summaryrefslogtreecommitdiffstats
path: root/scripts/external_libs/scapy-2.3.1/python3/scapy/tools/UTscapy.py
diff options
context:
space:
mode:
Diffstat (limited to 'scripts/external_libs/scapy-2.3.1/python3/scapy/tools/UTscapy.py')
-rw-r--r--scripts/external_libs/scapy-2.3.1/python3/scapy/tools/UTscapy.py677
1 files changed, 677 insertions, 0 deletions
diff --git a/scripts/external_libs/scapy-2.3.1/python3/scapy/tools/UTscapy.py b/scripts/external_libs/scapy-2.3.1/python3/scapy/tools/UTscapy.py
new file mode 100644
index 00000000..212aa123
--- /dev/null
+++ b/scripts/external_libs/scapy-2.3.1/python3/scapy/tools/UTscapy.py
@@ -0,0 +1,677 @@
+## This file is part of Scapy
+## See http://www.secdev.org/projects/scapy for more informations
+## Copyright (C) Philippe Biondi <phil@secdev.org>
+## This program is published under a GPLv2 license
+
+"""
+Unit testing infrastructure for Scapy
+"""
+
+import sys,getopt,imp
+import bz2, base64, os.path, time, traceback, zlib, hashlib
+
+
+#### Import tool ####
+
+def import_module(name):
+ name = os.path.realpath(name)
+ thepath = os.path.dirname(name)
+ name = os.path.basename(name)
+ if name.endswith(".py"):
+ name = name[:-3]
+ f,path,desc = imp.find_module(name,[thepath])
+
+ try:
+ return imp.load_module(name, f, path, desc)
+ finally:
+ if f:
+ f.close()
+
+
+#### INTERNAL/EXTERNAL FILE EMBEDDING ####
+
+class File:
+ def __init__(self, name, URL, local):
+ self.name = name
+ self.local = local
+ self.URL = URL
+ def get_local(self):
+ return bz2.decompress(base64.decodestring(self.local))
+ def get_URL(self):
+ return URL
+ def write(self, dir):
+ if dir:
+ dir += "/"
+ open(dir+self.name,"w").write(self.get_local())
+
+
+# Embed a base64 encoded bziped version of js and css files
+# to work if you can't reach Internet.
+class External_Files:
+ UTscapy_js = File("UTscapy.js", "http://www.secdev.org/projects/UTscapy/UTscapy.js",
+"""QlpoOTFBWSZTWWVijKQAAXxfgERUYOvAChIhBAC/79+qQAH8AFA0poANAMjQAAAG
+ABo0NGEZNBo00BhgAaNDRhGTQaNNAYFURJinplGaKbRkJiekzSenqmpA0Gm1LFMp
+RUklVQlK9WUTZYpNFI1IiEWEFT09Sfj5uO+qO6S5DQwKIxM92+Zku94wL6V/1KTK
+an2c66Ug6SmVKy1ZIrgauxMVLF5xLH0lJRQuKlqLF10iatlTzqvw7S9eS3+h4lu3
+GZyMgoOude3NJ1pQy8eo+X96IYZw+ynehsiPj73m0rnvQ3QXZ9BJQiZQYQ5/uNcl
+2WOlC5vyQqV/BWsnr2NZYLYXQLDs/Bffk4ZfR4/SH6GfA5Xlek4xHNHqbSsRbREO
+gueXo3kcYi94K6hSO3ldD2O/qJXOFqJ8o3TE2aQahxtQpCVUKQMvODHwu2YkaORY
+ZC6gihEallcHDIAtRPScBACAJnUggYhLDX6DEko7nC9GvAw5OcEkiyDUbLdiGCzD
+aXWMC2DuQ2Y6sGf6NcRuON7QSbhHsPc4KKmZ/xdyRThQkGVijKQ=""")
+ UTscapy_css = File("UTscapy.css","http://www.secdev.org/projects/UTscapy/UTscapy.css",
+"""QlpoOTFBWSZTWTbBCNEAAE7fgHxwSB//+Cpj2QC//9/6UAR+63dxbNzO3ccmtGEk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""")
+ def get_local_dict(cls):
+ #return dict(map(lambda (x,y): (x, y.name), filter(lambda (x,y): isinstance(y, File), cls.__dict__.items())))
+ return dict(map(lambda a: (a[0], a[1].name), filter(lambda a: isinstance(a[1], File), cls.__dict__.items())))
+ get_local_dict = classmethod(get_local_dict)
+ def get_URL_dict(cls):
+ #return dict(map(lambda (x,y): (x, y.URL), filter(lambda (x,y): isinstance(y, File), cls.__dict__.items())))
+ return dict(map(lambda a: (a[0], a[1].URL), filter(lambda a: isinstance(a[1], File), cls.__dict__.items())))
+ get_URL_dict = classmethod(get_URL_dict)
+
+
+#### HELPER CLASSES FOR PARAMETRING OUTPUT FORMAT ####
+
+class EnumClass:
+ def from_string(cls,x):
+ return cls.__dict__[x.upper()]
+ from_string = classmethod(from_string)
+
+class Format(EnumClass):
+ TEXT = 1
+ ANSI = 2
+ HTML = 3
+ LATEX = 4
+ XUNIT = 5
+
+
+#### TEST CLASSES ####
+
+class TestClass:
+ def __getitem__(self, item):
+ return getattr(self, item)
+ def add_keywords(self, kw):
+ if kw is str:
+ self.keywords.append(kw)
+ else:
+ self.keywords += kw
+
+class TestCampaign(TestClass):
+ def __init__(self, title):
+ self.title = title
+ self.filename = None
+ self.headcomments = ""
+ self.campaign = []
+ self.keywords = []
+ self.crc = None
+ self.sha = None
+ self.preexec = None
+ self.preexec_output = None
+ def add_testset(self, testset):
+ self.campaign.append(testset)
+ def __iter__(self):
+ return self.campaign.__iter__()
+ def all_tests(self):
+ for ts in self:
+ for t in ts:
+ yield t
+
+class TestSet(TestClass):
+ def __init__(self, name):
+ self.name = name
+ self.set = []
+ self.comments = ""
+ self.keywords = []
+ self.crc = None
+ self.expand = 1
+ def add_test(self, test):
+ self.set.append(test)
+ def __iter__(self):
+ return self.set.__iter__()
+
+class UnitTest(TestClass):
+ def __init__(self, name):
+ self.name = name
+ self.test = ""
+ self.comments = ""
+ self.result = ""
+ self.res = True # must be True at init to have a different truth value than None
+ self.output = ""
+ self.num = -1
+ self.keywords = []
+ self.crc = None
+ self.expand = 1
+ def __nonzero__(self):
+ return self.res
+
+
+#### PARSE CAMPAIGN ####
+
+def parse_campaign_file(campaign_file):
+ test_campaign = TestCampaign("Test campaign")
+ test_campaign.filename= campaign_file.name
+ testset = None
+ test = None
+ testnb = 0
+
+ for l in campaign_file.readlines():
+ if l[0] == '#':
+ continue
+ if l[0] == "~":
+ (test or testset or campaign_file).add_keywords(l[1:].split())
+ elif l[0] == "%":
+ test_campaign.title = l[1:].strip()
+ elif l[0] == "+":
+ testset = TestSet(l[1:].strip())
+ test_campaign.add_testset(testset)
+ test = None
+ elif l[0] == "=":
+ test = UnitTest(l[1:].strip())
+ test.num = testnb
+ testnb += 1
+ testset.add_test(test)
+ elif l[0] == "*":
+ if test is not None:
+
+ test.comments += l[1:]
+ elif testset is not None:
+ testset.comments += l[1:]
+ else:
+ test_campaign.headcomments += l[1:]
+ else:
+ if test is None:
+ if l.strip():
+ print("Unknown content [%s]" % l.strip(), file = sys.stderr)
+ else:
+ test.test += l
+ return test_campaign
+
+def dump_campaign(test_campaign):
+ print("#"*(len(test_campaign.title)+6))
+ print("## %(title)s ##" % test_campaign)
+ print("#"*(len(test_campaign.title)+6))
+ if test_campaign.sha and test_campaign.crc:
+ print("CRC=[%(crc)s] SHA=[%(sha)s]" % test_campaign)
+ print("from file %(filename)s" % test_campaign)
+ print()
+ for ts in test_campaign:
+ if ts.crc:
+ print("+--[%s]%s(%s)--" % (ts.name,"-"*max(2,80-len(ts.name)-18),ts.crc))
+ else:
+ print("+--[%s]%s" % (ts.name,"-"*max(2,80-len(ts.name)-6)))
+ if ts.keywords:
+ print(" kw=%s" % ",".join(ts.keywords))
+ for t in ts:
+ print("%(num)03i %(name)s" % t)
+ c = k = ""
+ if t.keywords:
+ k = "kw=%s" % ",".join(t.keywords)
+ if t.crc:
+ c = "[%(crc)s] " % t
+ if c or k:
+ print(" %s%s" % (c,k) )
+
+#### COMPUTE CAMPAIGN DIGESTS ####
+
+def crc32(x):
+ return "%08X" % (0xffffffff & zlib.crc32(x))
+
+def sha1(x):
+ return hashlib.sha1(x).hexdigest().upper()
+
+def compute_campaign_digests(test_campaign):
+ dc = b""
+ for ts in test_campaign:
+ dts = b""
+ for t in ts:
+ dt = t.test.strip().encode('ascii')
+ t.crc = crc32(dt)
+ dts += b"\0"+dt
+ ts.crc = crc32(dts)
+ dc += b"\0\x01"+dts
+ test_campaign.crc = crc32(dc)
+ if type(test_campaign.filename) is str and test_campaign.filename != '<stdin>':
+ test = open(test_campaign.filename, 'rb').read()
+ elif test_campaign.filename == '<stdin>':
+ test = sys.stdin.read().encode('ascii')
+ else:
+ raise Exception("Unknown test source %s" % test_campaign.filename)
+ test_campaign.sha = sha1(test)
+
+
+#### FILTER CAMPAIGN #####
+
+def filter_tests_on_numbers(test_campaign, num):
+ if num:
+ for ts in test_campaign:
+ #ts.set = filter(lambda t: t.num in num, ts.set)
+ ts.set = [ t for t in ts.set if t.num in num ]
+ #test_campaign.campaign = filter(lambda ts: len(ts.set) > 0, test_campaign.campaign)
+ test_campaign.campaign = [ ts for ts in test_campaign.campaign if len(ts.set) > 0 ]
+
+def filter_tests_keep_on_keywords(test_campaign, kw):
+ def kw_match(lst, kw):
+ for k in lst:
+ if k in kw:
+ return True
+ return False
+
+ if kw:
+ for ts in test_campaign:
+ #ts.set = filter(lambda t: kw_match(t.keywords, kw), ts.set)
+ ts.set = [ t for t in ts.set if kw_match(t.keywords, kw) ]
+
+def filter_tests_remove_on_keywords(test_campaign, kw):
+ def kw_match(lst, kw):
+ for k in kw:
+ if k not in lst:
+ return False
+ return True
+
+ if kw:
+ for ts in test_campaign:
+ #ts.set = filter(lambda t: not kw_match(t.keywords, kw), ts.set)
+ ts.set = [ t for t in ts.set if not kw_match(t.keywords, kw) ]
+
+
+def remove_empty_testsets(test_campaign):
+ #test_campaign.campaign = filter(lambda ts: len(ts.set) > 0, test_campaign.campaign)
+ test_campaign.campaign = [ ts for ts in test_campaign.campaign if len(ts.set) > 0 ]
+
+
+#### RUN CAMPAIGN #####
+
+def run_campaign(test_campaign, get_interactive_session, verb=2):
+ passed=failed=0
+ if test_campaign.preexec:
+ test_campaign.preexec_output = get_interactive_session(test_campaign.preexec.strip())[0]
+ for testset in test_campaign:
+ for t in testset:
+ t.output,res = get_interactive_session(t.test.strip())
+ the_res = False
+ try:
+ if res is None or res:
+ the_res= True
+ except Exception as msg:
+ t.output+="UTscapy: Error during result interpretation:\n"
+ t.output+="".join(traceback.format_exception(sys.exc_type, sys.exc_value, sys.exc_traceback,))
+ if the_res:
+ t.res = True
+ res = "passed"
+ passed += 1
+ else:
+ t.res = False
+ res = "failed"
+ failed += 1
+ t.result = res
+ if verb > 1:
+ print("%(result)6s %(crc)s %(name)s" % t, file = sys.stderr)
+ test_campaign.passed = passed
+ test_campaign.failed = failed
+ if verb:
+ print("Campaign CRC=%(crc)s SHA=%(sha)s" % test_campaign, file = sys.stderr)
+ print("PASSED=%i FAILED=%i" % (passed, failed), file = sys.stderr)
+
+
+#### INFO LINES ####
+
+def info_line(test_campaign):
+ filename = test_campaign.filename
+ if filename is None:
+ return "Run %s by UTscapy" % time.ctime()
+ else:
+ return "Run %s from [%s] by UTscapy" % (time.ctime(), filename)
+
+def html_info_line(test_campaign):
+ filename = test_campaign.filename
+ if filename is None:
+ return """Run %s by <a href="http://www.secdev.org/projects/UTscapy/">UTscapy</a><br>""" % time.ctime()
+ else:
+ return """Run %s from [%s] by <a href="http://www.secdev.org/projects/UTscapy/">UTscapy</a><br>""" % (time.ctime(), filename)
+
+
+#### CAMPAIGN TO something ####
+
+def campaign_to_TEXT(test_campaign):
+ output="%(title)s\n" % test_campaign
+ output += "-- "+info_line(test_campaign)+"\n\n"
+ output += "Passed=%(passed)i\nFailed=%(failed)i\n\n%(headcomments)s\n" % test_campaign
+
+ for testset in test_campaign:
+ output += "######\n## %(name)s\n######\n%(comments)s\n\n" % testset
+ for t in testset:
+ if t.expand:
+ output += "###(%(num)03i)=[%(result)s] %(name)s\n%(comments)s\n%(output)s\n\n" % t
+
+ return output
+
+def campaign_to_ANSI(test_campaign):
+ output="%(title)s\n" % test_campaign
+ output += "-- "+info_line(test_campaign)+"\n\n"
+ output += "Passed=%(passed)i\nFailed=%(failed)i\n\n%(headcomments)s\n" % test_campaign
+
+ for testset in test_campaign:
+ output += "######\n## %(name)s\n######\n%(comments)s\n\n" % testset
+ for t in testset:
+ if t.expand:
+ output += "###(%(num)03i)=[%(result)s] %(name)s\n%(comments)s\n%(output)s\n\n" % t
+
+ return output
+
+def campaign_to_xUNIT(test_campaign):
+ output='<?xml version="1.0" encoding="UTF-8" ?>\n<testsuite>\n'
+ for testset in test_campaign:
+ for t in testset:
+ output += ' <testcase classname="%s"\n' % testset.name.encode("string_escape").replace('"',' ')
+ output += ' name="%s"\n' % t.name.encode("string_escape").replace('"',' ')
+ output += ' duration="0">\n' % t
+ if not t.res:
+ output += '<error><![CDATA[%(output)s]]></error>\n' % t
+ output += "</testcase>\n"
+ output += '</testsuite>'
+ return output
+
+
+def campaign_to_HTML(test_campaign, local=0):
+ output = """<html>
+<head>
+<title>%(title)s</title>
+<link rel="stylesheet" href="%%(UTscapy_css)s" type="text/css">
+<script language="JavaScript" src="%%(UTscapy_js)s" type="text/javascript"></script>
+</head>
+<body>
+
+<h1>%(title)s</h1>
+
+<span class=button onClick="hide_all('tst')">Shrink All</span>
+<span class=button onClick="show_all('tst')">Expand All</span>
+<span class=button onClick="show_passed('tst')">Expand Passed</span>
+<span class=button onClick="show_failed('tst')">Expand Failed</span>
+<p>
+""" % test_campaign
+
+ if local:
+ External_Files.UTscapy_js.write(os.path.dirname(test_campaign.output_file.name))
+ External_Files.UTscapy_css.write(os.path.dirname(test_campaign.output_file.name))
+ output %= External_Files.get_local_dict()
+ else:
+ output %= External_Files.get_URL_dict()
+
+ if test_campaign.crc is not None and test_campaign.sha is not None:
+ output += "CRC=<span class=crc>%(crc)s</span> SHA=<span class=crc>%(sha)s</span><br>" % test_campaign
+ output += "<small><em>"+html_info_line(test_campaign)+"</em></small>"
+ output += test_campaign.headcomments + "\n<p>PASSED=%(passed)i FAILED=%(failed)i<p>\n\n" % test_campaign
+ for ts in test_campaign:
+ for t in ts:
+ output += """<span class=button%(result)s onClick="goto_id('tst%(num)il')">%(num)03i</span>\n""" % t
+ output += "\n\n"
+
+ for testset in test_campaign:
+ output += "<h2>" % testset
+ if testset.crc is not None:
+ output += "<span class=crc>%(crc)s</span> " % testset
+ output += "%(name)s</h2>\n%(comments)s\n<ul>\n" % testset
+ for t in testset:
+ output += """<li class=%(result)s id="tst%(num)il">\n""" % t
+ if t.expand == 2:
+ output +="""
+<span id="tst%(num)i+" class="button%(result)s" onClick="show('tst%(num)i')" style="POSITION: absolute; VISIBILITY: hidden;">+%(num)03i+</span>
+<span id="tst%(num)i-" class="button%(result)s" onClick="hide('tst%(num)i')">-%(num)03i-</span>
+""" % t
+ else:
+ output += """
+<span id="tst%(num)i+" class="button%(result)s" onClick="show('tst%(num)i')">+%(num)03i+</span>
+<span id="tst%(num)i-" class="button%(result)s" onClick="hide('tst%(num)i')" style="POSITION: absolute; VISIBILITY: hidden;">-%(num)03i-</span>
+""" % t
+ if t.crc is not None:
+ output += "<span class=crc>%(crc)s</span>\n" % t
+ output += """%(name)s\n<span class="comment %(result)s" id="tst%(num)i" """ % t
+ if t.expand < 2:
+ output += """ style="POSITION: absolute; VISIBILITY: hidden;" """
+ output += """><br>%(comments)s
+<pre>
+%(output)s</pre></span>
+""" % t
+ output += "\n</ul>\n\n"
+
+ output += "</body></html>"
+ return output
+
+def campaign_to_LATEX(test_campaign):
+ output = r"""\documentclass{report}
+\usepackage{alltt}
+\usepackage{xcolor}
+\usepackage{a4wide}
+\usepackage{hyperref}
+
+\title{%(title)s}
+\date{%%s}
+
+\begin{document}
+\maketitle
+\tableofcontents
+
+\begin{description}
+\item[Passed:] %(passed)i
+\item[Failed:] %(failed)i
+\end{description}
+
+%(headcomments)s
+
+""" % test_campaign
+ output %= info_line(test_campaign)
+
+ for testset in test_campaign:
+ output += "\\chapter{%(name)s}\n\n%(comments)s\n\n" % testset
+ for t in testset:
+ if t.expand:
+ output += r"""\section{%(name)s}
+
+[%(num)03i] [%(result)s]
+
+%(comments)s
+\begin{alltt}
+%(output)s
+\end{alltt}
+
+""" % t
+
+ output += "\\end{document}\n"
+ return output
+
+
+
+#### USAGE ####
+
+def usage():
+ print("""Usage: UTscapy [-m module] [-f {text|ansi|HTML|LaTeX}] [-o output_file]
+ [-t testfile] [-k keywords [-k ...]] [-K keywords [-K ...]]
+ [-l] [-d|-D] [-F] [-q[q]] [-P preexecute_python_code]
+ [-s /path/to/scpay]
+-l\t\t: generate local files
+-F\t\t: expand only failed tests
+-d\t\t: dump campaign
+-D\t\t: dump campaign and stop
+-C\t\t: don't calculate CRC and SHA
+-s\t\t: path to scapy.py
+-q\t\t: quiet mode
+-qq\t\t: [silent mode]
+-n <testnum>\t: only tests whose numbers are given (eg. 1,3-7,12)
+-m <module>\t: additional module to put in the namespace
+-k <kw1>,<kw2>,...\t: include only tests with one of those keywords (can be used many times)
+-K <kw1>,<kw2>,...\t: remove tests with one of those keywords (can be used many times)
+-P <preexecute_python_code>
+""", file = sys.stderr)
+ raise SystemExit
+
+
+#### MAIN ####
+
+def main(argv):
+ import builtins
+
+ # Parse arguments
+
+ FORMAT = Format.ANSI
+ TESTFILE = sys.stdin
+ OUTPUTFILE = sys.stdout
+ LOCAL = 0
+ NUM=None
+ KW_OK = []
+ KW_KO = []
+ DUMP = 0
+ CRC = 1
+ ONLYFAILED = 0
+ VERB=2
+ PREEXEC=""
+ SCAPY="scapy"
+ MODULES = []
+ try:
+ opts = getopt.getopt(argv, "o:t:f:hln:m:k:K:DdCFqP:s:")
+ for opt,optarg in opts[0]:
+ if opt == "-h":
+ usage()
+ elif opt == "-F":
+ ONLYFAILED = 1
+ elif opt == "-q":
+ VERB -= 1
+ elif opt == "-D":
+ DUMP = 2
+ elif opt == "-d":
+ DUMP = 1
+ elif opt == "-C":
+ CRC = 0
+ elif opt == "-s":
+ SCAPY = optarg
+ elif opt == "-P":
+ PREEXEC += "\n"+optarg
+ elif opt == "-f":
+ try:
+ FORMAT = Format.from_string(optarg)
+ except KeyError as msg:
+ raise getopt.GetoptError("Unknown output format %s" % msg)
+ elif opt == "-t":
+ TESTFILE = open(optarg)
+ elif opt == "-o":
+ OUTPUTFILE = open(optarg, "w")
+ elif opt == "-l":
+ LOCAL = 1
+ elif opt == "-n":
+ NUM = []
+ for v in map( lambda x: x.strip(), optarg.split(",") ):
+ try:
+ NUM.append(int(v))
+ except ValueError:
+ v1,v2 = map(int, v.split("-"))
+ for vv in range(v1,v2+1):
+ NUM.append(vv)
+ elif opt == "-m":
+ MODULES.append(optarg)
+ elif opt == "-k":
+ KW_OK.append(optarg.split(","))
+ elif opt == "-K":
+ KW_KO.append(optarg.split(","))
+
+
+ try:
+ from scapy import all as scapy
+ except ImportError as e:
+ raise getopt.GetoptError("cannot import [%s]: %s" % (SCAPY,e))
+
+ for m in MODULES:
+ try:
+ mod = import_module(m)
+ builtins.__dict__.update(mod.__dict__)
+ except ImportError as e:
+ raise getopt.GetoptError("cannot import [%s]: %s" % (m,e))
+
+ except getopt.GetoptError as msg:
+ print("ERROR:",msg, file = sys.stderr)
+ raise SystemExit
+
+ autorun_func = {
+ Format.TEXT: scapy.autorun_get_text_interactive_session,
+ Format.ANSI: scapy.autorun_get_ansi_interactive_session,
+ Format.HTML: scapy.autorun_get_html_interactive_session,
+ Format.LATEX: scapy.autorun_get_latex_interactive_session,
+ Format.XUNIT: scapy.autorun_get_text_interactive_session,
+ }
+
+ # Parse test file
+ test_campaign = parse_campaign_file(TESTFILE)
+
+ # Report parameters
+ if PREEXEC:
+ test_campaign.preexec = PREEXEC
+
+
+ # Compute campaign CRC and SHA
+ if CRC:
+ compute_campaign_digests(test_campaign)
+
+ # Filter out unwanted tests
+ filter_tests_on_numbers(test_campaign, NUM)
+ for k in KW_OK:
+ filter_tests_keep_on_keywords(test_campaign, k)
+ for k in KW_KO:
+ filter_tests_remove_on_keywords(test_campaign, k)
+
+ remove_empty_testsets(test_campaign)
+
+
+ # Dump campaign
+ if DUMP:
+ dump_campaign(test_campaign)
+ if DUMP > 1:
+ sys.exit()
+
+ # Run tests
+ test_campaign.output_file = OUTPUTFILE
+ run_campaign(test_campaign, autorun_func[FORMAT], verb=VERB)
+
+ # Shrink passed
+ if ONLYFAILED:
+ for t in test_campaign.all_tests():
+ if t:
+ t.expand = 0
+ else:
+ t.expand = 2
+
+ # Generate report
+ if FORMAT == Format.TEXT:
+ output = campaign_to_TEXT(test_campaign)
+ elif FORMAT == Format.ANSI:
+ output = campaign_to_ANSI(test_campaign)
+ elif FORMAT == Format.HTML:
+ output = campaign_to_HTML(test_campaign, local=LOCAL)
+ elif FORMAT == Format.LATEX:
+ output = campaign_to_LATEX(test_campaign)
+ elif FORMAT == Format.XUNIT:
+ output = campaign_to_xUNIT(test_campaign)
+
+ OUTPUTFILE.write(output)
+ OUTPUTFILE.close()
+
+if __name__ == "__main__":
+ main(sys.argv[1:])