da2a11adf9
This reverts commit 0494a36681
.
103 lines
3.7 KiB
Python
Executable file
103 lines
3.7 KiB
Python
Executable file
from collections import namedtuple
|
|
|
|
from django.conf import settings
|
|
from django.db import connection
|
|
|
|
from judge.models import Submission
|
|
from judge.timezone import from_database_time
|
|
|
|
PP_WEIGHT_TABLE = [
|
|
pow(settings.DMOJ_PP_STEP, i) for i in range(settings.DMOJ_PP_ENTRIES)
|
|
]
|
|
|
|
PPBreakdown = namedtuple(
|
|
"PPBreakdown",
|
|
"points weight scaled_points problem_name problem_code "
|
|
"sub_id sub_date sub_points sub_total sub_result_class "
|
|
"sub_short_status sub_long_status sub_lang",
|
|
)
|
|
|
|
|
|
def get_pp_breakdown(user, start=0, end=settings.DMOJ_PP_ENTRIES):
|
|
with connection.cursor() as cursor:
|
|
cursor.execute(
|
|
f"""
|
|
SELECT max_points_table.problem_code,
|
|
max_points_table.problem_name,
|
|
max_points_table.max_points,
|
|
judge_submission.id,
|
|
judge_submission.date,
|
|
judge_submission.case_points,
|
|
judge_submission.case_total,
|
|
judge_submission.result,
|
|
judge_language.short_name,
|
|
judge_language.key
|
|
FROM (
|
|
SELECT judge_problem.id problem_id,
|
|
judge_problem.name problem_name,
|
|
judge_problem.code problem_code,
|
|
MAX(judge_submission.points) AS max_points
|
|
FROM judge_problem
|
|
INNER JOIN judge_submission ON (judge_problem.id = judge_submission.problem_id)
|
|
WHERE (judge_problem.is_public AND
|
|
NOT judge_problem.is_organization_private AND
|
|
judge_submission.points IS NOT NULL AND
|
|
judge_submission.user_id = %s)
|
|
GROUP BY judge_problem.id
|
|
HAVING MAX(judge_submission.points) > 0.0
|
|
) AS max_points_table
|
|
INNER JOIN judge_submission ON (
|
|
judge_submission.problem_id = max_points_table.problem_id AND
|
|
judge_submission.points = max_points_table.max_points AND
|
|
judge_submission.user_id = %s
|
|
)
|
|
INNER JOIN judge_language ON (judge_submission.language_id = judge_language.id)
|
|
GROUP BY max_points_table.problem_id
|
|
ORDER BY max_points DESC, judge_submission.date DESC
|
|
LIMIT %s OFFSET %s
|
|
""",
|
|
(user.id, user.id, end - start + 1, start),
|
|
)
|
|
data = cursor.fetchall()
|
|
|
|
breakdown = []
|
|
for weight, contrib in zip(PP_WEIGHT_TABLE[start:end], data):
|
|
(
|
|
code,
|
|
name,
|
|
points,
|
|
id,
|
|
date,
|
|
case_points,
|
|
case_total,
|
|
result,
|
|
lang_short_name,
|
|
lang_key,
|
|
) = contrib
|
|
|
|
# Replicates a lot of the logic usually done on Submission objects
|
|
lang_short_display_name = lang_short_name or lang_key
|
|
result_class = Submission.result_class_from_code(
|
|
result, case_points, case_total
|
|
)
|
|
long_status = Submission.USER_DISPLAY_CODES.get(result, "")
|
|
|
|
breakdown.append(
|
|
PPBreakdown(
|
|
points=points,
|
|
weight=weight * 100,
|
|
scaled_points=points * weight,
|
|
problem_name=name,
|
|
problem_code=code,
|
|
sub_id=id,
|
|
sub_date=from_database_time(date),
|
|
sub_points=case_points,
|
|
sub_total=case_total,
|
|
sub_short_status=result,
|
|
sub_long_status=long_status,
|
|
sub_result_class=result_class,
|
|
sub_lang=lang_short_display_name,
|
|
)
|
|
)
|
|
has_more = end < min(len(PP_WEIGHT_TABLE), start + len(data))
|
|
return breakdown, has_more
|