from collections import defaultdict from math import e import os, zipfile from django.conf import settings from django.core.cache import cache from django.db.models import Case, Count, ExpressionWrapper, F, Max, Q, When from django.db.models.fields import FloatField from django.utils import timezone from django.utils.translation import gettext as _, gettext_noop from judge.models import Problem, Submission __all__ = ['contest_completed_ids', 'get_result_data', 'user_completed_ids', 'user_authored_ids', 'user_editable_ids'] def user_authored_ids(profile): result = set(Problem.objects.filter(authors=profile).values_list('id', flat=True)) return result def user_editable_ids(profile): result = set((Problem.objects.filter(authors=profile) | Problem.objects.filter(curators=profile)) .values_list('id', flat=True)) return result def contest_completed_ids(participation): key = 'contest_complete:%d' % participation.id result = cache.get(key) if result is None: result = set(participation.submissions.filter(submission__result='AC', points=F('problem__points')) .values_list('problem__problem__id', flat=True).distinct()) cache.set(key, result, 86400) return result def user_completed_ids(profile): key = 'user_complete:%d' % profile.id result = cache.get(key) if result is None: result = set(Submission.objects.filter(user=profile, result='AC', points=F('problem__points')) .values_list('problem_id', flat=True).distinct()) cache.set(key, result, 86400) return result def contest_attempted_ids(participation): key = 'contest_attempted:%s' % participation.id result = cache.get(key) if result is None: result = {id: {'achieved_points': points, 'max_points': max_points} for id, max_points, points in (participation.submissions .values_list('problem__problem__id', 'problem__points') .annotate(points=Max('points')) .filter(points__lt=F('problem__points')))} cache.set(key, result, 86400) return result def user_attempted_ids(profile): key = 'user_attempted:%s' % profile.id result = cache.get(key) if result is None: result = {id: {'achieved_points': points, 'max_points': max_points} for id, max_points, points in (Submission.objects.filter(user=profile) .values_list('problem__id', 'problem__points') .annotate(points=Max('points')) .filter(points__lt=F('problem__points')))} cache.set(key, result, 86400) return result def _get_result_data(results): return { 'categories': [ # Using gettext_noop here since this will be tacked into the cache, so it must be language neutral. # The caller, SubmissionList.get_result_data will run ugettext on the name. {'code': 'AC', 'name': gettext_noop('Accepted'), 'count': results['AC']}, {'code': 'WA', 'name': gettext_noop('Wrong'), 'count': results['WA']}, {'code': 'CE', 'name': gettext_noop('Compile Error'), 'count': results['CE']}, {'code': 'TLE', 'name': gettext_noop('Timeout'), 'count': results['TLE']}, {'code': 'ERR', 'name': gettext_noop('Error'), 'count': results['MLE'] + results['OLE'] + results['IR'] + results['RTE'] + results['AB'] + results['IE']}, ], 'total': sum(results.values()), } def get_result_data(*args, **kwargs): if args: submissions = args[0] if kwargs: raise ValueError(_("Can't pass both queryset and keyword filters")) else: submissions = Submission.objects.filter(**kwargs) if kwargs is not None else Submission.objects raw = submissions.values('result').annotate(count=Count('result')).values_list('result', 'count') return _get_result_data(defaultdict(int, raw)) def editable_problems(user, profile=None): subquery = Problem.objects.all() if profile is None: profile = user.profile if not user.has_perm('judge.edit_all_problem'): subfilter = Q(authors__id=profile.id) | Q(curators__id=profile.id) if user.has_perm('judge.edit_public_problem'): subfilter |= Q(is_public=True) subquery = subquery.filter(subfilter) return subquery def hot_problems(duration, limit): cache_key = 'hot_problems:%d:%d' % (duration.total_seconds(), limit) qs = cache.get(cache_key) if qs is None: qs = Problem.get_public_problems() \ .filter(submission__date__gt=timezone.now() - duration, points__gt=3, points__lt=25) qs0 = qs.annotate(k=Count('submission__user', distinct=True)).order_by('-k').values_list('k', flat=True) if not qs0: return [] # make this an aggregate mx = float(qs0[0]) qs = qs.annotate(unique_user_count=Count('submission__user', distinct=True)) # fix braindamage in excluding CE qs = qs.annotate(submission_volume=Count(Case( When(submission__result='AC', then=1), When(submission__result='WA', then=1), When(submission__result='IR', then=1), When(submission__result='RTE', then=1), When(submission__result='TLE', then=1), When(submission__result='OLE', then=1), output_field=FloatField(), ))) qs = qs.annotate(ac_volume=Count(Case( When(submission__result='AC', then=1), output_field=FloatField(), ))) qs = qs.filter(unique_user_count__gt=max(mx / 3.0, 1)) qs = qs.annotate(ordering=ExpressionWrapper( 0.5 * F('points') * (0.4 * F('ac_volume') / F('submission_volume') + 0.6 * F('ac_rate')) + 100 * e ** (F('unique_user_count') / mx), output_field=FloatField(), )).order_by('-ordering').defer('description')[:limit] cache.set(cache_key, qs, 900) return qs def get_visible_content(data): data = data or b'' data = data.replace(b'\r\n', b'\r').replace(b'\r', b'\n') data = data.decode('utf-8') if (len(data) > settings.TESTCASE_VISIBLE_LENGTH): data = data[:settings.TESTCASE_VISIBLE_LENGTH] data += '.' * 3 return data def get_cachekey_file(file): return file.replace(' ', '===') def get_problem_case(problem, files): result = {} uncached_files = [] for file in files: cache_key = 'problem_archive:%s:%s' % (problem.code, get_cachekey_file(file)) qs = cache.get(cache_key) if qs is None: uncached_files.append(file) else: result['file'] = qs if not uncached_files: return result archive_path = os.path.join(settings.DMOJ_PROBLEM_DATA_ROOT, str(problem.data_files.zipfile)) if not os.path.exists(archive_path): raise Exception( 'archive file "%s" does not exist' % archive_path) try: archive = zipfile.ZipFile(archive_path, 'r') except zipfile.BadZipfile: raise Exception('bad archive: "%s"' % archive_path) for file in uncached_files: cache_key = 'problem_archive:%s:%s' % (problem.code, get_cachekey_file(file)) with archive.open(file) as f: s = f.read(settings.TESTCASE_VISIBLE_LENGTH + 3) # add this so there are no characters left behind (ex, 'รก' = 2 utf-8 chars) while True: try: s.decode('utf-8') break except UnicodeDecodeError: s += f.read(1) qs = get_visible_content(s) cache.set(cache_key, qs, 86400) result[file] = qs return result