path: root/prolog/engine.py

#!/usr/bin/python3

# CodeQ: an online programming tutor.
# Copyright (C) 2015 UL FRI
#
# This program is free software: you can redistribute it and/or modify it under
# the terms of the GNU Affero General Public License as published by the Free
# Software Foundation, either version 3 of the License, or (at your option) any
# later version.
#
# This program is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License for more
# details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.

import collections
import html
import http.client
import json
from operator import itemgetter
import re
import time
import urllib

def strip_html(text):
    return html.unescape(re.sub(r'</?[a-z]+[^>]*>', '', text))

# Create a new pengine and initialize it with [code]. Return engine ID and a
# list of messages from Prolog.
def create(code='', timeout=10):
    opts = {'format': 'json-html', 'destroy': False, 'src_text': code}
    reply, output = request('POST', '/pengine/create', body=json.dumps(opts), timeout=timeout)
    return reply.get('id'), output

def ask(engine, query, timeout=10):
    # Strip trailing whitespace and periods from the query.
    query = query.rstrip().rstrip('.')
    return send(engine, 'ask(({}),[])'.format(query), timeout=timeout)

def next(engine, n=1, timeout=10):
    return send(engine, 'next({})'.format(n), timeout=timeout)

def stop(engine, timeout=10):
    return send(engine, 'stop', timeout=timeout)

def destroy(engine):
    params = urllib.parse.urlencode({'ids': engine})
    try:
        # We don't care about the answer here, so don't wait for it.
        request('GET', '/pengine/destroy_all?' + params, timeout=0.01)
    except:
        pass

def send(engine, event, timeout=10):
    params = urllib.parse.urlencode({
        'id': engine,
        'event': event,
        'format': 'json-html'})
    return request('GET', path='/pengine/send?' + params, timeout=timeout)

# Return the main reply and pull potential output replies.
address, port = '127.0.0.1', 3030  # TODO put this somewhere sane
def request(method, path, body=None, timeout=10):
    headers = {'Content-Type': 'application/json;charset=utf-8'}
    messages = []
    try:
        conn = http.client.HTTPConnection(address, port, timeout=timeout)
        conn.request(method, path, body, headers=headers)
        while True:
            response = conn.getresponse()
            if response.status != http.client.OK:
                raise Exception('server returned {}'.format(response.status))

            reply = json.loads(response.read().decode('utf-8'))
            if isinstance(reply, dict) and reply.get('event') == 'output':
                messages.append(_get_message(reply))

                # Pull the next output. These requests should return instantly
                # as no additional processing needs to be done in the pengine.
                params = urllib.parse.urlencode({
                    'id': reply['id'],
                    'format': 'json-html'})
                conn.request('GET', '/pengine/pull_response?' + params, headers=headers)
            else:
                return reply, messages
    finally:
        conn.close()

# Strip boilerplate from Prolog messages … ugly.
def _get_message(reply):
    data = strip_html(reply['data']).strip()
    # Prepend the message with formatted location.
    # NOTE in the final version we probably want to return the location object
    where = ''
    if reply['message'] in ('error', 'warning'):
        if 'location' in reply:
            loc = reply['location']
            where += 'near line ' + str(loc['line'])
            if 'ch' in loc:
                where += ', character ' + str(loc['ch'])
            where += ': '
    # Strip HTML and pengine IDs from the messages.
    text = strip_html(data)
    text = re.sub(r"pengine://[0-9]*/src:([0-9]*:)* ", '', text)
    text = re.sub(r"'[0-9]{10,}':", '', text)
    return reply['message'], where + text

# Return the value of variable [name] in the JSON object returned by Prolog.
def get_var(data, name):
    for binding in data['variables']:
        if name in binding['variables']:
            return strip_html(binding['value'])
    return None

# Return a string describing variable bindings and residuals in the JSON object
# returned by Prolog.
def pretty_vars(data):
    result = []
    for binding in data['variables']:
        var_list = binding['variables']
        value = binding['value']
        result.append(' = '.join(var_list) + ' = ' + strip_html(value))
    if 'residuals' in data:
        result += [strip_html(b) for b in data['residuals']]
    return ',\n'.join(result) if result else 'true'

# Run [query] in pengine with ID [engine] and check whether all solutions
# specified by [answers] (a list of binding dictionaries) are returned. This
# function succeeds if [query] finds each solution in [answers] at least once
# within [timeout] seconds, and fails when it finds any other solution.
def check_answers(engine, query, answers, timeout=10):
    seen = []
    # Return false (only) if there is no expected answer where values of all
    # specified variables match values in [answer].
    def check_answer(answer):
        if not answer:
            return True
        bindings, constraints = answer
        for expected in answers:
            if all(bindings.get(var) == val for var, val in expected.items()):
                if expected not in seen:
                    seen.append(expected)
                return True
        return False

    start = time.monotonic()
    try:
        # Limit inferences for each solution to curb unbounded recursion.
        limited = '''(call_with_inference_limit(({}), 100000, _ILR),
                      _ILR \= inference_limit_exceeded)'''.format(query)
        # Run the query.
        reply, output = ask(engine, limited, timeout)
        answer, error, more = process_answer(reply)
        if not check_answer(answer):
            return False

        # Continue while there are more potential answers and time remaining.
        while more:
            real_timeout = timeout - (time.monotonic()-start)
            if real_timeout <= 0:
                break
            reply, output = next(engine, timeout=real_timeout)
            answer, error, more = process_answer(reply)
            if not check_answer(answer):
                return False
    except:
        pass
    return len(seen) == len(answers)

# Run [query] in the pengine with id [engine] and return the first answer only
# found within [timeout] seconds.
# used for quick hint debugging by Sasha
def ask_one(engine, query, timeout=1):
    # Returns either an error message, true, false, timeout (see below), or bindings
    # Timeout is "returned" as an unhandled exception -- deliberately so

    # Run the query.
    reply, output = ask(engine, query, timeout)
    if 'error' in map(itemgetter(0), output):
        return 'query syntax error'
        # TODO: Tim, this doesn't really work. It just proceeds and returns as a runtime error.
    answer, error, more = process_answer(reply)
    if error:
        # query caused a runtime error
        return 'runtime error: ' + error[-1]
    elif answer:
        # no error, query satisfied
        if not answer[0]:       # empty bindings
            return 'true'
        else:
            return answer[0]    # for now return just bindings without constraints
    else:
        # no error, query failed
        return 'false'

def ask_truth(engine, query, timeout=1):
    # Returns either True or False
    # (Runtime) error is False!
    # Timeout is an unhandled exception -- deliberately so

    # Run the query.
    reply, output = ask(engine, query, timeout)
    if 'error' in map(itemgetter(0), output):
        return False
        # TODO: Tim, this doesn't really work. It just proceeds and returns as a runtime error.
    answer, error, more = process_answer(reply)
    if error:
        # query caused a runtime error
        return False
    elif answer:
        return True
    else:
        # no error, query failed
        return False

# reformats prolog's answer into a neater form
def process_answer(reply):
        # When an engine is destroyed, a nested data object has the actual
        # query result.
        if reply['event'] == 'destroy':
            reply = reply['data']

        if reply['event'] == 'success':
            # Return a dictionary {var: value} and a list of constraints (as
            # strings) from the JSON object returned by Prolog.
            data = reply['data'][0]
            bindings = {}
            for binding in data['variables']:
                value = strip_html(binding['value'])
                for name in binding['variables']:
                    bindings[name] = value
            constraints = [strip_html(r) for r in data.get('residuals', [])]
            return (bindings, constraints), None, reply['more']
        elif reply['event'] == 'failure':
            return None, None, False
        elif reply['event'] == 'error':
            # Remove potential module name (engine ID) from the error message.
            error = ('error', reply['data'].replace("'{}':".format(reply['id']), ''))
            return None, error, False

# Basic sanity check.
if __name__ == '__main__':
    engine, messages = create('b(Y). a(X) :- {X > 3}, (X = 5 ; {X > 4}).')
    print(engine, messages)
    if engine is not None:
        reply = ask(engine, 'a(X)', timeout=1.0)
        print(reply)
        destroy(engine)