| #! /usr/bin/env python |
| # GLib Testing Framework Utility -*- Mode: python; -*- |
| # Copyright (C) 2007 Imendio AB |
| # Authors: Tim Janik |
| # |
| # This library is free software; you can redistribute it and/or |
| # modify it under the terms of the GNU Lesser General Public |
| # License as published by the Free Software Foundation; either |
| # version 2.1 of the License, or (at your option) any later version. |
| # |
| # This library is distributed in the hope that it will be useful, |
| # but WITHOUT ANY WARRANTY; without even the implied warranty of |
| # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| # Lesser General Public License for more details. |
| # |
| # You should have received a copy of the GNU Lesser General Public |
| # License along with this library; if not, see <http://www.gnu.org/licenses/>. |
| import datetime |
| import optparse |
| import sys, re, xml.dom.minidom |
| |
| try: |
| import subunit |
| from subunit import iso8601 |
| from testtools.content import Content, ContentType |
| mime_utf8 = ContentType('text', 'plain', {'charset': 'utf8'}) |
| except ImportError: |
| subunit = None |
| |
| |
| pkginstall_configvars = { |
| #@PKGINSTALL_CONFIGVARS_IN24LINES@ # configvars are substituted upon script installation |
| } |
| |
| # xml utilities |
| def find_child (node, child_name): |
| for child in node.childNodes: |
| if child.nodeName == child_name: |
| return child |
| return None |
| def list_children (node, child_name): |
| rlist = [] |
| for child in node.childNodes: |
| if child.nodeName == child_name: |
| rlist += [ child ] |
| return rlist |
| def find_node (node, name = None): |
| if not node or node.nodeName == name or not name: |
| return node |
| for child in node.childNodes: |
| c = find_node (child, name) |
| if c: |
| return c |
| return None |
| def node_as_text (node, name = None): |
| if name: |
| node = find_node (node, name) |
| txt = '' |
| if node: |
| if node.nodeValue: |
| txt += node.nodeValue |
| for child in node.childNodes: |
| txt += node_as_text (child) |
| return txt |
| def attribute_as_text (node, aname, node_name = None): |
| node = find_node (node, node_name) |
| if not node: |
| return '' |
| attr = node.attributes.get (aname, '') |
| if hasattr (attr, 'value'): |
| return attr.value |
| return '' |
| |
| # HTML utilities |
| def html_indent_string (n): |
| uncollapsible_space = ' ' # HTML won't compress alternating sequences of ' ' and ' ' |
| string = '' |
| for i in range (0, int((n + 1) / 2)): |
| string += uncollapsible_space |
| return string |
| |
| # TestBinary object, instantiated per test binary in the log file |
| class TestBinary: |
| def __init__ (self, name): |
| self.name = name |
| self.testcases = [] |
| self.duration = 0 |
| self.success_cases = 0 |
| self.skipped_cases = 0 |
| self.file = '???' |
| self.random_seed = '' |
| |
| # base class to handle processing/traversion of XML nodes |
| class TreeProcess: |
| def __init__ (self): |
| self.nest_level = 0 |
| def trampoline (self, node): |
| name = node.nodeName |
| if name == '#text': |
| self.handle_text (node) |
| else: |
| try: method = getattr (self, 'handle_' + re.sub ('[^a-zA-Z0-9]', '_', name)) |
| except: method = None |
| if method: |
| return method (node) |
| else: |
| return self.process_recursive (name, node) |
| def process_recursive (self, node_name, node): |
| self.process_children (node) |
| def process_children (self, node): |
| self.nest_level += 1 |
| for child in node.childNodes: |
| self.trampoline (child) |
| self.nest_level += 1 |
| |
| # test report reader, this class collects some statistics and merges duplicate test binary runs |
| class ReportReader (TreeProcess): |
| def __init__ (self): |
| TreeProcess.__init__ (self) |
| self.binary_names = [] |
| self.binaries = {} |
| self.last_binary = None |
| self.info = {} |
| def binary_list (self): |
| lst = [] |
| for name in self.binary_names: |
| lst += [ self.binaries[name] ] |
| return lst |
| def get_info (self): |
| return self.info |
| def handle_info (self, node): |
| dn = find_child (node, 'package') |
| self.info['package'] = node_as_text (dn) |
| dn = find_child (node, 'version') |
| self.info['version'] = node_as_text (dn) |
| dn = find_child (node, 'revision') |
| if dn is not None: |
| self.info['revision'] = node_as_text (dn) |
| def handle_testcase (self, node): |
| self.last_binary.testcases += [ node ] |
| result = attribute_as_text (node, 'result', 'status') |
| if result == 'success': |
| self.last_binary.success_cases += 1 |
| if bool (int (attribute_as_text (node, 'skipped') + '0')): |
| self.last_binary.skipped_cases += 1 |
| def handle_text (self, node): |
| pass |
| def handle_testbinary (self, node): |
| path = node.attributes.get ('path', None).value |
| if self.binaries.get (path, -1) == -1: |
| self.binaries[path] = TestBinary (path) |
| self.binary_names += [ path ] |
| self.last_binary = self.binaries[path] |
| dn = find_child (node, 'duration') |
| dur = node_as_text (dn) |
| try: dur = float (dur) |
| except: dur = 0 |
| if dur: |
| self.last_binary.duration += dur |
| bin = find_child (node, 'binary') |
| if bin: |
| self.last_binary.file = attribute_as_text (bin, 'file') |
| rseed = find_child (node, 'random-seed') |
| if rseed: |
| self.last_binary.random_seed = node_as_text (rseed) |
| self.process_children (node) |
| |
| |
| class ReportWriter(object): |
| """Base class for reporting.""" |
| |
| def __init__(self, binary_list): |
| self.binaries = binary_list |
| |
| def _error_text(self, node): |
| """Get a string representing the error children of node.""" |
| rlist = list_children(node, 'error') |
| txt = '' |
| for enode in rlist: |
| txt += node_as_text (enode) |
| if txt and txt[-1] != '\n': |
| txt += '\n' |
| return txt |
| |
| |
| class HTMLReportWriter(ReportWriter): |
| # Javascript/CSS snippet to toggle element visibility |
| cssjs = r''' |
| <style type="text/css" media="screen"> |
| .VisibleSection { } |
| .HiddenSection { display: none; } |
| </style> |
| <script language="javascript" type="text/javascript"><!-- |
| function toggle_display (parentid, tagtype, idmatch, keymatch) { |
| ptag = document.getElementById (parentid); |
| tags = ptag.getElementsByTagName (tagtype); |
| for (var i = 0; i < tags.length; i++) { |
| tag = tags[i]; |
| var key = tag.getAttribute ("keywords"); |
| if (tag.id.indexOf (idmatch) == 0 && key && key.match (keymatch)) { |
| if (tag.className.indexOf ("HiddenSection") >= 0) |
| tag.className = "VisibleSection"; |
| else |
| tag.className = "HiddenSection"; |
| } |
| } |
| } |
| message_array = Array(); |
| function view_testlog (wname, file, random_seed, tcase, msgtitle, msgid) { |
| txt = message_array[msgid]; |
| var w = window.open ("", // URI |
| wname, |
| "resizable,scrollbars,status,width=790,height=400"); |
| var doc = w.document; |
| doc.write ("<h2>File: " + file + "</h2>\n"); |
| doc.write ("<h3>Case: " + tcase + "</h3>\n"); |
| doc.write ("<strong>Random Seed:</strong> <code>" + random_seed + "</code> <br /><br />\n"); |
| doc.write ("<strong>" + msgtitle + "</strong><br />\n"); |
| doc.write ("<pre>"); |
| doc.write (txt); |
| doc.write ("</pre>\n"); |
| doc.write ("<a href=\'javascript:window.close()\'>Close Window</a>\n"); |
| doc.close(); |
| } |
| --></script> |
| ''' |
| def __init__ (self, info, binary_list): |
| ReportWriter.__init__(self, binary_list) |
| self.info = info |
| self.bcounter = 0 |
| self.tcounter = 0 |
| self.total_tcounter = 0 |
| self.total_fcounter = 0 |
| self.total_duration = 0 |
| self.indent_depth = 0 |
| self.lastchar = '' |
| def oprint (self, message): |
| sys.stdout.write (message) |
| if message: |
| self.lastchar = message[-1] |
| def handle_info (self): |
| if 'package' in self.info and 'version' in self.info: |
| self.oprint ('<h3>Package: %(package)s, version: %(version)s</h3>\n' % self.info) |
| if 'revision' in self.info: |
| self.oprint ('<h5>Report generated from: %(revision)s</h5>\n' % self.info) |
| def handle_text (self, node): |
| self.oprint (node.nodeValue) |
| def handle_testcase (self, node, binary): |
| skipped = bool (int (attribute_as_text (node, 'skipped') + '0')) |
| if skipped: |
| return # skipped tests are uninteresting for HTML reports |
| path = attribute_as_text (node, 'path') |
| duration = node_as_text (node, 'duration') |
| result = attribute_as_text (node, 'result', 'status') |
| rcolor = { |
| 'success': 'bgcolor="lightgreen"', |
| 'failed': 'bgcolor="red"', |
| }.get (result, '') |
| if result != 'success': |
| duration = '-' # ignore bogus durations |
| self.oprint ('<tr id="b%u_t%u_" keywords="%s all" class="HiddenSection">\n' % (self.bcounter, self.tcounter, result)) |
| self.oprint ('<td>%s %s</td> <td align="right">%s</td> \n' % (html_indent_string (4), path, duration)) |
| perflist = list_children (node, 'performance') |
| if result != 'success': |
| txt = self._error_text(node) |
| txt = re.sub (r'"', r'\\"', txt) |
| txt = re.sub (r'\n', r'\\n', txt) |
| txt = re.sub (r'&', r'&', txt) |
| txt = re.sub (r'<', r'<', txt) |
| self.oprint ('<script language="javascript" type="text/javascript">message_array["b%u_t%u_"] = "%s";</script>\n' % (self.bcounter, self.tcounter, txt)) |
| self.oprint ('<td align="center"><a href="javascript:view_testlog (\'%s\', \'%s\', \'%s\', \'%s\', \'Output:\', \'b%u_t%u_\')">Details</a></td>\n' % |
| ('TestResultWindow', binary.file, binary.random_seed, path, self.bcounter, self.tcounter)) |
| elif perflist: |
| presults = [] |
| for perf in perflist: |
| pmin = bool (int (attribute_as_text (perf, 'minimize'))) |
| pmax = bool (int (attribute_as_text (perf, 'maximize'))) |
| pval = float (attribute_as_text (perf, 'value')) |
| txt = node_as_text (perf) |
| txt = re.sub (r'&', r'&', txt) |
| txt = re.sub (r'<', r'>', txt) |
| txt = '<strong>Performance(' + (pmin and '<em>minimized</em>' or '<em>maximized</em>') + '):</strong> ' + txt.strip() + '<br />\n' |
| txt = re.sub (r'"', r'\\"', txt) |
| txt = re.sub (r'\n', r'\\n', txt) |
| presults += [ (pval, txt) ] |
| presults.sort() |
| ptxt = ''.join ([e[1] for e in presults]) |
| self.oprint ('<script language="javascript" type="text/javascript">message_array["b%u_t%u_"] = "%s";</script>\n' % (self.bcounter, self.tcounter, ptxt)) |
| self.oprint ('<td align="center"><a href="javascript:view_testlog (\'%s\', \'%s\', \'%s\', \'%s\', \'Test Results:\', \'b%u_t%u_\')">Details</a></td>\n' % |
| ('TestResultWindow', binary.file, binary.random_seed, path, self.bcounter, self.tcounter)) |
| else: |
| self.oprint ('<td align="center">-</td>\n') |
| self.oprint ('<td align="right" %s>%s</td>\n' % (rcolor, result)) |
| self.oprint ('</tr>\n') |
| self.tcounter += 1 |
| self.total_tcounter += 1 |
| self.total_fcounter += result != 'success' |
| def handle_binary (self, binary): |
| self.tcounter = 1 |
| self.bcounter += 1 |
| self.total_duration += binary.duration |
| self.oprint ('<tr><td><strong>%s</strong></td><td align="right">%f</td> <td align="center">\n' % (binary.name, binary.duration)) |
| erlink, oklink = ('', '') |
| real_cases = len (binary.testcases) - binary.skipped_cases |
| if binary.success_cases < real_cases: |
| erlink = 'href="javascript:toggle_display (\'ResultTable\', \'tr\', \'b%u_\', \'failed\')"' % self.bcounter |
| if binary.success_cases: |
| oklink = 'href="javascript:toggle_display (\'ResultTable\', \'tr\', \'b%u_\', \'success\')"' % self.bcounter |
| if real_cases != 0: |
| self.oprint ('<a %s>ER</a>\n' % erlink) |
| self.oprint ('<a %s>OK</a>\n' % oklink) |
| self.oprint ('</td>\n') |
| perc = binary.success_cases * 100.0 / real_cases |
| pcolor = { |
| 100 : 'bgcolor="lightgreen"', |
| 0 : 'bgcolor="red"', |
| }.get (int (perc), 'bgcolor="yellow"') |
| self.oprint ('<td align="right" %s>%.2f%%</td>\n' % (pcolor, perc)) |
| self.oprint ('</tr>\n') |
| else: |
| self.oprint ('Empty\n') |
| self.oprint ('</td>\n') |
| self.oprint ('</tr>\n') |
| for tc in binary.testcases: |
| self.handle_testcase (tc, binary) |
| def handle_totals (self): |
| self.oprint ('<tr>') |
| self.oprint ('<td><strong>Totals:</strong> %u Binaries, %u Tests, %u Failed, %u Succeeded</td>' % |
| (self.bcounter, self.total_tcounter, self.total_fcounter, self.total_tcounter - self.total_fcounter)) |
| self.oprint ('<td align="right">%f</td>\n' % self.total_duration) |
| self.oprint ('<td align="center">-</td>\n') |
| if self.total_tcounter != 0: |
| perc = (self.total_tcounter - self.total_fcounter) * 100.0 / self.total_tcounter |
| else: |
| perc = 0.0 |
| pcolor = { |
| 100 : 'bgcolor="lightgreen"', |
| 0 : 'bgcolor="red"', |
| }.get (int (perc), 'bgcolor="yellow"') |
| self.oprint ('<td align="right" %s>%.2f%%</td>\n' % (pcolor, perc)) |
| self.oprint ('</tr>\n') |
| def printout (self): |
| self.oprint ('<html><head>\n') |
| self.oprint ('<title>GTester Unit Test Report</title>\n') |
| self.oprint (self.cssjs) |
| self.oprint ('</head>\n') |
| self.oprint ('<body>\n') |
| self.oprint ('<h2>GTester Unit Test Report</h2>\n') |
| self.handle_info () |
| self.oprint ('<table id="ResultTable" width="100%" border="1">\n<tr>\n') |
| self.oprint ('<th>Program / Testcase </th>\n') |
| self.oprint ('<th style="width:8em">Duration (sec)</th>\n') |
| self.oprint ('<th style="width:5em">View</th>\n') |
| self.oprint ('<th style="width:5em">Result</th>\n') |
| self.oprint ('</tr>\n') |
| for tb in self.binaries: |
| self.handle_binary (tb) |
| self.handle_totals() |
| self.oprint ('</table>\n') |
| self.oprint ('</body>\n') |
| self.oprint ('</html>\n') |
| |
| |
| class SubunitWriter(ReportWriter): |
| """Reporter to output a subunit stream.""" |
| |
| def printout(self): |
| reporter = subunit.TestProtocolClient(sys.stdout) |
| for binary in self.binaries: |
| for tc in binary.testcases: |
| test = GTestCase(tc, binary) |
| test.run(reporter) |
| |
| |
| class GTestCase(object): |
| """A representation of a gtester test result as a pyunit TestCase.""" |
| |
| def __init__(self, case, binary): |
| """Create a GTestCase for case 'case' from binary program 'binary'.""" |
| self._case = case |
| self._binary = binary |
| # the name of the case - e.g. /dbusmenu/glib/objects/menuitem/props_boolstr |
| self._path = attribute_as_text(self._case, 'path') |
| |
| def id(self): |
| """What test is this? Returns the gtester path for the testcase.""" |
| return self._path |
| |
| def _get_details(self): |
| """Calculate a details dict for the test - attachments etc.""" |
| details = {} |
| result = attribute_as_text(self._case, 'result', 'status') |
| details['filename'] = Content(mime_utf8, lambda:[self._binary.file]) |
| details['random_seed'] = Content(mime_utf8, |
| lambda:[self._binary.random_seed]) |
| if self._get_outcome() == 'addFailure': |
| # Extract the error details. Skips have no details because its not |
| # skip like unittest does, instead the runner just bypasses N test. |
| txt = self._error_text(self._case) |
| details['error'] = Content(mime_utf8, lambda:[txt]) |
| if self._get_outcome() == 'addSuccess': |
| # Sucessful tests may have performance metrics. |
| perflist = list_children(self._case, 'performance') |
| if perflist: |
| presults = [] |
| for perf in perflist: |
| pmin = bool (int (attribute_as_text (perf, 'minimize'))) |
| pmax = bool (int (attribute_as_text (perf, 'maximize'))) |
| pval = float (attribute_as_text (perf, 'value')) |
| txt = node_as_text (perf) |
| txt = 'Performance(' + (pmin and 'minimized' or 'maximized' |
| ) + '): ' + txt.strip() + '\n' |
| presults += [(pval, txt)] |
| presults.sort() |
| perf_details = [e[1] for e in presults] |
| details['performance'] = Content(mime_utf8, lambda:perf_details) |
| return details |
| |
| def _get_outcome(self): |
| if int(attribute_as_text(self._case, 'skipped') + '0'): |
| return 'addSkip' |
| outcome = attribute_as_text(self._case, 'result', 'status') |
| if outcome == 'success': |
| return 'addSuccess' |
| else: |
| return 'addFailure' |
| |
| def run(self, result): |
| time = datetime.datetime.utcnow().replace(tzinfo=iso8601.Utc()) |
| result.time(time) |
| result.startTest(self) |
| try: |
| outcome = self._get_outcome() |
| details = self._get_details() |
| # Only provide a duration IFF outcome == 'addSuccess' - the main |
| # parser claims bogus results otherwise: in that case emit time as |
| # zero perhaps. |
| if outcome == 'addSuccess': |
| duration = float(node_as_text(self._case, 'duration')) |
| duration = duration * 1000000 |
| timedelta = datetime.timedelta(0, 0, duration) |
| time = time + timedelta |
| result.time(time) |
| getattr(result, outcome)(self, details=details) |
| finally: |
| result.stopTest(self) |
| |
| |
| |
| # main program handling |
| def parse_opts(): |
| """Parse program options. |
| |
| :return: An options object and the program arguments. |
| """ |
| parser = optparse.OptionParser() |
| parser.version = pkginstall_configvars.get ('glib-version', '0.0-uninstalled') |
| parser.usage = "%prog [OPTIONS] <gtester-log.xml>" |
| parser.description = "Generate HTML reports from the XML log files generated by gtester." |
| parser.epilog = "gtester-report (GLib utils) version %s."% (parser.version,) |
| parser.add_option("-v", "--version", action="store_true", dest="version", default=False, |
| help="Show program version.") |
| parser.add_option("-s", "--subunit", action="store_true", dest="subunit", default=False, |
| help="Output subunit [See https://launchpad.net/subunit/" |
| " Needs python-subunit]") |
| options, files = parser.parse_args() |
| if options.version: |
| print(parser.epilog) |
| return None, None |
| if len(files) != 1: |
| parser.error("Must supply a log file to parse.") |
| if options.subunit and subunit is None: |
| parser.error("python-subunit is not installed.") |
| return options, files |
| |
| |
| def main(): |
| options, files = parse_opts() |
| if options is None: |
| return 0 |
| xd = xml.dom.minidom.parse (files[0]) |
| rr = ReportReader() |
| rr.trampoline (xd) |
| if not options.subunit: |
| HTMLReportWriter(rr.get_info(), rr.binary_list()).printout() |
| else: |
| SubunitWriter(rr.get_info(), rr.binary_list()).printout() |
| |
| |
| if __name__ == '__main__': |
| main() |