summaryrefslogtreecommitdiff
path: root/yaksh/bash_code_evaluator.py
diff options
context:
space:
mode:
Diffstat (limited to 'yaksh/bash_code_evaluator.py')
-rw-r--r--yaksh/bash_code_evaluator.py176
1 files changed, 134 insertions, 42 deletions
diff --git a/yaksh/bash_code_evaluator.py b/yaksh/bash_code_evaluator.py
index a468fd7..fbfa232 100644
--- a/yaksh/bash_code_evaluator.py
+++ b/yaksh/bash_code_evaluator.py
@@ -3,6 +3,7 @@ import traceback
import pwd
import os
from os.path import join, isfile
+import sys
import subprocess
import importlib
@@ -11,34 +12,32 @@ from code_evaluator import CodeEvaluator
class BashCodeEvaluator(CodeEvaluator):
- """Tests the Bash code obtained from Code Server"""
- def __init__(self, test_case_data, test, language, user_answer,
- ref_code_path=None, in_dir=None):
- super(BashCodeEvaluator, self).__init__(test_case_data, test, language, user_answer,
- ref_code_path, in_dir)
- self.test_case_args = self._setup()
+ # """Tests the Bash code obtained from Code Server"""
+ # def __init__(self, test_case_data, test, language, user_answer,
+ # ref_code_path=None, in_dir=None):
+ # super(BashCodeEvaluator, self).__init__(test_case_data, test, language, user_answer,
+ # ref_code_path, in_dir)
+ # self.test_case_args = self._setup()
# Private Protocol ##########
- def _setup(self):
- super(BashCodeEvaluator, self)._setup()
-
- self.submit_path = self.create_submit_code_file('submit.sh')
- self._set_file_as_executable(self.submit_path)
- get_ref_path, get_test_case_path = self.ref_code_path.strip().split(',')
- get_ref_path = get_ref_path.strip()
- get_test_case_path = get_test_case_path.strip()
- ref_path, test_case_path = self._set_test_code_file_path(get_ref_path,
- get_test_case_path)
-
- return ref_path, self.submit_path, test_case_path
-
- def _teardown(self):
+ def setup(self):
+ super(BashCodeEvaluator, self).setup()
+ self.submit_code_path = self.create_submit_code_file('submit.sh')
+ self._set_file_as_executable(self.submit_code_path)
+ # get_ref_path, get_test_case_path = self.ref_code_path.strip().split(',')
+ # get_ref_path = get_ref_path.strip()
+ # get_test_case_path = get_test_case_path.strip()
+ # ref_path, test_case_path = self._set_test_code_file_path(get_ref_path,
+ # get_test_case_path)
+
+ # return ref_path, self.submit_code_path, test_case_path
+
+ def teardown(self):
# Delete the created file.
- super(BashCodeEvaluator, self)._teardown()
- os.remove(self.submit_path)
+ super(BashCodeEvaluator, self).teardown()
+ os.remove(self.submit_code_path)
- def _check_code(self, ref_path, submit_path,
- test_case_path=None):
+ def check_code(self, user_answer, test_case_data):
""" Function validates student script using instructor script as
reference. Test cases can optionally be provided. The first argument
ref_path, is the path to instructor script, it is assumed to
@@ -61,23 +60,31 @@ class BashCodeEvaluator(CodeEvaluator):
the required permissions are not given to the file(s).
"""
- if not isfile(ref_path):
- return False, "No file at %s or Incorrect path" % ref_path
- if not isfile(submit_path):
- return False, "No file at %s or Incorrect path" % submit_path
- if not os.access(ref_path, os.X_OK):
- return False, "Script %s is not executable" % ref_path
- if not os.access(submit_path, os.X_OK):
- return False, "Script %s is not executable" % submit_path
+ ref_code_path = test_case_data[0]
+ get_ref_path, get_test_case_path = ref_code_path.strip().split(',')
+ get_ref_path = get_ref_path.strip()
+ get_test_case_path = get_test_case_path.strip()
+ clean_ref_code_path, clean_test_case_path = self._set_test_code_file_path(get_ref_path,
+ get_test_case_path)
+
+ if not isfile(clean_ref_code_path):
+ return False, "No file at %s or Incorrect path" % clean_ref_code_path
+ if not isfile(self.submit_code_path):
+ return False, "No file at %s or Incorrect path" % self.submit_code_path
+ if not os.access(clean_ref_code_path, os.X_OK):
+ return False, "Script %s is not executable" % clean_ref_code_path
+ if not os.access(self.submit_code_path, os.X_OK):
+ return False, "Script %s is not executable" % self.submit_code_path
success = False
+ self.write_to_submit_code_file(self.submit_code_path, user_answer)
- if test_case_path is None or "":
- ret = self._run_command(ref_path, stdin=None,
+ if clean_test_case_path is None or "":
+ ret = self._run_command(clean_ref_code_path, stdin=None,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
proc, inst_stdout, inst_stderr = ret
- ret = self._run_command(submit_path, stdin=None,
+ ret = self._run_command(self.submit_code_path, stdin=None,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
proc, stdnt_stdout, stdnt_stderr = ret
@@ -88,26 +95,26 @@ class BashCodeEvaluator(CodeEvaluator):
stdnt_stderr)
return False, err
else:
- if not isfile(test_case_path):
- return False, "No test case at %s" % test_case_path
- if not os.access(ref_path, os.R_OK):
- return False, "Test script %s, not readable" % test_case_path
+ if not isfile(clean_test_case_path):
+ return False, "No test case at %s" % clean_test_case_path
+ if not os.access(clean_ref_code_path, os.R_OK):
+ return False, "Test script %s, not readable" % clean_test_case_path
# valid_answer is True, so that we can stop once a test case fails
valid_answer = True
# loop_count has to be greater than or equal to one.
# Useful for caching things like empty test files,etc.
loop_count = 0
- test_cases = open(test_case_path).readlines()
+ test_cases = open(clean_test_case_path).readlines()
num_lines = len(test_cases)
for test_case in test_cases:
loop_count += 1
if valid_answer:
- args = [ref_path] + [x for x in test_case.split()]
+ args = [clean_ref_code_path] + [x for x in test_case.split()]
ret = self._run_command(args, stdin=None,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
proc, inst_stdout, inst_stderr = ret
- args = [submit_path]+[x for x in test_case.split()]
+ args = [self.submit_code_path]+[x for x in test_case.split()]
ret = self._run_command(args, stdin=None,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
@@ -120,3 +127,88 @@ class BashCodeEvaluator(CodeEvaluator):
stdnt_stdout+stdnt_stderr)
return False, err
+
+
+ # def _check_code(self, ref_path, submit_path,
+ # test_case_path=None):
+ # """ Function validates student script using instructor script as
+ # reference. Test cases can optionally be provided. The first argument
+ # ref_path, is the path to instructor script, it is assumed to
+ # have executable permission. The second argument submit_path, is
+ # the path to the student script, it is assumed to have executable
+ # permission. The Third optional argument is the path to test the
+ # scripts. Each line in this file is a test case and each test case is
+ # passed to the script as standard arguments.
+
+ # Returns
+ # --------
+
+ # returns (True, "Correct answer") : If the student script passes all
+ # test cases/have same output, when compared to the instructor script
+
+ # returns (False, error_msg): If the student script fails a single
+ # test/have dissimilar output, when compared to the instructor script.
+
+ # Returns (False, error_msg): If mandatory arguments are not files or if
+ # the required permissions are not given to the file(s).
+
+ # """
+ # if not isfile(ref_path):
+ # return False, "No file at %s or Incorrect path" % ref_path
+ # if not isfile(submit_path):
+ # return False, "No file at %s or Incorrect path" % submit_path
+ # if not os.access(ref_path, os.X_OK):
+ # return False, "Script %s is not executable" % ref_path
+ # if not os.access(submit_path, os.X_OK):
+ # return False, "Script %s is not executable" % submit_path
+
+ # success = False
+
+ # if test_case_path is None or "":
+ # ret = self._run_command(ref_path, stdin=None,
+ # stdout=subprocess.PIPE,
+ # stderr=subprocess.PIPE)
+ # proc, inst_stdout, inst_stderr = ret
+ # ret = self._run_command(submit_path, stdin=None,
+ # stdout=subprocess.PIPE,
+ # stderr=subprocess.PIPE)
+ # proc, stdnt_stdout, stdnt_stderr = ret
+ # if inst_stdout == stdnt_stdout:
+ # return True, "Correct answer"
+ # else:
+ # err = "Error: expected %s, got %s" % (inst_stderr,
+ # stdnt_stderr)
+ # return False, err
+ # else:
+ # if not isfile(test_case_path):
+ # return False, "No test case at %s" % test_case_path
+ # if not os.access(ref_path, os.R_OK):
+ # return False, "Test script %s, not readable" % test_case_path
+ # # valid_answer is True, so that we can stop once a test case fails
+ # valid_answer = True
+ # # loop_count has to be greater than or equal to one.
+ # # Useful for caching things like empty test files,etc.
+ # loop_count = 0
+ # test_cases = open(test_case_path).readlines()
+ # num_lines = len(test_cases)
+ # for test_case in test_cases:
+ # loop_count += 1
+ # if valid_answer:
+ # args = [ref_path] + [x for x in test_case.split()]
+ # ret = self._run_command(args, stdin=None,
+ # stdout=subprocess.PIPE,
+ # stderr=subprocess.PIPE)
+ # proc, inst_stdout, inst_stderr = ret
+ # args = [submit_path]+[x for x in test_case.split()]
+ # ret = self._run_command(args, stdin=None,
+ # stdout=subprocess.PIPE,
+ # stderr=subprocess.PIPE)
+ # proc, stdnt_stdout, stdnt_stderr = ret
+ # valid_answer = inst_stdout == stdnt_stdout
+ # if valid_answer and (num_lines == loop_count):
+ # return True, "Correct answer"
+ # else:
+ # err = "Error:expected %s, got %s" % (inst_stdout+inst_stderr,
+ # stdnt_stdout+stdnt_stderr)
+ # return False, err
+