = '1', '0'
            self.can_lock = False
        else:
            self.true, self.false = 'TRUE', 'FALSE'
            self.can_lock = True

        self.queue = queue

        self._changed_packages = set()

    def enqueue(self, func, *args, **kwargs):
        if self.queue is None:
            func(*args, **kwargs)
        else:
            self.queue.enqueue(func, *args, **kwargs)

    def last_id(self, tablename):
        ''' Return an SQL expression that returns the last inserted row,
        where the row is in the given table.
        '''
        if self.config.database_driver == 'sqlite3':
            return 'last_insert_rowid()'
        else:
            return "currval('%s_id_seq')" % tablename

    def trove(self):
        if not self._trove:
            self._trove = trove.Trove(self.get_cursor())
        return self._trove

    def add_journal_entry(self, name, version, action, submitted_date,
                                                submitted_by, submitted_from):
        cursor = self.get_cursor()
        safe_execute(cursor, """
            INSERT INTO journals
                (name, version, action, submitted_date, submitted_by,
                    submitted_from)
            VALUES
                (%s, %s, %s, %s, %s, %s)
        """, (name, version, action, submitted_date, submitted_by,
                                                            submitted_from))
        self._add_invalidation(name)

    def store_package(self, name, version, info):
        ''' Store info about the package to the database.

        If the name doesn't exist, we add a new package with the current
        user as the Owner of the package.

        If the version doesn't exist, we add a new release, hiding all
        previous releases.

        If the name and version do exist, we just edit (in place) and add a
        journal entry.
        '''
        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        cursor = self.get_cursor()
        # see if we're inserting or updating a package
        if not self.has_package(name):
            # insert the new package entry
            cols = 'name, normalized_name'
            vals = '%s, %s'
            args = (name, normalize_package_name(name))

            # if a bugtracker url is provided then insert it too
            if 'bugtrack_url' in info:
                cols += ', bugtrack_url'
                vals += ', %s'
                args += (info['bugtrack_url'], )

            sql = 'insert into packages (%s) values (%s)' % (cols, vals)
            safe_execute(cursor, sql, args)

            # journal entry
            self.add_journal_entry(name, None, "create", date,
                                                self.username, self.userip)

            # first person to add an entry may be considered owner - though
            # make sure they don't already have the Role (this might just
            # be a new version, or someone might have already given them
            # the Role)
            if not self.has_role('Owner', name):
                self.add_role(self.username, 'Owner', name)

            self._add_invalidation(None)

        # extract the Trove classifiers
        classifiers = info.get('classifiers', [])
        classifiers.sort()

        # now see if we're inserting or updating a release
        message = None
        relationships = defaultdict(set)
        old_cifiers = []
        html = None
        if self.has_release(name, version):
            # figure the changes
            existing = self.get_package(name, version)

            # handle the special vars that most likely won't have been
            # submitted
            for k in ('_pypi_ordering', '_pypi_hidden', 'bugtrack_url'):
                if not info.has_key(k):
                    info[k] = existing[k]

            # figure which cols in the table to update, if any
            specials = 'name version'.split()
            old = []
            cols = []
            vals = []
            for k, v in existing.items():
                if not info.has_key(k):
                    continue
                if k not in specials and info.get(k, None) != v:
                    if k == 'description':
                        cols.append('description_html')
                        html = processDescription(info[k])
                        vals.append(html)
                    old.append(k)
                    cols.append(k)
                    vals.append(info[k])
            vals.extend([name, version])

            # pull out the bugtrack_url and put it in the packages table
            # instead
            if 'bugtrack_url' in cols:
                sql = 'update packages set bugtrack_url=%s where name=%s'
                safe_execute(cursor, sql, (info['bugtrack_url'], name))
                del vals[cols.index('bugtrack_url')]
                cols.remove('bugtrack_url')

            # get old classifiers list
            old_cifiers = self.get_release_classifiers(name, version)
            old_cifiers.sort()
            if info.has_key('classifiers') and old_cifiers != classifiers:
                old.append('classifiers')

            # get old classifiers list
            for kind, specifier in self.get_release_dependencies(name, version):
                relationships[kind].add(specifier)
            for nkind, skind in dependency.by_val.items():
                # numerical kinds in relationships; string kinds in info
                try:
                    new_val = set(info[skind])
                except KeyError:
                    # value not provided
                    continue
                if relationships[skind] != new_val:
                    old.append(skind)

            # no update when nothing changes
            if not old:
                return None

            # create the journal/user message
            message = 'update %s'%', '.join(old)

            # update
            if cols:
                cols = ','.join(['%s=%%s'%x for x in cols])
                safe_execute(cursor, '''update releases set %s where name=%%s
                    and version=%%s'''%cols, vals)

            # journal the update
            self.add_journal_entry(name, version, message, date,
                                                self.username, self.userip)
        else:
            # round off the information (make sure name and version are in
            # the info dict)
            info['name'] = name
            info['version'] = version

            # figure the ordering
            info['_pypi_ordering'] = self.fix_ordering(name, version)

            # ReST-format the description
            if info.has_key('description'):
                info['description_html'] = html = processDescription(info['description'])
            else:
                info['description_html'] = ''

            # perform the insert
            cols = ('name version author author_email maintainer '
                    'maintainer_email home_page license summary description '
                    'description_html keywords platform requires_python '
                    'download_url _pypi_ordering _pypi_hidden').split()
            args = tuple([info.get(k, None) for k in cols])
            params = ','.join(['%s']*len(cols))
            scols = ','.join(cols)
            sql = 'insert into releases (%s) values (%s)'%(scols, params)
            safe_execute(cursor, sql, args)

            # journal entry
            self.add_journal_entry(name, version, "new release", date,
                                                self.username, self.userip)

            # first person to add an entry may be considered owner - though
            # make sure they don't already have the Role (this might just
            # be a new version, or someone might have already given them
            # the Role)
            if not self.has_role('Owner', name):
                self.add_role(self.username, 'Owner', name)

            # hide all other releases of this package if thus configured
            if self.get_package_autohide(name):
                safe_execute(cursor, 'update releases set _pypi_hidden=%s where '
                             'name=%s and version <> %s', (self.true, name, version))

        # add description urls
        if html:
            # grab the packages hosting_mode
            hosting_mode = self.get_package_hosting_mode(name)

            if hosting_mode in ["pypi-scrape-crawl", "pypi-scrape"]:
                self.update_description_urls(name, version, get_description_urls(html))

        # handle trove information
        if info.has_key('classifiers') and old_cifiers != classifiers:
            safe_execute(cursor, 'delete from release_classifiers where name=%s'
                ' and version=%s', (name, version))
            for classifier in classifiers:
                safe_execute(cursor, 'select id from trove_classifiers where'
                    ' classifier=%s', (classifier, ))
                trove_id = cursor.fetchone()[0]
                safe_execute(cursor, 'insert into release_classifiers '
                    '(name, version, trove_id) values (%s, %s, %s)',
                    (name, version, trove_id))

        # handle relationship specifiers
        for nkind, skind in dependency.by_val.items():
            if not info.has_key(skind) or relationships[nkind] == set(info[skind]):
                continue
            safe_execute(cursor, '''delete from release_dependencies where name=%s
                and version=%s and kind=%s''', (name, version, nkind))
            for specifier in info[skind]:
                safe_execute(cursor, '''insert into release_dependencies (name, version,
                    kind, specifier) values (%s, %s, %s, %s)''', (name,
                    version, nkind, specifier))

        self._add_invalidation(name)

        return message

    def fix_ordering(self, name, new_version=None):
        ''' Fix the _pypi_ordering column for a package's releases.

        If "new_version" is supplied, insert it into the sequence and
        return the ordering value for it.
        '''
        cursor = self.get_cursor()
        # load up all the version strings for this package and sort them
        safe_execute(cursor,
            'select version,_pypi_ordering from releases where name=%s',
            (name,))
        all_versions = list(cursor.fetchall())
        all_versions.append((new_version, None))

        versions = []
        current_ordering = {}
        norm_to_orig = {}
        try:
            # attempt to order using the PEP 386 implementation
            for version, ordering in all_versions:
                # fix up the 8% of PyPI versions that aren't PEP 386-strict
                norm_version = suggest_normalized_version(version)

                # give up if we can't even do that
                assert norm_version is not None

                norm_to_orig[norm_version] = version
                current_ordering[norm_version] = ordering
                versions.append(NormalizedVersion(norm_version))

                # just in case we did modify the new_version we need to update
                # it for later comparison
                if version == new_version:
                    new_version = norm_version
        except Exception:
            # fall back on the old distutils LooseVersion
            versions = []
            current_ordering = {}
            for version, ordering in all_versions:
                norm_to_orig[version] = version
                current_ordering[version] = ordering
                versions.append(LooseVersion(version))

        versions.sort()
        n = len(versions)

        # most packages won't need to renumber if we give them 100 releases
        max = 10 ** min(math.ceil(math.log10(n)), 2)

        # figure the ordering values for the releases
        for i, v in enumerate(versions):
            order = max + i
            v = str(v)
            if v == new_version:
                new_version = order
            elif order != current_ordering[v]:
                # map from normalised back to the actual project version
                v = norm_to_orig[v]
                # ordering has changed, update
                safe_execute(cursor, '''update releases set _pypi_ordering=%s
                    where name=%s and version=%s''', (order, name, v))

        self._add_invalidation(name)

        # return the ordering for this release
        return new_version

    def has_package(self, name):
        ''' Determine whether the package exists in the database.

            Returns true/false.
        '''
        cursor = self.get_cursor()
        sql = 'select count(*) from packages where name=%s'
        safe_execute(cursor, sql, (name, ))
        return int(cursor.fetchone()[0])

    def find_package(self, name):
        '''Return names of packages that differ from name only in case.'''
        cursor = self.get_cursor()
        name = normalize_package_name(name)
        sql = 'select name from packages where normalized_name=%s'
        safe_execute(cursor, sql, (name, ))
        return [r[0] for r in cursor.fetchall()]

    def has_release(self, name, version):
        ''' Determine whether the release exists in the database.

            Returns true/false.
        '''
        cursor = self.get_cursor()
        sql = 'select count(*) from releases where name=%s and version=%s'
        safe_execute(cursor, sql, (name, version))
        return int(cursor.fetchone()[0])

    def get_cheesecake_index(self, index_id):
        index = {'absolute': -1,
                 'relative': -1,
                 'subindices': []}

        cursor = self.get_cursor()

        sql = 'select absolute, relative from cheesecake_main_indices where id = %d'
        safe_execute(cursor, sql, (index_id,))

        index['absolute'], index['relative'] = cursor.fetchone()

        sql = 'select name, value, details from cheesecake_subindices where main_index_id = %d'
        safe_execute(cursor, sql, (index_id,))

        for name, value, details in cursor.fetchall():
            index['subindices'].append(dict(name=name, value=value, details=details))

        index['subindices'].sort(lambda x,y: cmp(x['name'], y['name']))

        return index

    _Package = FastResultRow('''name stable_version version author author_email
            maintainer maintainer_email home_page license summary description
            description_html keywords platform requires_python download_url
            _pypi_ordering! _pypi_hidden! cheesecake_installability_id!
            cheesecake_documentation_id! cheesecake_code_kwalitee_id! bugtrack_url!''')
    def get_package(self, name, version):
        ''' Retrieve info about the package from the database.

            Returns a mapping with the package info.
        '''
        cursor = self.get_cursor()
        sql = '''select packages.name as name, stable_version, version, author,
                  author_email, maintainer, maintainer_email, home_page,
                  license, summary, description, description_html, keywords,
                  platform, requires_python, download_url, _pypi_ordering,
                  _pypi_hidden,
                  cheesecake_installability_id,
                  cheesecake_documentation_id,
                  cheesecake_code_kwalitee_id, bugtrack_url
                 from packages, releases
                 where packages.name=%s and version=%s
                  and packages.name = releases.name'''
        safe_execute(cursor, sql, (name, version))
        return self._Package(None, cursor.fetchone())

    def get_package_urls(self, name, relative=None):
        '''Return all URLS (home, download, files) for a package,

        Return list of (link, rel, label) or None if there are no releases.
        '''
        cursor = self.get_cursor()
        result = []

        # grab the list of releases
        safe_execute(cursor, '''select version, home_page, download_url
            from releases where name=%s''', (name,))
        releases = list(cursor.fetchall())
        if not releases:
            return None

        # grab the packages hosting_mode
        hosting_mode = self.get_package_hosting_mode(name)

        if hosting_mode in ["pypi-scrape-crawl", "pypi-scrape"]:
            homerel = "homepage" if hosting_mode == "pypi-scrape-crawl" else "ext-homepage"
            downloadrel = "download" if hosting_mode == "pypi-scrape-crawl" else "ext-download"

            # homepage, download url
            for version, home_page, download_url in releases:
                # assume that home page and download URL are unescaped
                if home_page and home_page != 'UNKNOWN':
                    result.append((home_page, homerel, version + ' home_page'))
                if download_url and download_url != 'UNKNOWN':
                    result.append((download_url, downloadrel, version + ' download_url'))

        # uploaded files
        safe_execute(cursor, '''select filename, python_version, md5_digest
            from release_files where name=%s''', (name,))
        for fname, pyversion, md5 in cursor.fetchall():
            # Put files first, to have setuptools consider
            # them before going to other sites
            url = self.gen_file_url(pyversion, name, fname, relative) + \
                "#md5=" + md5
            result.insert(0, (url, "internal", fname))

        # urls from description - this also now includes explicit URLs provided
        # through the web interface
        for url in self.list_description_urls(name):
            # assume that description urls are escaped
            result.append((url['url'], None, url['url']))

        return result

    def get_uploaded_file_urls(self, name):
        cursor = self.get_cursor()
        urls = []
        safe_execute(cursor, '''select filename, python_version
            from release_files where name=%s''', (name,))
        for fname, pyversion in cursor.fetchall():
            urls.append(self.gen_file_url(pyversion, name, fname))
        return urls

    _Description_URLs = FastResultRow('id! version url')
    def list_description_urls(self, name, version=None):
        if version is None:
            sql = "SELECT id, version, url FROM description_urls WHERE name=%s"
            params = [name]
        else:
            sql = """SELECT id, version, url FROM description_urls
                WHERE name=%s AND version=%s"""
            params = [name, version]

        cursor = self.get_cursor()
        safe_execute(cursor, sql, params)
        return Result(None, cursor.fetchall(), self._Description_URLs)

    def add_description_url(self, name, version, url):
        cursor = self.get_cursor()
        safe_execute(cursor, """INSERT INTO description_urls (name, version, url)
             VALUES (%s, %s, %s)""", [name, version, url])

        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(name, version, "add url " + url, date,
                                                self.username, self.userip)

    def remove_description_url(self, url_id):
        cursor = self.get_cursor()
        sql = "SELECT name, version, url FROM description_urls WHERE id=%s"
        safe_execute(cursor, sql, [url_id])
        results = cursor.fetchone()
        if results is None:
            return

        name, version, url = results

        sql = "DELETE FROM description_urls WHERE id=%s"
        safe_execute(cursor, sql, [url_id])

        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(name, version, "remove url " + url, date,
                                                self.username, self.userip)

    def get_stable_version(self, name):
        ''' Retrieve the version marked as a package:s stable version.
        '''
        cursor = self.get_cursor()
        sql = 'select stable_version from packages where name=%s'
        safe_execute(cursor, sql, (name, ))
        return cursor.fetchone()[0]

    def top_packages(self, num=None):
        cursor = self.get_cursor()
        sql = """SELECT name, SUM(downloads) AS downloads FROM release_files
                    GROUP BY name ORDER BY downloads DESC"""
        if num is not None:
            sql += " LIMIT %s"
            safe_execute(cursor, sql, (num,))
        else:
            safe_execute(cursor, sql)

        return [(p[0], p[1]) for p in cursor.fetchall()]

    _Packages = FastResultRow('name stable_version')
    def get_packages(self):
        ''' Fetch the complete list of packages from the database.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, 'select name,stable_version from packages order by name')
        return Result(None, cursor.fetchall(), self._Packages)

    def get_packages_with_serial(self):
        cursor = self.get_cursor()
        safe_execute(cursor, "SELECT journals.name, max(id) FROM journals, packages WHERE journals.name = packages.name GROUP BY journals.name")
        return dict((n,i) for n, i in cursor.fetchall())

    def get_packages_utf8(self):
        '''Fetch the complete list of package names, UTF-8 encoded
        '''
        cursor = self.get_cursor()
        cursor.execute('select name from packages order by name')
        return [p[0] for p in cursor.fetchall()]

    _Journal = FastResultRow('action submitted_date! submitted_by submitted_from id!')
    def get_journal(self, name, version):
        ''' Retrieve info about the package from the database.

            Returns a list of the journal entries, giving those entries
            specific to the nominated version and those entries not specific
            to any version.
        '''
        cursor = self.get_cursor()
        # get the generic stuff or the stuff specific to the version
        sql = '''select action, submitted_date, submitted_by,
            submitted_from, id from journals where name=%s and (version=%s or
           version is NULL) order by submitted_date'''
        safe_execute(cursor, sql, (name, version))
        return Result(None, cursor.fetchall(), self._Journal)

    def count_packages(self):
        ''' Determine the number of packages registered with the index.
        '''
        cursor = self.get_cursor()
        cursor.execute('select count(*) from packages')
        return int(cursor.fetchone()[0])

    _Query_Packages = FastResultRow('name version summary _pypi_ordering!')
    def query_packages(self, spec, operator='and'):
        ''' Find packages that match the spec.

            Return a list of (name, version) tuples.
        '''
        if operator not in ('and', 'or'):
            operator = 'and'
        where = []
        for k, v in spec.items():
            if k not in ['name', 'version', 'author', 'author_email',
                         'maintainer', 'maintainer_email',
                         'home_page', 'license', 'summary',
                         'description', 'keywords', 'platform',
                         'download_url']:
                continue

            if type(v) != type([]): v = [v]
            # Quote the bits in the string that need it and then embed
            # in a "substring" search. Note - need to quote the '%' so
            # they make it through the python layer happily
            v = ['%%'+s.lower().replace("'", "''")+'%%' for s in v]

            # now add to the where clause
            where.append('(' + ' or '.join(["lower(%s) LIKE '%s'"%(k,
                s.encode('utf-8')) for s in v]) + ')')
        if where:
            where = ' %s '%operator.join(where)

        if '_pypi_hidden' in spec:
            if spec['_pypi_hidden'] in ('1', 1): v = self.true
            else: v = self.false
            if where:
                where += ' AND _pypi_hidden = %s'%v
            else:
                where = '_pypi_hidden = %s'%v

        # construct the SQL
        if where:
            where = ' where ' + where
        else:
            where = ''

        # do the fetch
        cursor = self.get_cursor()
        sql = '''select name, version, summary, _pypi_ordering
            from releases %s
            order by lower(name), _pypi_ordering'''%where
        safe_execute(cursor, sql)
        return Result(None, cursor.fetchall(), self._Query_Packages)

    _Classifiers = FastResultRow('classifier')
    def get_classifiers(self):
        ''' Fetch the list of valid classifiers from the database.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, 'select classifier from trove_classifiers'
            ' order by classifier')
        return Result(None, cursor.fetchall(), self._Classifiers)

    _ClassifierID = FastResultRow('classifier id')
    def get_classifier_ids(self, classifiers):
        '''Map list of classifiers to classifier IDs'''
        cursor = self.get_cursor()
        placeholders = ','.join(['%s'] * len(classifiers))
        safe_execute(cursor, 'select classifier, id from trove_classifiers '
           'where classifier in (%s)' % placeholders, classifiers)
        return dict(cursor.fetchall())

    _Release_Classifiers = FastResultRow('classifier trove_id!')
    def get_release_classifiers(self, name, version):
        ''' Fetch the list of classifiers for the release.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, '''select classifier, trove_id
            from trove_classifiers, release_classifiers where id=trove_id
            and name=%s and version=%s order by classifier''', (name, version))
        return Result(None, cursor.fetchall(), self._Release_Classifiers)

    _Release_Relationships = FastResultRow('specifier')
    def get_release_relationships(self, name, version, relationship):
        ''' Fetch the list of relationships of a particular type, either
            "requires", "provides" or "obsoletes".
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, '''select specifier from release_dependencies where
            name=%s and version=%s and kind=%s''', (name, version,
                                                    getattr(dependency, relationship)))
        return Result(None, cursor.fetchall(), self._Release_Relationships)

    _Release_Dependencies = FastResultRow('kind! specifier')
    def get_release_dependencies(self, name, version):
        '''Fetch all release dependencies of a release.'''
        cursor = self.get_cursor()
        safe_execute(cursor, '''select kind, specifier from release_dependencies
           where name=%s and version=%s''', (name, version))
        return Result(None, cursor.fetchall(), self._Release_Dependencies)

    def get_release_downloads(self, name, version):
        '''Fetch current download count for a release.'''
        cursor = self.get_cursor()
        safe_execute(cursor, '''select filename, downloads from release_files where
           name=%s and version=%s''', (name, version))
        return cursor.fetchall()

    _User_Packages_Roles = FastResultRow('role_name package_name')
    def get_user_packages(self, name):
        '''Fetch all packages and roles associated to user.'''
        cursor = self.get_cursor()
        safe_execute(cursor, '''select role_name, package_name from roles where
           user_name=%s''', (name,))
        return Result(None, cursor.fetchall(), self._User_Packages_Roles)

    _Package_Roles = FastResultRow('role_name user_name')
    def get_package_roles(self, name):
        ''' Fetch the list of Roles for the package.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, '''select role_name, user_name
            from roles where package_name=%s''', (name, ))
        return Result(None, cursor.fetchall(), self._Package_Roles)

    def get_package_autohide(self, name):
        cursor = self.get_cursor()
        safe_execute(cursor, 'select autohide from packages where name=%s',
                     [name])
        return cursor.fetchall()[0][0]

    def set_package_autohide(self, name, value):
        cursor = self.get_cursor()
        safe_execute(cursor, 'update packages set autohide=%s where name=%s',
                     [value, name])

    def get_package_hosting_mode(self, name):
        cursor = self.get_cursor()
        safe_execute(cursor, 'select hosting_mode from packages where name=%s',
                     [name])
        return cursor.fetchall()[0][0]

    def set_package_hosting_mode(self, name, value):
        if value not in ["pypi-explicit", "pypi-scrape", "pypi-scrape-crawl"]:
            raise ValueError("Invalid value for hosting_mode")

        cursor = self.get_cursor()
        safe_execute(cursor, 'update packages set hosting_mode=%s where name=%s',
                     [value, name])

        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(name, None, "update hosting_mode", date,
                                                    self.username, self.userip)

    def set_description(self, name, version, desc_text, desc_html,
            from_readme=False):
        cursor = self.get_cursor()
        safe_execute(cursor, '''update releases set description=%s,
            description_html=%s, description_from_readme=%s where name=%s
            and version=%s''', [desc_text, desc_html, from_readme, name,
            version])

        self._add_invalidation(name)

    def _get_package_url(self, name):
        name = name.split()[0]
        cursor = self.get_cursor()
        sql = 'select * from packages where name=%s'
        safe_execute(cursor, sql, (name, ))
        exists = cursor.fetchone() is not None
        if not exists:
            return None
        return self.config.url + '/' + name

    def get_package_requires_dist(self, name, version):
        cursor = self.get_cursor()
        safe_execute(cursor, '''select specifier from release_dependencies
            where name=%s and version=%s and kind=%s''', (name, version,
                                                          dependency.requires_dist))
        packages = []
        for package in cursor.fetchall():
            pack = {'name': package[0],
                    'href': self._get_package_url(package[0])}
            packages.append(pack)
        return packages

    def get_package_provides_dist(self, name, version):
        cursor = self.get_cursor()
        safe_execute(cursor, '''select specifier from release_dependencies
            where name=%s and version=%s and kind=%s''', (name, version,
                                                          dependency.provides_dist))
        packages = []
        for package in cursor.fetchall():
            pack = {'name': package[0],
                    'href': self._get_package_url(package[0])}
            packages.append(pack)
        return packages

    def get_package_obsoletes_dist(self, name, version):
        cursor = self.get_cursor()
        safe_execute(cursor, '''select specifier from release_dependencies
            where name=%s and version=%s and kind=%s''', (name, version,
                                                          dependency.obsoletes_dist))
        packages = []
        for package in cursor.fetchall():
            pack = {'name': package[0],
                    'href': self._get_package_url(package[0])}
            packages.append(pack)
        return packages

    def get_package_requires_external(self, name, version):
        cursor = self.get_cursor()
        safe_execute(cursor, '''select specifier from release_dependencies
            where name=%s and version=%s and kind=%s''', (name, version,
                                                          dependency.requires_external))
        return [package[0] for package in cursor.fetchall()]

    def get_package_project_url(self, name, version):
        cursor = self.get_cursor()
        safe_execute(cursor, '''select specifier from release_dependencies
            where name=%s and version=%s and kind=%s''', (name, version,
                                                          dependency.project_url))
        project_urls = []
        for project in cursor.fetchall():
            project_urls.append(project[0].split(','))
        return project_urls

    def get_unique(self, iterable):
        ''' Iterate over list of (name,version,date,summary) tuples
            and return list of unique (taking name and version into
            account) elements.
        '''
        d = {}
        L = []

        for name,version,date,summary in iterable:
            k = (name, version)
            if not d.has_key(k):
                L.append((name,version,date,summary))
                d[k] = 1

        return L

    _Updated_Releases = FastResultRow('name version submitted_date! summary')
    def updated_releases(self, since):
        '''Fetch all releases younger than "since" argument.
        '''
        assert isinstance(since, int)

        cursor = self.get_cursor()
        safe_execute(cursor, '''
            select j.name,j.version,j.submitted_date,r.summary
            from journals j, releases r
            where j.version is not NULL
                  and j.action = 'new release'
                  and j.name = r.name and j.version = r.version
                  and r._pypi_hidden = '''+self.false+'''
                  and j.submitted_date > %s
            order by submitted_date desc
        ''', (time.strftime('%Y-%m-%d %H:%M:%S +0000', time.gmtime(since)),))

        return Result(None, self.get_unique(cursor.fetchall()),
                self._Updated_Releases)

    _Changelog = FastResultRow('name version submitted_date! action id!')
    def changelog(self, since):
        '''Fetch (name, version, submitted_date, action, id) since 'since'
        argument.
        '''
        assert isinstance(since, int)

        cursor = self.get_cursor()
        safe_execute(cursor, '''
            select name, version, submitted_date, action, id
            from journals j
            where j.submitted_date > %s
        ''', (time.strftime('%Y-%m-%d %H:%M:%S +0000', time.gmtime(since)),))

        return Result(None, cursor.fetchall(), self._Changelog)

    def changelog_since_serial(self, since):
        '''Fetch (name, version, submitted_date, action, id) since 'since' id
        argument.
        '''
        assert isinstance(since, int)

        cursor = self.get_cursor()
        safe_execute(cursor, '''
            select name, version, submitted_date, action, id
            from journals j where j.id > %s
        ''', (since,))

        return Result(None, cursor.fetchall(), self._Changelog)

    def changelog_last_serial(self):
        '''Fetch the last event's serial id.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, 'select max(id) from journals j')
        return cursor.fetchone()[0]

    def last_serial_for_package(self, package):
        cursor = self.get_cursor()
        safe_execute(cursor, """
            SELECT id FROM journals WHERE name = %s ORDER BY id DESC LIMIT 1
        """, (package,))
        row = cursor.fetchone()
        if row:
            return row[0]

    def changed_packages(self, since):
        "Fetch list of names of packages changed 'since'"
        assert isinstance(since, int)
        cursor = self.get_cursor()
        safe_execute(cursor, 'select distinct(name) from journals where submitted_date > %s',
                     (time.strftime('%Y-%m-%d %H:%M:%S +0000', time.gmtime(since)),))
        return [r[0] for r in cursor.fetchall()]

    def changelog_last_hour(self):
        return self.changelog(int(time.time())-3600)

    _Latest_Packages = FastResultRow('name version submitted_date! summary')
    def latest_packages(self, num=40):
        '''Fetch "number" latest packages registered, youngest to oldest.
        '''
        cursor = self.get_cursor()
        # After the limited query below, we still have to do
        # filtering. Assume that doubling the number of records
        # we look for will still allow for sufficient room for
        # filtering out unneeded records. If this was wrong,
        # try again without limit.
        limit = ' limit %s' % (2*num)
        # This query is designed to run from the journals_latest_releases
        # index, doing a reverse index scan, then lookups in the releases
        # table to find the description and whether the package is hidden.
        # Postgres will only do that if the number of expected results
        # is "small".

        statement = '''
            select j.name, r.version, j.submitted_date%s, r.summary
              from releases r
                   JOIN (SELECT name, max(submitted_Date) submitted_date
                         FROM  journals GROUP BY name) j ON j.name = r.name
             where r.version is not NULL
               and not r._pypi_hidden
               and r.name in (SELECT name FROM journals
                              WHERE  action='create'
                              ORDER BY submitted_date DESC %%s)
             order by j.submitted_date desc'''

        if self.config.database_driver == 'sqlite3':
            statement = statement % ' as "sd [timestamp]"'
        else:
            statement = statement % ''

        #print ' '.join((statement % limit).split())
        safe_execute(cursor, statement % limit)
        result = Result(None, self.get_unique(cursor.fetchall())[:num],
                self._Latest_Packages)
        if len(result) == num:
            return result
        # try again without limit
        safe_execute(cursor, statement % '')
        return Result(None, self.get_unique(cursor.fetchall())[:num],
                self._Latest_Packages)

    _Latest_Releases = FastResultRow('name version submitted_date! summary')
    def latest_releases(self, num=40):
        ''' Fetch "number" latest releases, youngest to oldest.
        '''
        cursor = self.get_cursor()
        # After the limited query below, we still have to do
        # filtering. Assume that doubling the number of records
        # we look for will still allow for sufficient room for
        # filtering out unneeded records. If this was wrong,
        # try again without limit.
        limit = ' limit %s' % (2*num)
        # This query is designed to run from the journals_latest_releases
        # index, doing a reverse index scan, then lookups in the releases
        # table to find the description and whether the package is hidden.
        # Postgres will only do that if the number of expected results
        # is "small".
        statement = '''
             select j.name, j.version, j.submitted_date, r.summary
             from (select name,version,submitted_date from journals
             where version is not null and action='new release'
             order by submitted_date desc %s) j, releases r
             where  j.name=r.name and j.version=r.version
             and not r._pypi_hidden order by j.submitted_date desc'''
        #print ' '.join((statement % limit).split())
        safe_execute(cursor, statement % limit)
        result = Result(None, self.get_unique(cursor.fetchall())[:num],
                self._Latest_Releases)
        if len(result) == num:
            return result
        # try again without limit
        safe_execute(cursor, statement % '')
        return Result(None, self.get_unique(cursor.fetchall())[:num],
                self._Latest_Releases)

    _Latest_Updates = FastResultRow('name version submitted_date! summary')
    def latest_updates(self, num=20):
        ''' Fetch "number" latest updates, youngest to oldest.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, '''
            select j.name,j.version,j.submitted_date,r.summary
            from journals j, releases r
            where j.version is not NULL
                  and j.name = r.name and j.version = r.version
                  and r._pypi_hidden = '''+self.false+'''
            order by submitted_date desc
        ''')

        return Result(None, self.get_unique(cursor.fetchall())[:num],
                self._Latest_Updates)

    _Latest_Release = FastResultRow('''name version submitted_date! summary
            _pypi_hidden! _pypi_ordering!''')
    def get_latest_release(self, name, hidden=None):
        ''' Fetch all releses for the package name, including hidden.

        The "latest" version determined by ordering version numbers, not by
        submission date.
        '''
        args = [name, name]
        if hidden is not None:
            args.append(hidden)
            hidden = 'and _pypi_hidden = %s'
        else:
            hidden = ''
        cursor = self.get_cursor()
        safe_execute(cursor, '''
            select r.name as name, r.version as version, j.submitted_date,
                r.summary as summary, _pypi_hidden, _pypi_ordering
            from journals j, releases r
            where j.version is not NULL
                  and j.action = 'new release'
                  and j.name = %%s
                  and r.name = %%s
                  and j.version = r.version
                  %s
            order by _pypi_ordering desc
        '''%hidden, tuple(args))
        res = cursor.fetchall()
        if res is None:
            return []
        return Result(None, res, self._Latest_Release)

    _Package_Releases = FastResultRow('name version summary _pypi_hidden!')
    def get_package_releases(self, name, hidden=None):
        ''' Fetch all releses for the package name, including hidden.
        '''
        args = [name]
        if hidden is not None:
            args.append(hidden)
            hidden = 'and _pypi_hidden = %s'
        else:
            hidden = ''
        cursor = self.get_cursor()
        safe_execute(cursor, '''
            select name, version, summary, _pypi_hidden
            from releases
            where name = %%s %s
            order by _pypi_ordering desc
        '''%hidden, tuple(args))
        res = cursor.fetchall()
        if res is None:
            return []
        return Result(None, res, self._Package_Releases)

    def update_description_urls(self, name, version, urls):
        cursor = self.get_cursor()
        safe_execute(cursor, '''delete from description_urls
        where name=%s and version=%s''', (name, version))
        for url in urls:
            url = url.encode('utf-8')
            safe_execute(cursor, '''insert into description_urls(name, version, url)
            values(%s, %s, %s)''', (name, version, url))

        self._add_invalidation(name)

    def updateurls(self):
        cursor = self.get_cursor()
        safe_execute(cursor, '''select name, version, description_html from
                releases where description_html is not null''')
        for name, version, desc in cursor.fetchall():
            urls = get_description_urls(desc)
            self.update_description_urls(name, version, urls)

    def updateurls2(self):
        cursor = self.get_cursor()
        safe_execute(cursor, 'select name, version, url from description_urls')
        for name, version, url in cursor.fetchall():
            url2 = xmlescape(url)
            if url==url2:continue
            safe_execute(cursor, 'update description_urls set url=%s where name=%s and version=%s and url=%s',
                         (url2, name, version, url))

        self._add_invalidation(name)

    def update_normalized_text(self):
        cursor = self.get_cursor()
        safe_execute(cursor, 'select name from packages')
        for name, in cursor.fetchall():
            safe_execute(cursor, 'update packages set normalized_name=%s where name=%s',
                         [normalize_package_name(name), name])

        self._add_invalidation(name)

    def update_description_html(self, name):
        cursor = self.get_cursor()
        safe_execute(cursor, 'select version,description from releases where name=%s', (name,))
        for version, description in cursor.fetchall():
            safe_execute(cursor, 'update releases set description_html=%s where name=%s and version=%s',
                         (processDescription(description), name, version))

        self._add_invalidation(name)

    def remove_release(self, name, version):
        ''' Delete a single release from the database.
        '''
        cursor = self.get_cursor()

        self._add_invalidation(name)

        # delete the files
        for file in self.list_files(name, version):
            os.remove(self.gen_file_path(file['python_version'], name,
                file['filename']))

        # delete ancillary table entries
        for tab in ('files', 'dependencies', 'classifiers'):
            safe_execute(cursor, '''delete from release_%s where
                name=%%s and version=%%s'''%tab, (name, version))
        safe_execute(cursor, 'delete from description_urls where name=%s and version=%s',
               (name, version))

        # delete releases table entry
        safe_execute(cursor, 'delete from releases where name=%s and version=%s',
            (name, version))

        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(name, version, "remove", date,
                                                    self.username, self.userip)

    def remove_package(self, name):
        ''' Delete an entire package from the database.
        '''
        cursor = self.get_cursor()
        for release in self.get_package_releases(name):
            for file in self.list_files(name, release['version']):
                os.remove(self.gen_file_path(file['python_version'], name,
                    file['filename']))

        # delete ancillary table entries
        for tab in ('files', 'dependencies', 'classifiers'):
            safe_execute(cursor, 'delete from release_%s where name=%%s'%tab,
                (name, ))

        safe_execute(cursor, 'delete from description_urls where name=%s', (name,))
        safe_execute(cursor, 'delete from releases where name=%s', (name,))
        safe_execute(cursor, 'delete from journals where name=%s', (name,))
        safe_execute(cursor, 'delete from roles where package_name=%s', (name,))
        safe_execute(cursor, 'delete from packages where name=%s', (name,))

        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(name, None, "remove", date,
                                                    self.username, self.userip)

        self._add_invalidation(name)
        self._add_invalidation(None)

    def rename_package(self, old, new):
        ''' Rename a package. Relies on cascaded updates.
        '''
        cursor = self.get_cursor()
        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        safe_execute(cursor, '''update packages
        set name=%s, normalized_name=%s where name=%s''',
                     (new, normalize_package_name(new), old))
        safe_execute(cursor, '''update journals set name=%s where name=%s''',
                     (new, old))
        # move all files on disk
        sql = '''select python_version, filename
            from release_files where name=%s'''
        safe_execute(cursor, sql, (new,))
        for pyversion, filename in cursor.fetchall():
            oldname = self.gen_file_path(pyversion, old, filename)
            newname = self.gen_file_path(pyversion, new, filename)
            if not os.path.exists(oldname):
                continue
            dirpath = os.path.split(newname)[0]
            if not os.path.exists(dirpath):
                os.makedirs(dirpath)
            os.rename(oldname, newname)

        self.add_journal_entry(new, None, "rename from %s" % old, date,
                                                    self.username, self.userip)

        self._add_invalidation(new)
        self._add_invalidation(old)
        self._add_invalidation(None)

    def save_cheesecake_score(self, name, version, score_data):
        '''Save Cheesecake score for a release.
        '''
        cursor = self.get_cursor()

        execute = lambda query, *args: safe_execute(cursor, query, args)
        fetchone = cursor.fetchone

        def save_main_index(main_index):
            # Insert main_index into main_indices table.
            execute('''INSERT INTO cheesecake_main_indices
                       (absolute, relative)
                       VALUES (%d, %d)''',
                    main_index[0],
                    main_index[1])
            execute('SELECT last_value FROM cheesecake_main_indices_id_seq')
            main_index_id = fetchone()[0]

            # Insert each of its subindices.
            for sub_name, sub_value, sub_details in main_index[2]:
                execute('''INSERT INTO cheesecake_subindices
                           VALUES (%d, %s, %d, %s)''',
                        main_index_id,
                        sub_name,
                        sub_value,
                        sub_details)

            return main_index_id

        def release_exists(name, version):
            execute('''SELECT *
                       FROM releases
                       WHERE name = %s AND version = %s''',
                    name,
                    version)

            if fetchone():
                return True
            return False

        def remove_indices_for_release(name, version):
            execute('''SELECT cheesecake_installability_id,
                              cheesecake_documentation_id,
                              cheesecake_code_kwalitee_id
                       FROM releases
                       WHERE name = %s AND version = %s''',
                    name,
                    version)

            main_index_ids = fetchone()
            for index in main_index_ids:
                execute('''DELETE FROM cheesecake_subindices
                           WHERE main_index_id = %d''',
                        index)
                execute('''UPDATE releases
                           SET cheesecake_installability_id=NULL,
                               cheesecake_documentation_id=NULL,
                               cheesecake_code_kwalitee_id=NULL
                           WHERE name = %s AND version = %s''',
                        name,
                        version)
                execute('''DELETE FROM cheesecake_main_indices
                           WHERE id = %d''',
                        index)

        def insert_score_for_release(name, version, installability_id, documentation_id, code_kwalitee_id):
            execute('''UPDATE releases
                       SET cheesecake_installability_id=%d,
                           cheesecake_documentation_id=%d,
                           cheesecake_code_kwalitee_id=%d
                       WHERE name = %s AND version = %s''',
                    installability_id,
                    documentation_id,
                    code_kwalitee_id,
                    name,
                    version)

        installability_id = save_main_index(score_data['INSTALLABILITY'])
        documentation_id = save_main_index(score_data['DOCUMENTATION'])
        code_kwalitee_id = save_main_index(score_data['CODE_KWALITEE'])

        # If score for a release already exist, remove it first.
        if release_exists(name, version):
            remove_indices_for_release(name, version)

        insert_score_for_release(name, version, installability_id, documentation_id, code_kwalitee_id)

        self._add_invalidation(name)


    #
    # Users interface
    #
    def has_user(self, name, case_sensitive=True):
        ''' Determine whether the user exists in the database.

            Returns true/false.
        '''
        if case_sensitive:
            sql = "select count(*) from users where name=%s"
        else:
            sql = "select count(*) from users where LOWER(name)=LOWER(%s)"
        cursor = self.get_cursor()
        safe_execute(cursor, sql, (name,))
        return int(cursor.fetchone()[0])

    def store_user(self, name, password, email, gpg_keyid="", otk=True):
        ''' Store info about the user to the database.

            The "password" argument is passed in cleartext and sha-ed
            before storage.

            New user entries create a rego_otk entry too and return the OTK.
        '''
        cursor = self.get_cursor()
        if self.has_user(name, case_sensitive=False):
            if password:
                # update existing user, including password
                password = self.config.passlib.encrypt(password)
                safe_execute(cursor,
                   'update users set password=%s, email=%s where name=%s',
                    (password, email, name))
            else:
                # update existing user - but not password
                safe_execute(cursor, 'update users set email=%s where name=%s',
                    (email, name))
            if gpg_keyid is not None:
                safe_execute(cursor, 'update users set gpg_keyid=%s where name=%s',
                    (gpg_keyid, name))
            return None

        # Make sure email addresses are unique
        safe_execute(cursor, 'select count(*) from users where lower(email)=%s',
                     (email.lower(),))
        if cursor.fetchone()[0] > 0:
            raise ValueError, "Email address already belongs to a different user"

        password = self.config.passlib.encrypt(password)

        # new user
        safe_execute(cursor,
           'insert into users (name, password, email) values (%s, %s, %s)',
           (name, password, email))
        if not otk:
            return None
        otk = ''.join([random.choice(chars) for x in range(32)])
        safe_execute(cursor, 'insert into rego_otk (name, otk, date) values (%s, %s, current_timestamp)',
            (name, otk))
        return otk

    _User = FastResultRow('name password email gpg_keyid last_login!')
    def get_user(self, name, case_sensitive=True):
        ''' Retrieve info about the user from the database.

            Returns a mapping with the user info or None if there is no
            such user.
        '''
        cursor = self.get_cursor()
        if case_sensitive:
            sql = '''select name, password, email, gpg_keyid, last_login
                from users where name=%s'''
        else:
            sql = '''select name, password, email, gpg_keyid, last_login
                from users where lower(name)=lower(%s)'''
        safe_execute(cursor, sql, (name,))
        return self._User(None, cursor.fetchone())

    def get_user_by_email(self, email):
        ''' Retrieve info about the user from the database, looked up by
            email address.

            Returns a mapping with the user info or None if there is no
            such user.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, '''select name, password, email, gpg_keyid
            from users where lower(email)=%s''', (email.lower(),))
        return self._User(None, cursor.fetchone())

    def get_user_by_openid(self, openid):
        ''' Retrieve info about the user from the database, looked up by
            email address.

            Returns a mapping with the user info or None if there is no
            such user.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, '''select users.name, password, email, gpg_keyid
            from users,openids where users.name=openids.name
            and openids.id=%s''', (openid,))
        return self._User(None, cursor.fetchone())

    _Users = FastResultRow('name email')
    def get_users(self):
        ''' Fetch the complete list of users from the database.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, 'select name,email from users order by lower(name)')
        return Result(None, cursor.fetchall(), self._Users)

    _Openid = FastResultRow('id')
    def get_openids(self, username):
        cursor = self.get_cursor()
        safe_execute(cursor, 'select id from openids where name=%s', (username,))
        return Result(None, cursor.fetchall(), self._Openid)

    _Sshkey = FastResultRow('id! key')
    def get_sshkeys(self, username):
        '''Fetch the list of SSH keys for a user.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, 'select id,key from sshkeys where name=%s', (username,))
        return Result(None, cursor.fetchall(), self._Sshkey)

    def add_sshkey(self, username, key):
        '''Add a new SSH key for a user.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, 'insert into sshkeys(name, key) values(%s, %s)', (username, key))

    def delete_sshkey(self, id):
        '''Delete an SSH key given by ID.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, 'delete from sshkeys where id=%s', (id,))

    def has_role(self, role_name, package_name=None, user_name=None):
        ''' Determine whether the current user has the named Role for the
            named package.
        '''
        if user_name is None:
            user_name = self.username
        if user_name is None:
            return 0
        sql = '''select count(*) from roles where user_name=%s and
            role_name=%s and (package_name=%s or package_name is NULL)'''
        cursor = self.get_cursor()
        safe_execute(cursor, sql, (user_name, role_name, package_name))
        return int(cursor.fetchone()[0])

    def add_role(self, user_name, role_name, package_name):
        ''' Add a role to the user for the package.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, '''
            insert into roles (user_name, role_name, package_name)
            values (%s, %s, %s)''', (user_name, role_name, package_name))
        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(
            package_name, None, "add %s %s" % (role_name, user_name), date,
            self.username, self.userip)

    def delete_role(self, user_name, role_name, package_name):
        ''' Delete a role for the user for the package.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, '''
            delete from roles where user_name=%s and role_name=%s
            and package_name=%s''', (user_name, role_name, package_name))
        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(
                package_name, None, "remove %s %s" % (role_name, user_name),
                date, self.username, self.userip)

    def delete_otk(self, otk):
        ''' Delete the One Time Key.
        '''
        safe_execute(self.get_cursor(), "delete from rego_otk where otk=%s",
                     (otk,))

    def get_otk(self, username):
        ''' Retrieve the One Time Key for the user.

        Username must be a case-sensitive match.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, 'select otk from rego_otk where name=%s',
            (username, ))
        res = cursor.fetchone()
        if res is None:
            return ''
        return res[0]

    def get_user_by_otk(self, otk):
        '''Find a user by its otk.
        '''
        cursor = self.get_cursor()
        safe_execute(cursor, "select name from rego_otk where otk=%s", (otk, ))
        res = cursor.fetchone()
        if res is None:
            return ''
        return res[0]

    _User_Packages = FastResultRow('package_name')
    def user_packages(self, user, only_owner=False):
        ''' Retrieve package info for all packages of a user
        '''
        cursor = self.get_cursor()
        owner_sql = ''
        if only_owner:
            owner_sql = "and roles.role_name='Owner'"
        sql = '''select distinct(package_name),lower(package_name) from roles
                 where roles.user_name=%s and package_name is not NULL
                 ''' + owner_sql + '''
                 order by lower(package_name)'''
        safe_execute(cursor, sql, (user,))
        res = cursor.fetchall()
        if res is None:
            res = []
        return Result(None, res, self._User_Packages)

    def delete_user(self, user):
        '''Delete a user. Return None.'''
        cursor = self.get_cursor()
        # delete all maintainer roles
        safe_execute(cursor,
                     '''delete from roles where role_name='Maintainer'
                        and user_name=%s''',
                     (user,))
        # point all journal entries to the "deleted user"
        safe_execute(cursor,
                     '''update journals set submitted_by='deleted user' where submitted_by=%s''',
                     (user,))
        # delete all cookies
        safe_execute(cursor,
                     '''delete from cookies where name=%s''',
                     (user,))
        # every other reference should either be cascading,
        # or it's a bug to break it

        # delete user account itself
        safe_execute(cursor, 'delete from users where name=%s', (user,))

    #
    # Trove
    #

    def check_trove(self):
        cursor = self.get_cursor()
        trove = self.trove()
        # Verify that all l2, l3, l4 fields are set properly
        for field, depth in (('l2', 2), ('l3', 3), ('l4', 4), ('l5', 5)):
            cursor.execute('select id from trove_classifiers where %s is null' % field)
            for id, in cursor.fetchall():
                t = trove.trove[id]
                if len(t.path_split) < depth:
                    value = 0
                else:
                    value = trove.getid(t.path_split[:depth])
                cursor.execute('update trove_classifiers set %s=%d where id=%d' % (field, value, id))

    def browse_tally(self):
        import time
        cursor = self.get_cursor()
        t = self.trove()
        cursor.execute("select value from timestamps where name='browse_tally'")
        date = cursor.fetchone()[0]
        if time.time() - time.mktime(date.timetuple()) > 10*60:
            # Regenerate tally. First, release locks we hold on the timestamps
            self._conn.commit()
            # Clear old tally
            if self.can_lock:
                cursor.execute("lock table browse_tally")
            cursor.execute("delete from browse_tally")
            # Regenerate tally; see browse() below
            cursor.execute("""insert into browse_tally
            select res.l2, count(*) from (select t.l2, rc.name, rc.version
            from trove_classifiers t, release_classifiers rc, releases r
            where rc.name=r.name and rc.version=r.version and not r._pypi_hidden and rc.trove_id=t.id
            group by t.l2, rc.name, rc.version) res group by res.l2""")
            cursor.execute("update timestamps set value=current_timestamp where name='browse_tally'")
            self._conn.commit()
        cursor.execute("select trove_id, tally from browse_tally")
        return [], cursor.fetchall()

    def browse(self, selected_classifiers):
        t = self.trove()
        cursor = self.get_cursor()
        if not selected_classifiers:
            # This is not used; see browse_tally above
            tally = """select res.l2, count(*) from (select t.l2, rc.name, rc.version
            from trove_classifiers t, release_classifiers rc, releases r
            where rc.name=r.name and rc.version=r.version and not r._pypi_hidden and rc.trove_id=t.id
            group by t.l2, rc.name, rc.version) res group by res.l2"""
            cursor.execute(tally)
            return [], cursor.fetchall()

        # First compute statement to produce all packages still selected
        pkgs = "select name, version, summary from releases where _pypi_hidden="+self.false
        for c in selected_classifiers:
            level = t.trove[c].level
            pkgs = """select distinct a.name, a.version, summary from (%s) a, release_classifiers rc, trove_classifiers t
             where a.name=rc.name and a.version=rc.version and rc.trove_id=t.id and
             t.l%d=%d""" % (pkgs, level, c)
        # Next download all selected releases
        cursor.execute(pkgs)
        releases = []
        for name, version, summary in cursor.fetchall():
            if summary: summary = summary.decode('utf-8')
            releases.append((name.decode('utf-8'), version, summary))
        # Finally, compute the tally
        tally = """select tl.id,count(*) from (select distinct t.id, a.name,
        a.version from (%s) a, release_classifiers rc, trove_classifiers t, trove_classifiers t2
        where a.name=rc.name and a.version=rc.version and rc.trove_id=t2.id""" % pkgs
        # tally all level-2 classifiers
        tally += " and (t.id=t2.l2"
        # then tally for all level n+1 classifiers of selected_classifiers
        for c in selected_classifiers:
            level = t.trove[c].level
            if level==5:
                # There are no level 6 classifiers
                continue
            tally += " or (t.id=t2.l%d and t2.l%d=%s)" % (level+1, level, c)
        tally += ")) tl group by tl.id"
        cursor.execute(tally)
        tally = cursor.fetchall()
        return releases, tally

    #
    # File handling
    #
    def gen_file_url(self, pyversion, name, filename, prefix=None):
        '''Generate the URL for a given file download.'''
        if not prefix:
            prefix = self.config.files_url
        return os.path.join(prefix, pyversion,
                                       name[0], name, filename)

    def gen_file_path(self, pyversion, name, filename):
        '''Generate the path to the file on disk.'''
        return os.path.join(self.config.database_files_dir, pyversion,
            name[0], name, filename)

    def add_file(self, name, version, content, md5_digest, filetype,
            pyversion, comment, filename, signature):
        '''Add to the database and store the content to disk.'''
        # add database entry
        cursor = self.get_cursor()
        sql = '''insert into release_files (name, version, python_version,
            packagetype, comment_text, filename, md5_digest, upload_time) values
            (%s, %s, %s, %s, %s, %s, %s, current_timestamp)'''
        safe_execute(cursor, sql, (name, version, pyversion, filetype,
            comment, filename, md5_digest))

        # store file to disk
        filepath = self.gen_file_path(pyversion, name, filename)
        dirpath = os.path.split(filepath)[0]
        if not os.path.exists(dirpath):
            os.makedirs(dirpath)
        f = open(filepath, 'wb')
        try:
            f.write(content)
        finally:
            f.close()

        # Store signature next to the file
        if signature:
            f = open(filepath + ".asc", "wb")
            try:
                f.write(signature)
            finally:
                f.close()

        # add journal entry
        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(
            name, version, "add %s file %s" % (pyversion, filename), date,
            self.username, self.userip)

    _List_Files = FastResultRow('''packagetype python_version comment_text
    filename md5_digest size! has_sig! downloads! upload_time!''')
    def list_files(self, name, version, show_missing=False):
        cursor = self.get_cursor()
        sql = '''select packagetype, python_version, comment_text,
            filename, md5_digest, downloads, upload_time from release_files
            where name=%s and version=%s'''
        safe_execute(cursor, sql, (name, version))
        l = []
        for pt, pv, ct, fn, m5, dn, ut in cursor.fetchall():
            path = self.gen_file_path(pv, name, fn)
            try:
                size = os.stat(path)[stat.ST_SIZE]
            except OSError, error:
                if error.errno != errno.ENOENT: raise

                if show_missing:
                    size = 0
                else:
                    continue
            has_sig = os.path.exists(path+'.asc')
            l.append(self._List_Files(None, (pt, pv, ct, fn, m5, size, has_sig, dn, ut)))
        l.sort(key=lambda r:r['filename'])
        return l

    def has_file(self, name, version, filename):
        cursor = self.get_cursor()
        sql = '''select count(*) from release_files
            where name=%s and version=%s and filename=%s'''
        safe_execute(cursor, sql, (name, version, filename))
        return int(cursor.fetchone()[0])

    def remove_file(self, digest):
        cursor = self.get_cursor()
        sql = '''select python_version, name, version, filename from release_files
            where md5_digest=%s'''
        safe_execute(cursor, sql, (digest, ))
        info = cursor.fetchone()
        if not info:
            raise KeyError, 'no such file'
        pyversion, name, version, filename = info
        safe_execute(cursor, 'delete from release_files where md5_digest=%s',
            (digest, ))
        filepath = self.gen_file_path(pyversion, name, filename)
        dirpath = os.path.split(filepath)[0]
        os.remove(filepath)
        if os.path.exists(filepath+'.asc'):
            os.remove(filepath+'.asc')
        while True:
            if os.listdir(dirpath):
                break
            if dirpath == self.config.database_files_dir:
                break
            os.rmdir(dirpath)
            dirpath = os.path.split(dirpath)[0]
        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(name, version, "remove file %s" % filename,
                                            date, self.username, self.userip)

        self._add_invalidation(name)

    _File_Info = FastResultRow('''python_version packagetype name comment_text
                                filename''')
    def get_file_info(self, digest):
        '''Get the file info based on the md5 hash.

        Raise KeyError if the digest doesn:t match any file in the
        database.
        '''
        cursor = self.get_cursor()
        sql = '''select python_version, packagetype, name, comment_text,
            filename from release_files where md5_digest=%s'''
        safe_execute(cursor, sql, (digest, ))
        row = cursor.fetchone()
        if not row:
            raise KeyError, 'invalid digest %r'%digest
        return self._File_Info(None, row)

    def log_docs(self, name, version):
        cursor = self.get_cursor()
        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        self.add_journal_entry(name, version, "docupdate", date,
                                                self.username, self.userip)

    def docs_url(self, name):
        '''Determine the local (pythonhosted.org) documentation URL, if any.

        Returns the URL or '' if there are no docs.
        '''
        for sub in [[], ['html']]:
            path = [self.config.database_docs_dir,
                name.encode('utf8')] + sub + ['index.html']
            if os.path.exists(os.path.join(*path)):
                return '/'.join([self.config.package_docs_url, name] + sub)
        return ''

    def update_upload_times(self):
        cursor = self.get_cursor()
        safe_execute(cursor,
                     "select python_version, name, filename from release_files "
                     "where upload_time is null")
        for pyversion, name, filename in cursor.fetchall():
            fn = self.gen_file_path(pyversion, name, filename)
            try:
                st = os.stat(fn)
            except OSError:
                continue
            date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime(st.st_mtime))
            safe_execute(cursor, "update release_files "
            "set upload_time=%s where python_version=%s and name=%s "
            "and filename = %s", (date, pyversion, name, filename))

            self._add_invalidation(name)
    #
    # Mirrors managment
    #
    def add_mirror(self, ip, user):
        ''' Add a mirror to the mirrors list
        '''
        cursor = self.get_cursor()
        sql = ('insert into mirrors (ip, user_name)'
               'values (%s, %s)')

        values = (ip, user)
        safe_execute(cursor, sql, values)

    def delete_mirror(self, ip):
        ''' Delete a mirror
        '''
        cursor = self.get_cursor()
        cursor.execute('delete from mirrors where ip=%s', [ip])

    def list_mirrors(self):
        ''' Returns a list of mirrors
        '''
        cursor = self.get_cursor()
        sql = '''select ip from mirrors'''
        safe_execute(cursor, sql)
        return cursor.fetchall()

    def find_user_by_cookie(self, cookie):
        '''Return username of user if cookie is valid, else None.'''
        if not cookie:
            return None
        cursor = self.get_cursor()
        sql = 'select name, last_seen from cookies where cookie=%s'
        safe_execute(cursor, sql, (cookie,))
        users = cursor.fetchall()
        if users:
            # cookie was found
            name, last_seen = users[0]
            if datetime.datetime.now()-datetime.timedelta(0,60) > last_seen:
                # refresh cookie and login time every minute
                sql = 'update cookies set last_seen=current_timestamp where cookie=%s'
                safe_execute(cursor, sql, (cookie,))
                sql ='update users set last_login=current_timestamp where name=%s'
                safe_execute(cursor, sql, (name,))
            return name
        return None

    def create_cookie(self, username):
        '''Create and return a new cookie for the user.'''
        cursor = self.get_cursor()
        cookie = binascii.hexlify(os.urandom(16))
        sql = '''insert into cookies(cookie, name, last_seen)
                 values(%s, %s, current_timestamp)'''
        safe_execute(cursor, sql, (cookie, username))
        return cookie

    def delete_cookie(self, cookie):
        cursor = self.get_cursor()
        safe_execute(cursor, 'delete from cookies where cookie=%s', (cookie,))

    # CSRF Protection

    def get_token(self, username):
        '''Return csrf current token for user.'''
        cursor = self.get_cursor()
        sql = '''select token from csrf_tokens where name=%s
                 and end_date > %s'''
        now = datetime.datetime.now()
        safe_execute(cursor, sql, (username, now))
        token = cursor.fetchall()
        if not token:
            return self.create_token(username)
        return token[0][0]

    def create_token(self, username):
        '''Create and return a new csrf token for user.'''
        alphanum = string.ascii_letters + string.digits
        # dependency on cookie existence
        cursor = self.get_cursor()
        safe_execute(cursor, 'select cookie from cookies where name=%s',
                (username,))
        try:
            cookie = cursor.fetchall()[0][0]
        except IndexError:
            # no cookie, make one
            cookie = ''.join(random.choice(alphanum) for i in range(10))

        # create random data
        rand = [random.choice(alphanum) for i in range(12)]
        rand.append(str(int(time.time())))
        rand.append(cookie)
        random.shuffle(rand)
        rand = hmac.new(''.join(random.choice(alphanum) for i in range(16)),
                ''.join(rand),digestmod=hashlib.sha1).hexdigest()
        rand = b64encode(rand)

        # we may have a current entry which is out of date, delete
        safe_execute(cursor, 'delete from csrf_tokens where name=%s', (username,))
        target = datetime.datetime.now() + datetime.timedelta(minutes=15)
        sql = 'insert into csrf_tokens values(%s, %s, %s)'
        safe_execute(cursor, sql, (username, rand, target))

        return rand

    # OpenID

    def store_discovered(self, url, services, op_endpoint, op_local):
        cursor = self.get_cursor()
        sql = '''delete from openid_discovered where url = %s'''
        safe_execute(cursor, sql, (url,))
        services = binary(cursor, pickle.dumps(services, pickle.HIGHEST_PROTOCOL))
        sql = '''insert into openid_discovered(created, url, services, op_endpoint, op_local)
        values(%s, %s, %s, %s, %s)'''
        now = datetime.datetime.now()
        safe_execute(cursor, sql, (now, url, services, op_endpoint, op_local))

    def discovered(self, url):
        cursor = self.get_cursor()
        sql = '''select services, op_endpoint, op_local from openid_discovered where url=%s'''
        safe_execute(cursor, sql, (url,))
        result = cursor.fetchall()
        if result:
            services, endpoint, local = result[0]
            services = pickle.loads(str(services))
            return services, endpoint, local
        else:
            return None

    def get_provider_session(self, provider):
        cursor = self.get_cursor()
        # discover service URL, possibly from cache
        res = self.discovered(provider[2])
        if not res:
            res = openid2rp.discover(provider[2])
            assert res
            self.store_discovered(provider[2], *res)
        stypes, url, op_local = res
        # Check for existing session
        sql = '''select assoc_handle from openid_sessions
                 where url=%s and expires>current_timestamp'''
        safe_execute(cursor, sql, (url,))
        sessions = cursor.fetchall()
        if sessions:
            assoc_handle = sessions[0][0]
            return stypes, url, assoc_handle

        # start from scratch:
        # associate session
        now = datetime.datetime.now()
        session = openid2rp.associate(stypes, url)
        # store it
        sql = '''insert into openid_sessions
                 (url, assoc_handle, expires, mac_key)
                 values (%s, %s, %s, %s)'''
        safe_execute(cursor, sql, (url,
                                   session['assoc_handle'],
                                   now+datetime.timedelta(0,int(session['expires_in'])),
                                   session['mac_key']))
        return stypes, url, session['assoc_handle']

    def get_session_for_endpoint(self, endpoint, stypes):
        '''Return the assoc_handle for the a claimed ID/endpoint pair;
        create a new session if necessary. Discovery is supposed to be
        done by the caller.'''
        cursor = self.get_cursor()
        # Check for existing session
        sql = '''select assoc_handle from openid_sessions
                 where url=%s and expires>current_timestamp'''
        safe_execute(cursor, sql, (endpoint,))
        sessions = cursor.fetchall()
        if sessions:
            return sessions[0][0]

        # associate new session
        now = datetime.datetime.now()
        session = openid2rp.associate(stypes, endpoint)
        # store it
        sql = '''insert into openid_sessions
                 (url, assoc_handle, expires, mac_key)
                 values (%s, %s, %s, %s)'''
        safe_execute(cursor, sql, (endpoint,
                                   session['assoc_handle'],
                                   now+datetime.timedelta(0,int(session['expires_in'])),
                                   session['mac_key']))
        safe_execute(cursor, 'select %s' % self.last_id('openid_sessions'))
        session_id = cursor.fetchone()[0]
        return session['assoc_handle']

    def find_association(self, assoc_handle):
        cursor = self.get_cursor()
        sql ='select mac_key from openid_sessions where assoc_handle=%s'
        safe_execute(cursor, sql, (assoc_handle,))
        sessions = cursor.fetchall()
        if sessions:
            return {'assoc_handle':assoc_handle, 'mac_key':sessions[0][0]}
        return None

    def duplicate_nonce(self, nonce, checkonly = False):
        '''Return true if we might have seen this nonce before.'''
        stamp = openid2rp.parse_nonce(nonce)
        utc = calendar.timegm(stamp.utctimetuple())
        if utc < time.time()-3600:
            # older than 1h: this is probably a replay
            # the cronjob deletes old nonces after 1day
            return True
        cursor = self.get_cursor()
        safe_execute(cursor, 'select * from openid_nonces where nonce=%s',
                     (nonce,))
        if cursor.fetchone():
            return True
        if not checkonly:
            safe_execute(cursor, '''insert into openid_nonces(created, nonce)
            values(%s,%s)''', (stamp, nonce))
        return False

    def check_nonce(self, nonce):
        return self.duplicate_nonce(nonce, checkonly=True)

    def associate_openid(self, username, openid):
        cursor = self.get_cursor()
        safe_execute(cursor, 'insert into openids(id, name) values(%s,%s)',
                     (openid, username))

    def drop_openid(self, openid):
        cursor = self.get_cursor()
        safe_execute(cursor, 'delete from openids where id=%s', (openid,))

    def set_openid_trustedroot(self, username, trusted_root):
        now = datetime.datetime.now()
        cursor = self.get_cursor()
        safe_execute(cursor, '''select * from openid_whitelist
                     where name=%s and trust_root=%s''',
                     (username, trusted_root))
        if not cursor.fetchone():
            safe_execute(cursor, '''insert into openid_whitelist(
                     name, trust_root, created) values(%s,%s,%s)''',
                     (username, trusted_root, now))

    def check_openid_trustedroot(self, username, trusted_root):
        """Check trusted_root is in user's whitelist"""
        cursor = self.get_cursor()
        safe_execute(cursor, '''select * from openid_whitelist
                                where name=%s and trust_root=%s''',
                                (username, trusted_root))
        if cursor.fetchone():
            return True
        else:
            return False

    def log_keyrotate(self):
        cursor = self.get_cursor()
        date = time.strftime('%Y-%m-%d %H:%M:%S', time.gmtime())
        safe_execute(cursor, '''insert into journals (
              name, version, action, submitted_date, submitted_by,
              submitted_from) values (%s, %s, %s, %s, %s, %s)''',
            ('', '', 'keyrotate ', date,
            None, None))


    #
    # Handle the underlying database
    #
    def get_cursor(self):
        if self._cursor is None:
            self.open()
        return self._cursor

    def open(self):
        ''' Open the database, initialising if necessary.
        '''
        global connection
        # ensure files are group readable and writable
        cd = dict(database=self.config.database_name,
            user=self.config.database_user)
        if self.config.database_pw:
            cd['password'] = self.config.database_pw
        if self.config.database_host:
            cd['host'] = self.config.database_host
        if self.config.database_port:
            cd['port'] = self.config.database_port
        if keep_conn and connection:
            self._conn = connection
            # Rollback any uncommitted earlier change
            try:
                self._conn.rollback()
            except psycopg2.InterfaceError:
                # already closed
                connection = None
                return self.open()
        elif self.config.database_driver == 'sqlite3':
            self._conn = connection = sqlite3.connect(self.config.database_name,
                detect_types=sqlite3.PARSE_DECLTYPES|sqlite3.PARSE_COLNAMES)
            # we store encoded strings in the db, not unicode objects
            self._conn.text_factory = str
        else:
            self._conn = connection = psycopg2.connect(**cd)

        cursor = self._cursor = self._conn.cursor()
        self._changed_packages = set()

    def oid_store(self):
        if self.config.database_driver == 'sqlite3':
            return openid.store.sqlstore.SQLiteStore(self._conn)
        return openid.store.sqlstore.PostgreSQLStore(self._conn)

    def force_close(self):
        '''Force closure of the current persistent connection.
        '''
        global connection
        try:
            connection.close()
        except Exception:
            pass
        connection = None
        self._changed_packages = set()

    def set_user(self, username, userip, update_last_login):
        ''' Set the user who is doing the changes.
        '''
        # now check the user
        if username is not None:
            if self.has_user(username):
                self.username = username
                if update_last_login:
                    safe_execute(self.get_cursor(), '''update users
                        set last_login=current_timestamp
                        where name=%s''', (username,))
        self.userip = userip

    def setpasswd(self, username, password, hashed=False):
        if not hashed:
            password = self.config.passlib.encrypt(password)

        safe_execute(self.get_cursor(), '''update users set password=%s
            where name=%s''', (password, username))

    def _add_invalidation(self, package=None):
        self._changed_packages.add(package)

    def _invalidate_cache(self):
        if self.config.fastly_api_key:
            # Build up a list of tags we want to purge
            tags = []
            for pkg in self._changed_packages:
                if pkg is None:
                    tags += ["simple-index"]
                else:
                    tags += ["pkg~%s" % safe_name(pkg).lower()]

            # We only need to bother to enqueue a task if we have something
            #   to purge
            if tags:
                # Enqueue the purge
                self.enqueue(tasks.purge_fastly_tags,
                            self.config.fastly_api_domain,
                            self.config.fastly_api_key,
                            self.config.fastly_service_id,
                            tags,
                        )

        # Empty our changed packages
        self._changed_packages = set()

    def close(self):
        if self._conn is None:
            return
        if keep_conn:
            # rollback any aborted transaction
            self._conn.rollback()
        else:
            self._conn.close()
        if not keep_trove:
            self._trove = None
        self._conn = None
        self._cursor = None

    def commit(self):
        if self._conn is None:
            return
        self._conn.commit()
        self._invalidate_cache()

    def rollback(self):
        if self._conn is None:
            return
        self._conn.rollback()
        self._cursor = None
        self._changed_packages = set()

    def changed(self):
        '''A journalled change has been made. Notify listeners'''
        self.commit()
        # XXX consider running this in a separate thread
        if self.config.pubsubhubbub:
            try:
                urllib2.urlopen(self.config.pubsubhubbub, "hub.mode=publish&hub.url="+
                                urllib2.quote(self.config.url+'?:action=lasthour'))
            except urllib2.HTTPError, e:
                if e.code == 204:
                    # no content, ok
                    return
                # ignore all other errors
            except Exception:
                pass

def generate_random(length, chars = string.letters + string.digits):
    return ''.join([random.choice(chars) for n in range(length)])

class OAuthDataStore(oauth.OAuthDataStore):
    '''Manages an OAuth data store over the Store.
    '''
    def __init__(self, store):
        self.store = store

    def lookup_consumer(self, key):
        cursor = self.store.get_cursor()
        sql = 'select secret from oauth_consumers where consumer = %s'
        safe_execute(cursor, sql, (key,))
        for row in cursor.fetchall():
            return oauth.OAuthConsumer(key, row[0])

    def lookup_token(self, token_type, token_token):
        '''Look up the given token in either the oauth_request_tokens or
        oauth_access_tokens table.
        '''
        cursor = self.store.get_cursor()
        sql = 'select secret from oauth_%s_tokens where token = %%s'%token_type
        safe_execute(cursor, sql, (token_token,))
        for row in cursor.fetchall():
            return oauth.OAuthToken(token_token, row[0])

    def lookup_nonce(self, oauth_consumer, oauth_token, nonce, timestamp):
        '''Check that the indicated timestamp + consumer + token + nonce haven't been seen
        before.

        Return True if it has. Store the information if it hasn't.
        '''
        cursor = self.store.get_cursor()
        if oauth_token is None:
            sql = '''select * from oauth_nonce where
                timestamp=%s and consumer=%s and token is NULL and nonce=%s'''
            safe_execute(cursor, sql, (timestamp, oauth_consumer.key, nonce))
            token = None
        else:
            sql = '''select * from oauth_nonce where
                timestamp=%s and consumer=%s and token=%s and nonce=%s'''
            safe_execute(cursor, sql, (timestamp, oauth_consumer.key,
                oauth_token.key, nonce))
            token = oauth_token.key
        for row in cursor.fetchall():
            return True
        sql = '''insert into oauth_nonce (timestamp, consumer, token, nonce)
            values (%s, %s, %s, %s)'''
        safe_execute(cursor, sql, (timestamp, oauth_consumer.key, token, nonce))
        return False

    def fetch_request_token(self, oauth_consumer, callback):
        '''When it says "fetch" it really means "create".

        oauth_consumer is an OAuthConsumer instance

        Create a token in the oauth_request_tokens table.
        '''
        # generate some randomish token / secret combo
        token = generate_random(32)
        secret = generate_random(64)

        # generate the token in the db
        sql = '''insert into oauth_request_tokens (token, secret, consumer,
            callback, date_created) values (%s, %s, %s, %s, %s)'''
        now = datetime.datetime.now()
        safe_execute(self.store.get_cursor(), sql, (token, secret,
            oauth_consumer.key, callback, now))
        return oauth.OAuthToken(token, secret)

    def authorize_request_token(self, oauth_token, user):
        '''Fill in the user id in the indicated request token.

        oauth_token is just a string
        user is the username of the account to associate with
        '''
        sql = 'update oauth_request_tokens set user_name=%s where token=%s'
        safe_execute(self.store.get_cursor(), sql, (user, oauth_token))
        # XXX note: no return value. Spec says to return OAuthToken but it's
        # not used and I'd have to do a separate SELECT so I just don't see the
        # point.

    def fetch_access_token(self, oauth_consumer, oauth_token, oauth_verifier):
        '''When it says "fetch" it really means "create".

        oauth_consumer is an OAuthConsumer instance
        oauth_token is an OAuthToken instance representing the request token
        oauth_verifier is ignored

        Create a token in the oauth_access_tokens table.
        '''
        # find user in the oauth_request_tokens table
        cursor = self.store.get_cursor()
        sql = '''select user_name from oauth_request_tokens
            where consumer = %s and token = %s'''
        safe_execute(cursor, sql, (oauth_consumer.key, oauth_token.key))
        for row in cursor.fetchall():
            user = row[0]
            break
        else:
            raise ValueError('request token consumer=%r, token=%r not found'%(
                oauth_consumer.key, oauth_token.token))

        # sanity check - if the user_name is None then the request token has
        # NOT been authorised
        if user is None:
            return None

        # check that there's not already an access token for this consumer / user
        sql = '''select token from oauth_access_tokens
            where consumer = %s and user_name = %s'''
        safe_execute(cursor, sql, (oauth_consumer.key, user))
        for row in cursor.fetchall():
            # return the existing token
            return self.lookup_token('access', row[0])

        # generate some randomish token / secret combo
        token = generate_random(32)
        secret = generate_random(64)

        # generate the token in the db
        sql = '''insert into oauth_access_tokens
            (token, secret, consumer, user_name, date_created, last_modified)
            values (%s, %s, %s, %s, %s, %s)'''
        now = datetime.datetime.now()
        safe_execute(cursor, sql, (token, secret, oauth_consumer.key, user,
            now, now))
        return oauth.OAuthToken(token, secret)

    def _get_user(self, token):
        '''Given an access token, determine the user associated with it.
        '''
        sql = 'select user_name from oauth_access_tokens where token = %s'
        cursor = self.store.get_cursor()
        safe_execute(cursor, sql, (token.key, ))
        for row in cursor.fetchall():
            return row[0]
        raise ValueError('token %r not valid (or too valid)'%token.key)

    def _get_consumer_description(self, request_token=None):
        if request_token is not None:
            sql = '''select description from oauth_consumers, oauth_request_tokens
                where oauth_consumers.consumer = oauth_request_tokens.consumer
                and token = %s'''
            cursor = self.store.get_cursor()
            safe_execute(cursor, sql, (request_token, ))
            for row in cursor.fetchall():
                return row[0]
        raise ValueError('token lookup failed')

def xmlescape(url):
    '''Make sure a URL is valid XML'''
    try:
        ElementTree.fromstring('<x y="%s"/>' % url)
    except ElementTree.ParseError:
        return cgi.escape(url)
    else:
        return url

def get_description_urls(html):
    from htmllib import HTMLParser
    from formatter import NullFormatter
    import urlparse, sgmllib
    try:
        parser = HTMLParser(NullFormatter())
        parser.feed(html)
        parser.close()
    except sgmllib.SGMLParseError:
        return []
    result = []
    for url in parser.anchorlist:
        if urlparse.urlparse(url)[0]:
            result.append(xmlescape(url))
    return result

if __name__ == '__main__':
    import config
    cfg = config.Config(sys.argv[1])
    store = Store(cfg)
    store.open()
    if sys.argv[2] == 'changepw':
        store.setpasswd(sys.argv[3], sys.argv[4])
        store.commit()
    elif sys.argv[2] == 'adduser':
        otk = store.store_user(sys.argv[3], sys.argv[4], sys.argv[5])
        store.delete_otk(otk)
        store.commit()
    elif sys.argv[2] == 'checktrove':
        store.check_trove()
        store.commit()
    elif sys.argv[2] == 'updateurls':
        store.updateurls()
        store.commit()
    elif sys.argv[2] == 'updateurls2':
        store.updateurls2()
        store.commit()
    elif sys.argv[2] == 'update_normalized_text':
        store.update_normalized_text()
        store.commit()
    elif sys.argv[2] == 'update_upload_times':
        store.update_upload_times()
        store.commit()
    elif sys.argv[2] == 'update_html':
        store.update_description_html(sys.argv[3])
        store.commit()
    else:
        print "UNKNOWN COMMAND", sys.argv[2]
    store.close()