965 lines
34 KiB
Python
Executable File
965 lines
34 KiB
Python
Executable File
#!/usr/bin/env python
|
|
|
|
from __future__ import unicode_literals
|
|
|
|
from datetime import date
|
|
import io
|
|
import locale
|
|
import operator
|
|
import optparse
|
|
import os
|
|
import sys
|
|
import subprocess
|
|
|
|
from collections import Counter
|
|
from pygments import highlight
|
|
from pygments.lexers import guess_lexer, guess_lexer_for_filename
|
|
from pygments.formatters import HtmlFormatter # pylint: disable=no-name-in-module
|
|
from pygments.util import ClassNotFound
|
|
from xml.sax import parse as xml_parse
|
|
from xml.sax import SAXParseException as XmlParseException
|
|
from xml.sax.handler import ContentHandler as XmlContentHandler
|
|
from xml.sax.saxutils import escape
|
|
"""
|
|
Turns a cppcheck xml file into a browsable html report along
|
|
with syntax highlighted source code.
|
|
"""
|
|
|
|
STYLE_FILE = """
|
|
body {
|
|
font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, "Noto Sans", sans-serif;
|
|
font-size: 13px;
|
|
line-height: 1.5;
|
|
height: 100%;
|
|
margin: 0;
|
|
}
|
|
|
|
#wrapper {
|
|
position: fixed;
|
|
height: 100vh;
|
|
width: 100vw;
|
|
display: grid;
|
|
grid-template-rows: fit-content(8rem) auto fit-content(8rem);
|
|
grid-template-columns: fit-content(25%) 1fr;
|
|
grid-template-areas:
|
|
"header header"
|
|
"menu content"
|
|
"footer footer";
|
|
}
|
|
|
|
h1 {
|
|
margin: 0 0 8px -2px;
|
|
font-size: 175%;
|
|
}
|
|
|
|
.header {
|
|
padding: 0 0 5px 15px;
|
|
grid-area: header;
|
|
border-bottom: thin solid #aaa;
|
|
}
|
|
|
|
.footer {
|
|
grid-area: footer;
|
|
border-top: thin solid #aaa;
|
|
font-size: 85%;
|
|
|
|
}
|
|
|
|
.footer > p {
|
|
margin: 4px;
|
|
}
|
|
|
|
#menu,
|
|
#menu_index {
|
|
grid-area: menu;
|
|
text-align: left;
|
|
overflow: auto;
|
|
padding: 0 23px 15px 15px;
|
|
border-right: thin solid #aaa;
|
|
min-width: 200px;
|
|
}
|
|
|
|
#menu > a {
|
|
display: block;
|
|
margin-left: 10px;
|
|
font-size: 12px;
|
|
}
|
|
|
|
#content,
|
|
#content_index {
|
|
grid-area: content;
|
|
padding: 0px 5px 15px 15px;
|
|
overflow: auto;
|
|
}
|
|
|
|
label {
|
|
white-space: nowrap;
|
|
}
|
|
|
|
label.checkBtn.disabled {
|
|
color: #606060;
|
|
background: #e0e0e0;
|
|
font-style: italic;
|
|
}
|
|
|
|
label.checkBtn, input[type="text"] {
|
|
border: 1px solid grey;
|
|
border-radius: 4px;
|
|
box-shadow: 1px 1px inset;
|
|
padding: 1px 5px;
|
|
}
|
|
|
|
label.checkBtn {
|
|
white-space: nowrap;
|
|
background: #ccddff;
|
|
}
|
|
|
|
label.unchecked {
|
|
background: #eff8ff;
|
|
box-shadow: 1px 1px 1px;
|
|
}
|
|
|
|
label.checkBtn:hover, label.unchecked:hover{
|
|
box-shadow: 0 0 2px;
|
|
}
|
|
|
|
label.disabled:hover {
|
|
box-shadow: 1px 1px inset;
|
|
}
|
|
|
|
label.checkBtn > input {
|
|
display:none;
|
|
}
|
|
|
|
.summaryTable {
|
|
width: 100%;
|
|
}
|
|
|
|
table.summaryTable td { padding: 0 5px 0 5px; }
|
|
|
|
.statHeader, .severityHeader {
|
|
font-weight: bold;
|
|
}
|
|
|
|
.warning {
|
|
background-color: #ffffa7;
|
|
}
|
|
|
|
.error {
|
|
background-color: #ffb7b7;
|
|
}
|
|
|
|
.error2 {
|
|
background-color: #faa;
|
|
display: inline-block;
|
|
margin-left: 4px;
|
|
}
|
|
|
|
.inconclusive {
|
|
background-color: #b6b6b4;
|
|
}
|
|
|
|
.inconclusive2 {
|
|
background-color: #b6b6b4;
|
|
display: inline-block;
|
|
margin-left: 4px;
|
|
}
|
|
|
|
.verbose {
|
|
display: inline-block;
|
|
vertical-align: top;
|
|
cursor: help;
|
|
}
|
|
|
|
.verbose .content {
|
|
display: none;
|
|
position: absolute;
|
|
padding: 10px;
|
|
margin: 4px;
|
|
max-width: 40%;
|
|
white-space: pre-wrap;
|
|
border: 1px solid #000;
|
|
background-color: #ffffcc;
|
|
cursor: auto;
|
|
}
|
|
|
|
.highlight .hll {
|
|
padding: 1px;
|
|
}
|
|
|
|
.highlighttable {
|
|
background-color: #fff;
|
|
position: relative;
|
|
margin: -10px;
|
|
}
|
|
|
|
.linenos {
|
|
border-right: thin solid #aaa;
|
|
color: #d3d3d3;
|
|
padding-right: 6px;
|
|
}
|
|
|
|
.id-filtered, .severity-filtered, .file-filtered, .tool-filtered, .text-filtered {
|
|
visibility: collapse;
|
|
}
|
|
"""
|
|
|
|
HTML_HEAD = """
|
|
<!doctype html>
|
|
<html lang="en">
|
|
<head>
|
|
<meta charset="utf-8">
|
|
<title>Cppcheck - HTML report - %s</title>
|
|
<link rel="stylesheet" href="style.css">
|
|
<style>
|
|
%s
|
|
</style>
|
|
<script>
|
|
function getStyle(el, styleProp) {
|
|
var y;
|
|
|
|
if (el.currentStyle) {
|
|
y = el.currentStyle[styleProp];
|
|
} else if (window.getComputedStyle) {
|
|
y = document.defaultView.getComputedStyle(el, null).getPropertyValue(styleProp);
|
|
}
|
|
|
|
return y;
|
|
}
|
|
|
|
function toggle() {
|
|
var el = this.expandable_content;
|
|
var mark = this.expandable_marker;
|
|
|
|
if (el.style.display === "block") {
|
|
el.style.display = "none";
|
|
mark.textContent = "[+]";
|
|
} else {
|
|
el.style.display = "block";
|
|
mark.textContent = "[-]";
|
|
}
|
|
}
|
|
|
|
function initExpandables() {
|
|
var elements = document.querySelectorAll(".expandable");
|
|
|
|
for (var i = 0, len = elements.length; i < len; i++) {
|
|
var el = elements[i];
|
|
var clickable = el.querySelector("span");
|
|
var marker = clickable.querySelector(".marker");
|
|
var content = el.querySelector(".content");
|
|
var width = clickable.clientWidth - parseInt(getStyle(content, "padding-left")) - parseInt(getStyle(content, "padding-right"));
|
|
content.style.width = width + "px";
|
|
clickable.expandable_content = content;
|
|
clickable.expandable_marker = marker;
|
|
clickable.addEventListener("click", toggle);
|
|
}
|
|
}
|
|
|
|
function toggleDisplay(cb) {
|
|
var elements = document.querySelectorAll("." + cb.id);
|
|
|
|
for (var i = 0, len = elements.length; i < len; i++) {
|
|
elements[i].classList.toggle("id-filtered", !cb.checked);
|
|
}
|
|
|
|
updateFileRows();
|
|
}
|
|
|
|
function toggleSeverity(cb) {
|
|
cb.parentElement.classList.toggle("unchecked", !cb.checked);
|
|
var elements = document.querySelectorAll(".sev_" + cb.id);
|
|
|
|
for (var i = 0, len = elements.length; i < len; i++) {
|
|
elements[i].classList.toggle("severity-filtered", !cb.checked);
|
|
}
|
|
|
|
updateFileRows();
|
|
}
|
|
|
|
function toggleTool(cb) {
|
|
cb.parentElement.classList.toggle("unchecked", !cb.checked);
|
|
|
|
var elements;
|
|
if (cb.id == "clang-tidy")
|
|
elements = document.querySelectorAll("[class^=clang-tidy-]");
|
|
else
|
|
elements = document.querySelectorAll(".issue:not([class^=clang-tidy-])");
|
|
|
|
for (var i = 0, len = elements.length; i < len; i++) {
|
|
elements[i].classList.toggle("tool-filtered", !cb.checked);
|
|
}
|
|
|
|
updateFileRows();
|
|
}
|
|
|
|
function toggleAll() {
|
|
var elements = document.querySelectorAll(".idToggle");
|
|
|
|
// starting from 1 since 0 is the "toggle all" input
|
|
for (var i = 1, len = elements.length; i < len; i++) {
|
|
var changed = elements[i].checked != elements[0].checked;
|
|
if (changed) {
|
|
elements[i].checked = elements[0].checked;
|
|
toggleDisplay(elements[i]);
|
|
}
|
|
}
|
|
}
|
|
|
|
function filterFile(filter) {
|
|
var elements = document.querySelectorAll(".fileEntry");
|
|
|
|
for (var i = 0, len = elements.length; i < len; i++) {
|
|
var visible = elements[i].querySelector("tr").querySelector("td").textContent.toLowerCase().includes(filter.toLowerCase());
|
|
elements[i].classList.toggle("text-filtered", !visible);
|
|
}
|
|
}
|
|
|
|
function filterText(text) {
|
|
filter = text.toLowerCase();
|
|
var elements = document.querySelectorAll(".issue");
|
|
|
|
for (var i = 0, len = elements.length; i < len; i++) {
|
|
var visible = false;
|
|
var fields = elements[i].querySelectorAll("td");
|
|
for (var n = 0, num = fields.length; n < num; n++) {
|
|
if (fields[n].textContent.toLowerCase().includes(filter)) {
|
|
visible = true;
|
|
break;
|
|
}
|
|
}
|
|
elements[i].classList.toggle("text-filtered", !visible);
|
|
}
|
|
|
|
updateFileRows();
|
|
}
|
|
|
|
function updateFileRows(element) {
|
|
var elements = document.querySelectorAll(".fileEntry");
|
|
|
|
for (var i = 0, len = elements.length; i < len; i++) {
|
|
var visible = elements[i].querySelector(".issue:not(.id-filtered):not(.severity-filtered):not(.tool-filtered):not(.text-filtered)");
|
|
elements[i].classList.toggle("file-filtered", !visible);
|
|
}
|
|
}
|
|
|
|
window.addEventListener("load", initExpandables);
|
|
</script>
|
|
</head>
|
|
<body>
|
|
<div id="wrapper">
|
|
<div id="header" class="header">
|
|
<h1>Cppcheck report - %s%s</h1>
|
|
"""
|
|
|
|
HTML_HEAD_END = """
|
|
</div>
|
|
"""
|
|
|
|
HTML_MENU = """
|
|
<div id="menu">
|
|
<p><a href="index.html">Defects:</a> %s</p>
|
|
"""
|
|
|
|
HTML_MENU_END = """
|
|
</div>
|
|
<div id="content">
|
|
"""
|
|
|
|
HTML_FOOTER = """
|
|
</div>
|
|
<div id="footer" class="footer">
|
|
<p>
|
|
Created by Cppcheck %s (<a href="https://cppcheck.sourceforge.io">Sourceforge</a>, <a href="irc://irc.freenode.net/cppcheck">IRC</a>)
|
|
</p>
|
|
</div>
|
|
</div>
|
|
</body>
|
|
</html>
|
|
"""
|
|
|
|
HTML_ERROR = "<span class=\"error2\"><--- %s</span>\n"
|
|
HTML_INCONCLUSIVE = "<span class=\"inconclusive2\"><--- %s</span>\n"
|
|
|
|
HTML_EXPANDABLE_ERROR = "<div class=\"verbose expandable\"><span class=\"error2\"><--- %s <span class=\"marker\">[+]</span></span><div class=\"content\">%s</div></div>\n"""
|
|
HTML_EXPANDABLE_INCONCLUSIVE = "<div class=\"verbose expandable\"><span class=\"inconclusive2\"><--- %s <span class=\"marker\">[+]</span></span><div class=\"content\">%s</div></div>\n"""
|
|
|
|
# escape() and unescape() takes care of &, < and >.
|
|
html_escape_table = {
|
|
'"': """,
|
|
"'": "'"
|
|
}
|
|
html_unescape_table = {v: k for k, v in html_escape_table.items()}
|
|
|
|
|
|
def html_escape(text):
|
|
return escape(text, html_escape_table)
|
|
|
|
def filter_button(enabled_filters, id, function):
|
|
enabled = enabled_filters.get(id, False)
|
|
return '\n <label class="checkBtn%s"><input type="checkbox" onclick="%s(this)" id="%s"%s/>%s</label>'\
|
|
% (' disabled' if not enabled else '', function, id, 'checked' if enabled else 'disabled', id)
|
|
|
|
def filter_bar(enabled):
|
|
return ''.join([
|
|
' <div id="filters">\n'
|
|
,''.join([filter_button(enabled, severity, 'toggleSeverity') for severity in ['error', 'warning', 'portability', 'performance', 'style', 'information']])
|
|
,'\n | '
|
|
,''.join([filter_button(enabled, tool, 'toggleTool') for tool in ['cppcheck', 'clang-tidy']])
|
|
,'\n | '
|
|
,'\n <label class="severityHeader">File: <input type="text" oninput="filterFile(this.value)"/></label>'
|
|
,'\n <label class="severityHeader">Filter: <input type="text" oninput="filterText(this.value)"/></label>'
|
|
,'\n </div>\n'])
|
|
|
|
def git_blame(errors, path, file, blame_options):
|
|
last_line= errors[-1]['line']
|
|
if last_line == 0:
|
|
return {}
|
|
|
|
first_line = next((error for error in errors if error['line'] > 0))['line']
|
|
|
|
full_path = os.path.join(path, file)
|
|
path, filename = os.path.split(full_path)
|
|
|
|
if path:
|
|
os.chdir(path)
|
|
|
|
cmd_args = ['git', 'blame', '-L %d,%d' % (first_line, last_line)]
|
|
if '-w' in blame_options:
|
|
cmd_args.append('-w')
|
|
if '-M' in blame_options:
|
|
cmd_args.append('-M')
|
|
cmd_args = cmd_args + ['--porcelain', '--incremental', '--', filename]
|
|
|
|
try:
|
|
result = subprocess.check_output(cmd_args)
|
|
result = result.decode(locale.getpreferredencoding())
|
|
except:
|
|
return []
|
|
finally:
|
|
os.chdir(cwd)
|
|
|
|
if result.startswith('fatal'):
|
|
return []
|
|
|
|
blame_data = []
|
|
line_from = 0
|
|
line_to = 0
|
|
blame_info = dict()
|
|
|
|
for line_str in result.splitlines():
|
|
field, _, value = line_str.partition(' ')
|
|
if len(field) == 40:
|
|
line_nr, count = value.split(' ')[1:]
|
|
line_from = int(line_nr)
|
|
line_to = line_from + int(count)
|
|
elif field.startswith('author'):
|
|
blame_info[field] = value
|
|
if field == 'author-time':
|
|
author_date = date.fromtimestamp(int(blame_info['author-time']))
|
|
blame_info['author-time'] = author_date.strftime("%d/%m/%Y")
|
|
elif field == 'filename':
|
|
blame_data.append([line_from, line_to, blame_info.copy()])
|
|
|
|
return blame_data
|
|
|
|
|
|
def blame_lookup(blame_data, line):
|
|
return next((data for start, end, data in blame_data if line >= start and line < end), {})
|
|
|
|
|
|
def tr_str(td_th, line, id, cwe, severity, message, author, author_mail, date, add_author, tr_class=None, htmlfile=None, message_class=None):
|
|
ret = ''
|
|
if htmlfile:
|
|
ret += '<%s><a href="%s#line-%d">%d</a></%s>' % (td_th, htmlfile, line, line, td_th)
|
|
for item in (id, cwe, severity):
|
|
ret += '<%s>%s</%s>' % (td_th, item, td_th)
|
|
else:
|
|
for item in (line, id, cwe, severity):
|
|
ret += '<%s>%s</%s>' % (td_th, item, td_th)
|
|
if message_class:
|
|
message_attribute = ' class="%s"' % message_class
|
|
else:
|
|
message_attribute = ''
|
|
ret += '<%s%s>%s</%s>' % (td_th, message_attribute, html_escape(message), td_th)
|
|
|
|
for field in add_author:
|
|
if field == 'name':
|
|
ret += '<%s>%s</%s>' % (td_th, html_escape(author), td_th)
|
|
elif field == 'email':
|
|
ret += '<%s>%s</%s>' % (td_th, html_escape(author_mail), td_th)
|
|
elif field == 'date':
|
|
ret += '<%s>%s</%s>' % (td_th, date, td_th)
|
|
|
|
if tr_class:
|
|
tr_attributes = ' class="%s"' % tr_class
|
|
else:
|
|
tr_attributes = ''
|
|
return '<tr%s>%s</tr>' % (tr_attributes, ret)
|
|
|
|
|
|
class AnnotateCodeFormatter(HtmlFormatter):
|
|
errors = []
|
|
|
|
def wrap(self, source, outfile):
|
|
line_no = 1
|
|
for i, t in HtmlFormatter.wrap(self, source, outfile):
|
|
# If this is a source code line we want to add a span tag at the
|
|
# end.
|
|
if i == 1:
|
|
for error in self.errors:
|
|
if error['line'] == line_no:
|
|
try:
|
|
if error['inconclusive'] == 'true':
|
|
# only print verbose msg if it really differs
|
|
# from actual message
|
|
if error.get('verbose') and (error['verbose'] != error['msg']):
|
|
index = t.rfind('\n')
|
|
t = t[:index] + HTML_EXPANDABLE_INCONCLUSIVE % (error['msg'], html_escape(error['verbose'].replace("\\012", '\n'))) + t[index + 1:]
|
|
else:
|
|
t = t.replace('\n', HTML_INCONCLUSIVE % error['msg'])
|
|
except KeyError:
|
|
if error.get('verbose') and (error['verbose'] != error['msg']):
|
|
index = t.rfind('\n')
|
|
t = t[:index] + HTML_EXPANDABLE_ERROR % (error['msg'], html_escape(error['verbose'].replace("\\012", '\n'))) + t[index + 1:]
|
|
else:
|
|
t = t.replace('\n', HTML_ERROR % error['msg'])
|
|
|
|
line_no = line_no + 1
|
|
yield i, t
|
|
|
|
|
|
class CppCheckHandler(XmlContentHandler):
|
|
|
|
"""Parses the cppcheck xml file and produces a list of all its errors."""
|
|
|
|
def __init__(self):
|
|
XmlContentHandler.__init__(self)
|
|
self.errors = []
|
|
self.version = '1'
|
|
self.versionCppcheck = ''
|
|
|
|
def startElement(self, name, attributes):
|
|
if name == 'results':
|
|
self.version = attributes.get('version', self.version)
|
|
|
|
if self.version == '1':
|
|
self.handleVersion1(name, attributes)
|
|
else:
|
|
self.handleVersion2(name, attributes)
|
|
|
|
def handleVersion1(self, name, attributes):
|
|
if name != 'error':
|
|
return
|
|
|
|
self.errors.append({
|
|
'file': attributes.get('file', ''),
|
|
'line': int(attributes.get('line', 0)),
|
|
'locations': [{
|
|
'file': attributes.get('file', ''),
|
|
'line': int(attributes.get('line', 0)),
|
|
}],
|
|
'id': attributes['id'],
|
|
'severity': attributes['severity'],
|
|
'msg': attributes['msg']
|
|
})
|
|
|
|
def handleVersion2(self, name, attributes):
|
|
if name == 'cppcheck':
|
|
self.versionCppcheck = attributes['version']
|
|
if name == 'error':
|
|
error = {
|
|
'locations': [],
|
|
'file': '',
|
|
'line': 0,
|
|
'id': attributes['id'],
|
|
'severity': attributes['severity'],
|
|
'msg': attributes['msg'],
|
|
'verbose': attributes.get('verbose')
|
|
}
|
|
|
|
if 'inconclusive' in attributes:
|
|
error['inconclusive'] = attributes['inconclusive']
|
|
if 'cwe' in attributes:
|
|
error['cwe'] = attributes['cwe']
|
|
|
|
self.errors.append(error)
|
|
elif name == 'location':
|
|
assert self.errors
|
|
error = self.errors[-1]
|
|
locations = error['locations']
|
|
file = attributes['file']
|
|
line = int(attributes['line'])
|
|
if not locations:
|
|
error['file'] = file
|
|
error['line'] = line
|
|
locations.append({
|
|
'file': file,
|
|
'line': line,
|
|
'info': attributes.get('info')
|
|
})
|
|
|
|
if __name__ == '__main__':
|
|
# Configure all the options this little utility is using.
|
|
parser = optparse.OptionParser()
|
|
parser.add_option('--title', dest='title',
|
|
help='The title of the project.',
|
|
default='[project name]')
|
|
parser.add_option('--file', dest='file', action="append",
|
|
help='The cppcheck xml output file to read defects '
|
|
'from. You can combine results from several '
|
|
'xml reports i.e. "--file file1.xml --file file2.xml ..". '
|
|
'Default is reading from stdin.')
|
|
parser.add_option('--report-dir', dest='report_dir',
|
|
help='The directory where the HTML report content is '
|
|
'written.')
|
|
parser.add_option('--source-dir', dest='source_dir',
|
|
help='Base directory where source code files can be '
|
|
'found.')
|
|
parser.add_option('--add-author-information', dest='add_author_information',
|
|
help='Blame information to include. '
|
|
'Adds specified author information. '
|
|
'Specify as comma-separated list of either "name", "email", "date" or "n","e","d". '
|
|
'Default: "n,e,d"')
|
|
parser.add_option('--source-encoding', dest='source_encoding',
|
|
help='Encoding of source code.', default='utf-8')
|
|
parser.add_option('--blame-options', dest='blame_options',
|
|
help='[-w, -M] blame options which you can use to get author and author mail '
|
|
'-w --> not including white spaces and returns original author of the line '
|
|
'-M --> not including moving of lines and returns original author of the line')
|
|
|
|
# Parse options and make sure that we have an output directory set.
|
|
options, args = parser.parse_args()
|
|
|
|
try:
|
|
sys.argv[1]
|
|
except IndexError: # no arguments give, print --help
|
|
parser.print_help()
|
|
quit()
|
|
|
|
if not options.report_dir:
|
|
parser.error('No report directory set.')
|
|
|
|
# Get the directory where source code files are located.
|
|
cwd = os.getcwd()
|
|
source_dir = os.getcwd()
|
|
if options.source_dir:
|
|
source_dir = options.source_dir
|
|
|
|
add_author_information = []
|
|
if options.add_author_information:
|
|
fields = [x.strip() for x in options.add_author_information.split(',')]
|
|
for x in fields:
|
|
if x.lower() in ['n', 'name']:
|
|
add_author_information.append('name')
|
|
elif x.lower() in ['e', 'email']:
|
|
add_author_information.append('email')
|
|
elif x.lower() in ['d', 'date']:
|
|
add_author_information.append('date')
|
|
else:
|
|
print('Unrecognized value "%s" for author information, using default (name, email, date)' % x)
|
|
add_author_information = ['name', 'email', 'date']
|
|
break
|
|
|
|
blame_options = ''
|
|
if options.blame_options:
|
|
blame_options = options.blame_options
|
|
add_author_information = add_author_information or ['name', 'email', 'date']
|
|
|
|
# Parse the xml from all files defined in file argument
|
|
# or from stdin. If no input is provided, stdin is used
|
|
# Produce a simple list of errors.
|
|
print('Parsing xml report.')
|
|
try:
|
|
contentHandler = CppCheckHandler()
|
|
for fname in options.file or [sys.stdin]:
|
|
xml_parse(fname, contentHandler)
|
|
except (XmlParseException, ValueError) as msg:
|
|
print('Failed to parse cppcheck xml file: %s' % msg)
|
|
sys.exit(1)
|
|
|
|
# We have a list of errors. But now we want to group them on
|
|
# each source code file. Lets create a files dictionary that
|
|
# will contain a list of all the errors in that file. For each
|
|
# file we will also generate a HTML filename to use.
|
|
files = {}
|
|
file_no = 0
|
|
for error in contentHandler.errors:
|
|
filename = error['file']
|
|
if filename not in files:
|
|
files[filename] = {
|
|
'errors': [], 'htmlfile': str(file_no) + '.html'}
|
|
file_no = file_no + 1
|
|
files[filename]['errors'].append(error)
|
|
|
|
# Make sure that the report directory is created if it doesn't exist.
|
|
print('Creating %s directory' % options.report_dir)
|
|
if not os.path.exists(options.report_dir):
|
|
os.makedirs(options.report_dir)
|
|
|
|
# Generate a HTML file with syntax highlighted source code for each
|
|
# file that contains one or more errors.
|
|
print('Processing errors')
|
|
|
|
decode_errors = []
|
|
for filename, data in sorted(files.items()):
|
|
htmlfile = data['htmlfile']
|
|
errors = []
|
|
|
|
for error in data['errors']:
|
|
for location in error['locations']:
|
|
if filename == location['file']:
|
|
newError = dict(error)
|
|
|
|
del newError['locations']
|
|
newError['line'] = location['line']
|
|
if location.get('info'):
|
|
newError['msg'] = location['info']
|
|
newError['severity'] = 'information'
|
|
del newError['verbose']
|
|
|
|
errors.append(newError)
|
|
|
|
lines = []
|
|
for error in errors:
|
|
lines.append(error['line'])
|
|
|
|
if filename == '':
|
|
continue
|
|
|
|
source_filename = os.path.join(source_dir, filename)
|
|
try:
|
|
with io.open(source_filename, 'r', encoding=options.source_encoding) as input_file:
|
|
content = input_file.read()
|
|
except IOError:
|
|
if error['id'] == 'unmatchedSuppression':
|
|
continue # file not found, bail out
|
|
else:
|
|
sys.stderr.write("ERROR: Source file '%s' not found.\n" %
|
|
source_filename)
|
|
continue
|
|
except UnicodeDecodeError:
|
|
sys.stderr.write("WARNING: Unicode decode error in '%s'.\n" %
|
|
source_filename)
|
|
decode_errors.append(source_filename[2:]) # "[2:]" gets rid of "./" at beginning
|
|
continue
|
|
|
|
htmlFormatter = AnnotateCodeFormatter(linenos=True,
|
|
style='colorful',
|
|
hl_lines=lines,
|
|
lineanchors='line',
|
|
encoding=options.source_encoding)
|
|
htmlFormatter.errors = errors
|
|
|
|
with io.open(os.path.join(options.report_dir, htmlfile), 'w', encoding='utf-8') as output_file:
|
|
output_file.write(HTML_HEAD %
|
|
(options.title,
|
|
htmlFormatter.get_style_defs('.highlight'),
|
|
options.title,
|
|
': ' + filename))
|
|
output_file.write(HTML_HEAD_END)
|
|
|
|
output_file.write(HTML_MENU % (filename.split('/')[-1]))
|
|
for error in sorted(errors, key=lambda k: k['line']):
|
|
output_file.write("<a href=\"%s#line-%d\"> %s %s</a>" % (data['htmlfile'], error['line'], error['id'], error['line']))
|
|
output_file.write(HTML_MENU_END)
|
|
|
|
try:
|
|
lexer = guess_lexer_for_filename(source_filename, '', stripnl=False)
|
|
except ClassNotFound:
|
|
try:
|
|
lexer = guess_lexer(content, stripnl=False)
|
|
except ClassNotFound:
|
|
sys.stderr.write("ERROR: Couldn't determine lexer for the file' " + source_filename + " '. Won't be able to syntax highlight this file.")
|
|
output_file.write("\n <tr><td colspan=\"5\"> Could not generate content because pygments failed to determine the code type.</td></tr>")
|
|
output_file.write("\n <tr><td colspan=\"5\"> Sorry about this.</td></tr>")
|
|
continue
|
|
|
|
if options.source_encoding:
|
|
lexer.encoding = options.source_encoding
|
|
|
|
output_file.write(
|
|
highlight(content, lexer, htmlFormatter).decode(
|
|
options.source_encoding))
|
|
|
|
output_file.write(HTML_FOOTER % contentHandler.versionCppcheck)
|
|
|
|
print(' ' + filename)
|
|
|
|
# Generate a master index.html file that will contain a list of
|
|
# all the errors created.
|
|
print('Creating index.html')
|
|
|
|
with io.open(os.path.join(options.report_dir, 'index.html'),
|
|
'w') as output_file:
|
|
|
|
stats_count = 0
|
|
stats = []
|
|
filter_enabled = {}
|
|
for filename, data in sorted(files.items()):
|
|
for error in data['errors']:
|
|
stats.append(error['id']) # get the stats
|
|
filter_enabled[error['severity']] = True
|
|
filter_enabled['clang-tidy' if error['id'].startswith('clang-tidy-') else 'cppcheck'] = True
|
|
stats_count += 1
|
|
|
|
counter = Counter(stats)
|
|
|
|
stat_html = []
|
|
# the following lines sort the stat primary by value (occurrences),
|
|
# but if two IDs occur equally often, then we sort them alphabetically by warning ID
|
|
try:
|
|
cnt_max = counter.most_common()[0][1]
|
|
except IndexError:
|
|
cnt_max = 0
|
|
|
|
try:
|
|
cnt_min = counter.most_common()[-1][1]
|
|
except IndexError:
|
|
cnt_min = 0
|
|
|
|
stat_fmt = "\n <tr><td><input type=\"checkbox\" class=\"idToggle\" onclick=\"toggleDisplay(this)\" id=\"{}\" name=\"{}\" checked></td><td>{}</td><td>{}</td></tr>"
|
|
for occurrences in reversed(range(cnt_min, cnt_max + 1)):
|
|
for _id in [k for k, v in sorted(counter.items()) if v == occurrences]:
|
|
stat_html.append(stat_fmt.format(_id, _id, dict(counter.most_common())[_id], _id))
|
|
|
|
output_file.write(HTML_HEAD % (options.title, '', options.title, ''))
|
|
output_file.write(filter_bar(filter_enabled))
|
|
output_file.write(HTML_HEAD_END)
|
|
output_file.write(HTML_MENU.replace('id="menu"', 'id="menu_index"', 1).replace("Defects:", "Defect summary", 1) % (''))
|
|
output_file.write('\n <label><input type="checkbox" class=\"idToggle\" onclick="toggleAll()" checked> Toggle all</label>')
|
|
output_file.write('\n <table>')
|
|
output_file.write('\n <tr><th>Show</th><th>#</th><th>Defect ID</th></tr>')
|
|
output_file.write(''.join(stat_html))
|
|
output_file.write('\n <tr><td></td><td>' + str(stats_count) + '</td><td>total</td></tr>')
|
|
output_file.write('\n </table>')
|
|
output_file.write('\n <p><a href="stats.html">Statistics</a></p>')
|
|
output_file.write(HTML_MENU_END.replace("content", "content_index", 1))
|
|
|
|
output_file.write('\n <table class=\"summaryTable\">')
|
|
output_file.write(
|
|
'\n %s' %
|
|
tr_str('th', 'Line', 'Id', 'CWE', 'Severity', 'Message', 'Author', 'Author mail', 'Date (DD/MM/YYYY)', add_author=add_author_information))
|
|
|
|
for filename, data in sorted(files.items()):
|
|
file_error = filename in decode_errors or filename.endswith('*')
|
|
is_file = filename != '' and not file_error
|
|
row_content = filename if file_error else "<a href=\"%s\">%s</a>" % (data['htmlfile'], filename)
|
|
htmlfile = data.get('htmlfile') if is_file else None
|
|
|
|
output_file.write("\n <tbody class=\"fileEntry\">")
|
|
output_file.write("\n <tr><td colspan=\"5\">%s</td></tr>" % row_content)
|
|
|
|
if filename in decode_errors:
|
|
output_file.write("\n <tr><td colspan=\"5\">Could not generated due to UnicodeDecodeError</td></tr>")
|
|
|
|
sorted_errors = sorted(data['errors'], key=lambda k: k['line'])
|
|
blame_data = git_blame(sorted_errors, source_dir, filename, blame_options) if add_author_information else []
|
|
for error in sorted_errors:
|
|
git_blame_dict = blame_lookup(blame_data, error['line'])
|
|
message_class = None
|
|
try:
|
|
if error['inconclusive'] == 'true':
|
|
message_class = 'inconclusive'
|
|
error['severity'] += ", inconcl."
|
|
except KeyError:
|
|
pass
|
|
|
|
try:
|
|
if error['cwe']:
|
|
cwe_url = "<a href=\"https://cwe.mitre.org/data/definitions/" + error['cwe'] + ".html\">" + error['cwe'] + "</a>"
|
|
except KeyError:
|
|
cwe_url = ""
|
|
|
|
if error['severity'] in ['error', 'warning']:
|
|
message_class = error['severity']
|
|
|
|
line = error["line"] if is_file else ""
|
|
|
|
output_file.write(
|
|
'\n %s' %
|
|
tr_str('td', line, error["id"], cwe_url, error["severity"], error["msg"],
|
|
git_blame_dict.get('author', 'Unknown'), git_blame_dict.get('author-mail', '---'),
|
|
git_blame_dict.get('author-time', '---'),
|
|
tr_class=error["id"] + ' sev_' + error["severity"] + ' issue',
|
|
message_class=message_class,
|
|
add_author=add_author_information,
|
|
htmlfile=htmlfile))
|
|
output_file.write("\n </tbody>")
|
|
output_file.write('\n </table>')
|
|
output_file.write(HTML_FOOTER % contentHandler.versionCppcheck)
|
|
|
|
if decode_errors:
|
|
sys.stderr.write("\nGenerating html failed for the following files: " + ' '.join(decode_errors))
|
|
sys.stderr.write("\nConsider changing source-encoding (for example: \"htmlreport ... --source-encoding=\"iso8859-1\"\"\n")
|
|
|
|
print('Creating style.css file')
|
|
os.chdir(cwd) # going back to the cwd to find style.css
|
|
with io.open(os.path.join(options.report_dir, 'style.css'), 'w') as css_file:
|
|
css_file.write(STYLE_FILE)
|
|
|
|
print("Creating stats.html (statistics)\n")
|
|
stats_countlist = {}
|
|
|
|
for filename, data in sorted(files.items()):
|
|
if filename == '':
|
|
continue
|
|
stats_tmplist = []
|
|
for error in sorted(data['errors'], key=lambda k: k['line']):
|
|
stats_tmplist.append(error['severity'])
|
|
|
|
stats_countlist[filename] = dict(Counter(stats_tmplist))
|
|
|
|
# get top ten for each severity
|
|
SEVERITIES = "error", "warning", "portability", "performance", "style", "unusedFunction", "information", "missingInclude", "internal"
|
|
|
|
with io.open(os.path.join(options.report_dir, 'stats.html'), 'w') as stats_file:
|
|
|
|
stats_file.write(HTML_HEAD % (options.title, '', options.title, ': Statistics'))
|
|
stats_file.write(HTML_HEAD_END)
|
|
|
|
stats_file.write(HTML_MENU.replace('id="menu"', 'id="menu_index"', 1).replace("Defects:", "Back to summary", 1) % (''))
|
|
stats_file.write(HTML_MENU_END.replace("content", "content_index", 1))
|
|
|
|
|
|
for sev in SEVERITIES:
|
|
_sum = 0
|
|
stats_templist = {}
|
|
|
|
# if the we have an style warning but we are checking for
|
|
# portability, we have to skip it to prevent KeyError
|
|
try:
|
|
for filename in stats_countlist:
|
|
try: # also bail out if we have a file with no sev-results
|
|
_sum += stats_countlist[filename][sev]
|
|
stats_templist[filename] = int(stats_countlist[filename][sev]) # file : amount,
|
|
except KeyError:
|
|
continue
|
|
# don't print "0 style" etc, if no style warnings were found
|
|
if _sum == 0:
|
|
continue
|
|
except KeyError:
|
|
continue
|
|
stats_file.write("<p><span class=\"statHeader\">Top 10 files for " + sev + " severity, total findings: " + str(_sum) + "</span><br>\n")
|
|
|
|
# sort, so that the file with the most severities per type is first
|
|
stats_list_sorted = sorted(stats_templist.items(), key=operator.itemgetter(1, 0), reverse=True)
|
|
it = 0
|
|
LENGTH = 0
|
|
|
|
for i in stats_list_sorted: # printing loop
|
|
# for aesthetics: if it's the first iteration of the loop, get
|
|
# the max length of the number string
|
|
if it == 0:
|
|
LENGTH = len(str(i[1])) # <- length of longest number, now get the difference and try to make other numbers align to it
|
|
|
|
stats_file.write(" " * 3 + str(i[1]) + " " * (1 + LENGTH - len(str(i[1]))) + "<a href=\"" + files[i[0]]['htmlfile'] + "\"> " + i[0] + "</a><br>\n")
|
|
it += 1
|
|
if it == 10: # print only the top 10
|
|
break
|
|
stats_file.write("</p>\n")
|
|
|
|
stats_file.write(HTML_FOOTER % contentHandler.versionCppcheck)
|
|
|
|
print("\nOpen '" + options.report_dir + "/index.html' to see the results.")
|