#!/usr/bin/env python
+from __future__ import print_function
import os, sys, re
import argparse, time
+import signal, atexit
-import pty
from subprocess import Popen, STDOUT, PIPE
from select import select
-# TODO: do we need to support '\n' too
-sep = "\r\n"
-#sep = "\n"
+# Pseudo-TTY and terminal manipulation
+import pty, array, fcntl, termios
+
+IS_PY_3 = sys.version_info[0] == 3
+
+debug_file = None
+log_file = None
+
+def debug(data):
+ if debug_file:
+ debug_file.write(data)
+ debug_file.flush()
+
+def log(data, end='\n'):
+ if log_file:
+ log_file.write(data + end)
+ log_file.flush()
+ print(data, end=end)
+ sys.stdout.flush()
+
+sep = "\n"
rundir = None
parser = argparse.ArgumentParser(
help="default timeout for each individual test action")
parser.add_argument('--pre-eval', default=None, type=str,
help="Mal code to evaluate prior to running the test")
-parser.add_argument('--redirect', action='store_true',
- help="Run implementation in bash and redirect output to /dev/null")
+parser.add_argument('--no-pty', action='store_true',
+ help="Use direct pipes instead of pseudo-tty")
+parser.add_argument('--log-file', type=str,
+ help="Write messages to the named file in addition the screen")
+parser.add_argument('--debug-file', type=str,
+ help="Write all test interaction the named file")
+parser.add_argument('--hard', action='store_true',
+ help="Turn soft tests (soft, deferrable, optional) into hard failures")
-parser.add_argument('test_file', type=argparse.FileType('r'),
+# Control whether deferrable and optional tests are executed
+parser.add_argument('--deferrable', dest='deferrable', action='store_true',
+ help="Enable deferrable tests that follow a ';>>> deferrable=True'")
+parser.add_argument('--no-deferrable', dest='deferrable', action='store_false',
+ help="Disable deferrable tests that follow a ';>>> deferrable=True'")
+parser.set_defaults(deferrable=True)
+parser.add_argument('--optional', dest='optional', action='store_true',
+ help="Enable optional tests that follow a ';>>> optional=True'")
+parser.add_argument('--no-optional', dest='optional', action='store_false',
+ help="Disable optional tests that follow a ';>>> optional=True'")
+parser.set_defaults(optional=True)
+
+parser.add_argument('test_file', type=str,
help="a test file formatted as with mal test data")
parser.add_argument('mal_cmd', nargs="*",
help="Mal implementation command line. Use '--' to "
"specify a Mal command line with dashed options.")
+parser.add_argument('--crlf', dest='crlf', action='store_true',
+ help="Write \\r\\n instead of \\n to the input")
class Runner():
- def __init__(self, args, redirect=False):
- print "args: %s" % repr(args)
- if redirect:
- print "using redirect"
- self.p = Popen(args, bufsize=0, stdin=PIPE, stdout=PIPE, stderr=STDOUT)
+ def __init__(self, args, no_pty=False, line_break="\n"):
+ #print "args: %s" % repr(args)
+ self.no_pty = no_pty
+
+ # Cleanup child process on exit
+ atexit.register(self.cleanup)
+
+ self.p = None
+ env = os.environ
+ env['TERM'] = 'dumb'
+ env['INPUTRC'] = '/dev/null'
+ env['PERL_RL'] = 'false'
+ if no_pty:
+ self.p = Popen(args, bufsize=0,
+ stdin=PIPE, stdout=PIPE, stderr=STDOUT,
+ preexec_fn=os.setsid,
+ env=env)
self.stdin = self.p.stdin
self.stdout = self.p.stdout
else:
# provide tty to get 'interactive' readline to work
master, slave = pty.openpty()
- self.p = Popen(args, bufsize=0, stdin=slave, stdout=slave, stderr=STDOUT)
+
+ # Set terminal size large so that readline will not send
+ # ANSI/VT escape codes when the lines are long.
+ buf = array.array('h', [100, 200, 0, 0])
+ fcntl.ioctl(master, termios.TIOCSWINSZ, buf, True)
+
+ self.p = Popen(args, bufsize=0,
+ stdin=slave, stdout=slave, stderr=STDOUT,
+ preexec_fn=os.setsid,
+ env=env)
+ # Now close slave so that we will get an exception from
+ # read when the child exits early
+ # http://stackoverflow.com/questions/11165521
+ os.close(slave)
self.stdin = os.fdopen(master, 'r+b', 0)
self.stdout = self.stdin
self.buf = ""
self.last_prompt = ""
+ self.line_break = line_break
+
def read_to_prompt(self, prompts, timeout):
end_time = time.time() + timeout
while time.time() < end_time:
- [outs,_,_] = select([self.stdin], [], [], 1)
- if self.stdin in outs:
- new_data = self.stdin.read(1)
- #print "new_data: '%s'" % new_data
- self.buf += new_data
+ [outs,_,_] = select([self.stdout], [], [], 1)
+ if self.stdout in outs:
+ new_data = self.stdout.read(1)
+ new_data = new_data.decode("utf-8") if IS_PY_3 else new_data
+ #print("new_data: '%s'" % new_data)
+ debug(new_data)
+ # Perform newline cleanup
+ self.buf += new_data.replace("\r", "")
for prompt in prompts:
regexp = re.compile(prompt)
match = regexp.search(self.buf)
if match:
end = match.end()
- buf = self.buf[0:end-len(prompt)]
+ buf = self.buf[0:match.start()]
self.buf = self.buf[end:]
self.last_prompt = prompt
return buf
return None
- def write(self, str):
- self.stdout.write(str)
+ def writeline(self, str):
+ def _to_bytes(s):
+ return bytes(s, "utf-8") if IS_PY_3 else s
+
+ self.stdin.write(_to_bytes(str.replace('\r', '\x16\r') + self.line_break))
def cleanup(self):
+ #print "cleaning up"
if self.p:
- self.p.terminate()
+ try:
+ os.killpg(self.p.pid, signal.SIGTERM)
+ except OSError:
+ pass
self.p = None
+class TestReader:
+ def __init__(self, test_file):
+ self.line_num = 0
+ f = open(test_file, newline='') if IS_PY_3 else open(test_file)
+ self.data = f.read().split('\n')
+ self.soft = False
+ self.deferrable = False
+ self.optional = False
+
+ def next(self):
+ self.msg = None
+ self.form = None
+ self.out = ""
+ self.ret = None
+
+ while self.data:
+ self.line_num += 1
+ line = self.data.pop(0)
+ if re.match(r"^\s*$", line): # blank line
+ continue
+ elif line[0:3] == ";;;": # ignore comment
+ continue
+ elif line[0:2] == ";;": # output comment
+ self.msg = line[3:]
+ return True
+ elif line[0:5] == ";>>> ": # settings/commands
+ settings = {}
+ exec(line[5:], {}, settings)
+ if 'soft' in settings:
+ self.soft = settings['soft']
+ if 'deferrable' in settings and settings['deferrable']:
+ self.deferrable = "\nSkipping deferrable and optional tests"
+ return True
+ if 'optional' in settings and settings['optional']:
+ self.optional = "\nSkipping optional tests"
+ return True
+ continue
+ elif line[0:1] == ";": # unexpected comment
+ raise Exception("Test data error at line %d:\n%s" % (self.line_num, line))
+ self.form = line # the line is a form to send
+
+ # Now find the output and return value
+ while self.data:
+ line = self.data[0]
+ if line[0:3] == ";=>":
+ self.ret = line[3:]
+ self.line_num += 1
+ self.data.pop(0)
+ break
+ elif line[0:2] == ";/":
+ self.out = self.out + line[2:] + sep
+ self.line_num += 1
+ self.data.pop(0)
+ else:
+ self.ret = ""
+ break
+ if self.ret != None: break
+
+ if self.out[-1:] == sep and not self.ret:
+ # If there is no return value, output should not end in
+ # separator
+ self.out = self.out[0:-1]
+ return self.form
args = parser.parse_args(sys.argv[1:])
-test_data = args.test_file.read().split('\n')
+# Workaround argparse issue with two '--' on command line
+if sys.argv.count('--') > 0:
+ args.mal_cmd = sys.argv[sys.argv.index('--')+1:]
if args.rundir: os.chdir(args.rundir)
-r = Runner(args.mal_cmd, redirect=args.redirect)
-
-
-test_idx = 0
-def read_test(data):
- global test_idx
- form, output, ret = None, "", None
- while data:
- test_idx += 1
- line = data.pop(0)
- if re.match(r"^\s*$", line): # blank line
- continue
- elif line[0:3] == ";;;": # ignore comment
- continue
- elif line[0:2] == ";;": # output comment
- print line[3:]
- continue
- elif line[0:2] == ";": # unexpected comment
- print "Test data error at line %d:\n%s" % (test_idx, line)
- return None, None, None, test_idx
- form = line # the line is a form to send
-
- # Now find the output and return value
- while data:
- line = data[0]
- if line[0:3] == ";=>":
- ret = line[3:].replace('\\r', '\r').replace('\\n', '\n')
- test_idx += 1
- data.pop(0)
- break
- elif line[0:2] == "; ":
- output = output + line[2:] + sep
- test_idx += 1
- data.pop(0)
- else:
- ret = "*"
- break
- if ret: break
+if args.log_file: log_file = open(args.log_file, "a")
+if args.debug_file: debug_file = open(args.debug_file, "a")
+
+r = Runner(args.mal_cmd, no_pty=args.no_pty, line_break="\r\n" if args.crlf else "\n")
+t = TestReader(args.test_file)
- return form, output, ret, test_idx
-def assert_prompt(timeout):
+def assert_prompt(runner, prompts, timeout):
# Wait for the initial prompt
- header = r.read_to_prompt(['user> ', 'mal-user> '], timeout=timeout)
+ header = runner.read_to_prompt(prompts, timeout=timeout)
if not header == None:
if header:
- print "Started with:\n%s" % header
+ log("Started with:\n%s" % header)
else:
- print "Did not get 'user> ' or 'mal-user> ' prompt"
- print " Got : %s" % repr(r.buf)
- r.cleanup()
+ log("Did not receive one of following prompt(s): %s" % repr(prompts))
+ log(" Got : %s" % repr(r.buf))
sys.exit(1)
# Wait for the initial prompt
-assert_prompt(args.start_timeout)
+try:
+ assert_prompt(r, ['[^\s()<>]+> '], args.start_timeout)
+except:
+ _, exc, _ = sys.exc_info()
+ log("\nException: %s" % repr(exc))
+ log("Output before exception:\n%s" % r.buf)
+ sys.exit(1)
# Send the pre-eval code if any
if args.pre_eval:
sys.stdout.write("RUNNING pre-eval: %s" % args.pre_eval)
- p.write(args.pre_eval)
- assert_prompt(args.test_timeout)
+ r.writeline(args.pre_eval)
+ assert_prompt(r, ['[^\s()<>]+> '], args.test_timeout)
+test_cnt = 0
+pass_cnt = 0
fail_cnt = 0
+soft_fail_cnt = 0
+failures = []
-while test_data:
- form, out, ret, line_num = read_test(test_data)
- if form == None:
+while t.next():
+ if args.deferrable == False and t.deferrable:
+ log(t.deferrable)
break
- sys.stdout.write("TEST: %s -> [%s,%s]" % (form, repr(out), repr(ret)))
- sys.stdout.flush()
- expected = "%s%s%s%s" % (form, sep, out, ret)
- r.write(form + "\n")
+ if args.optional == False and t.optional:
+ log(t.optional)
+ break
+
+ if t.msg != None:
+ log(t.msg)
+ continue
+
+ if t.form == None: continue
+
+ log("TEST: %s -> [%s,%s]" % (repr(t.form), repr(t.out), t.ret), end='')
+
+ # The repeated form is to get around an occasional OS X issue
+ # where the form is repeated.
+ # https://github.com/kanaka/mal/issues/30
+ expects = [".*%s%s%s" % (sep, t.out, re.escape(t.ret)),
+ ".*%s.*%s%s%s" % (sep, sep, t.out, re.escape(t.ret))]
+
+ r.writeline(t.form)
try:
- res = r.read_to_prompt(['\r\nuser> ', '\nuser> ',
- '\r\nmal-user> ', '\nmal-user> '],
+ test_cnt += 1
+ res = r.read_to_prompt(['\r\n[^\s()<>]+> ', '\n[^\s()<>]+> '],
timeout=args.test_timeout)
#print "%s,%s,%s" % (idx, repr(p.before), repr(p.after))
- if ret == "*" or res == expected:
- print " -> SUCCESS"
+ if (t.ret == "" and t.out == ""):
+ log(" -> SUCCESS (result ignored)")
+ pass_cnt += 1
+ elif (re.search(expects[0], res, re.S) or
+ re.search(expects[1], res, re.S)):
+ log(" -> SUCCESS")
+ pass_cnt += 1
else:
- print " -> FAIL (line %d):" % line_num
- print " Expected : %s" % repr(expected)
- print " Got : %s" % repr(res)
- fail_cnt += 1
+ if t.soft and not args.hard:
+ log(" -> SOFT FAIL (line %d):" % t.line_num)
+ soft_fail_cnt += 1
+ fail_type = "SOFT "
+ else:
+ log(" -> FAIL (line %d):" % t.line_num)
+ fail_cnt += 1
+ fail_type = ""
+ log(" Expected : %s" % repr(expects[0]))
+ log(" Got : %s" % repr(res))
+ failed_test = """%sFAILED TEST (line %d): %s -> [%s,%s]:
+ Expected : %s
+ Got : %s""" % (fail_type, t.line_num, t.form, repr(t.out),
+ t.ret, repr(expects[0]), repr(res))
+ failures.append(failed_test)
except:
- print "Got Exception"
- r.cleanup()
+ _, exc, _ = sys.exc_info()
+ log("\nException: %s" % repr(exc))
+ log("Output before exception:\n%s" % r.buf)
sys.exit(1)
-r.cleanup()
+
+if len(failures) > 0:
+ log("\nFAILURES:")
+ for f in failures:
+ log(f)
+
+results = """
+TEST RESULTS (for %s):
+ %3d: soft failing tests
+ %3d: failing tests
+ %3d: passing tests
+ %3d: total tests
+""" % (args.test_file, soft_fail_cnt, fail_cnt,
+ pass_cnt, test_cnt)
+log(results)
+
+debug("\n") # add some separate to debug log
if fail_cnt > 0:
- print "FAILURES: %d" % fail_cnt
- sys.exit(2)
+ sys.exit(1)
sys.exit(0)