diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 19:33:14 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 19:33:14 +0000 |
commit | 36d22d82aa202bb199967e9512281e9a53db42c9 (patch) | |
tree | 105e8c98ddea1c1e4784a60a5a6410fa416be2de /third_party/libwebrtc/build/android/emma_coverage_stats.py | |
parent | Initial commit. (diff) | |
download | firefox-esr-36d22d82aa202bb199967e9512281e9a53db42c9.tar.xz firefox-esr-36d22d82aa202bb199967e9512281e9a53db42c9.zip |
Adding upstream version 115.7.0esr.upstream/115.7.0esr
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'third_party/libwebrtc/build/android/emma_coverage_stats.py')
-rwxr-xr-x | third_party/libwebrtc/build/android/emma_coverage_stats.py | 483 |
1 files changed, 483 insertions, 0 deletions
diff --git a/third_party/libwebrtc/build/android/emma_coverage_stats.py b/third_party/libwebrtc/build/android/emma_coverage_stats.py new file mode 100755 index 0000000000..9ea8baa321 --- /dev/null +++ b/third_party/libwebrtc/build/android/emma_coverage_stats.py @@ -0,0 +1,483 @@ +#!/usr/bin/env vpython3 +# Copyright 2015 The Chromium Authors. All rights reserved. +# Use of this source code is governed by a BSD-style license that can be +# found in the LICENSE file. + +"""Generates incremental code coverage reports for Java code in Chromium. + +Usage: + + build/android/emma_coverage_stats.py -v --out <output file path> --emma-dir + <EMMA file directory> --lines-for-coverage-file + <path to file containing lines for coverage> + + Creates a JSON representation of the overall and file coverage stats and saves + this information to the specified output file. +""" + +import argparse +import collections +import json +import logging +import os +import re +import sys +from xml.etree import ElementTree + +import devil_chromium +from devil.utils import run_tests_helper + +NOT_EXECUTABLE = -1 +NOT_COVERED = 0 +COVERED = 1 +PARTIALLY_COVERED = 2 + +# Coverage information about a single line of code. +LineCoverage = collections.namedtuple( + 'LineCoverage', + ['lineno', 'source', 'covered_status', 'fractional_line_coverage']) + + +class _EmmaHtmlParser(object): + """Encapsulates HTML file parsing operations. + + This class contains all operations related to parsing HTML files that were + produced using the EMMA code coverage tool. + + Example HTML: + + Package links: + <a href="_files/1.html">org.chromium.chrome</a> + This is returned by the selector |XPATH_SELECT_PACKAGE_ELEMENTS|. + + Class links: + <a href="1e.html">DoActivity.java</a> + This is returned by the selector |XPATH_SELECT_CLASS_ELEMENTS|. + + Line coverage data: + <tr class="p"> + <td class="l" title="78% line coverage (7 out of 9)">108</td> + <td title="78% line coverage (7 out of 9 instructions)"> + if (index < 0 || index = mSelectors.size()) index = 0;</td> + </tr> + <tr> + <td class="l">109</td> + <td> </td> + </tr> + <tr class="c"> + <td class="l">110</td> + <td> if (mSelectors.get(index) != null) {</td> + </tr> + <tr class="z"> + <td class="l">111</td> + <td> for (int i = 0; i < mSelectors.size(); i++) {</td> + </tr> + Each <tr> element is returned by the selector |XPATH_SELECT_LOC|. + + We can parse this to get: + 1. Line number + 2. Line of source code + 3. Coverage status (c, z, or p) + 4. Fractional coverage value (% out of 100 if PARTIALLY_COVERED) + """ + # Selector to match all <a> elements within the rows that are in the table + # that displays all of the different packages. + _XPATH_SELECT_PACKAGE_ELEMENTS = './/BODY/TABLE[4]/TR/TD/A' + + # Selector to match all <a> elements within the rows that are in the table + # that displays all of the different classes within a package. + _XPATH_SELECT_CLASS_ELEMENTS = './/BODY/TABLE[3]/TR/TD/A' + + # Selector to match all <tr> elements within the table containing Java source + # code in an EMMA HTML file. + _XPATH_SELECT_LOC = './/BODY/TABLE[4]/TR' + + # Children of HTML elements are represented as a list in ElementTree. These + # constants represent list indices corresponding to relevant child elements. + + # Child 1 contains percentage covered for a line. + _ELEMENT_PERCENT_COVERED = 1 + + # Child 1 contains the original line of source code. + _ELEMENT_CONTAINING_SOURCE_CODE = 1 + + # Child 0 contains the line number. + _ELEMENT_CONTAINING_LINENO = 0 + + # Maps CSS class names to corresponding coverage constants. + _CSS_TO_STATUS = {'c': COVERED, 'p': PARTIALLY_COVERED, 'z': NOT_COVERED} + + # UTF-8 no break space. + _NO_BREAK_SPACE = '\xc2\xa0' + + def __init__(self, emma_file_base_dir): + """Initializes _EmmaHtmlParser. + + Args: + emma_file_base_dir: Path to the location where EMMA report files are + stored. Should be where index.html is stored. + """ + self._base_dir = emma_file_base_dir + self._emma_files_path = os.path.join(self._base_dir, '_files') + self._index_path = os.path.join(self._base_dir, 'index.html') + + def GetLineCoverage(self, emma_file_path): + """Returns a list of LineCoverage objects for the given EMMA HTML file. + + Args: + emma_file_path: String representing the path to the EMMA HTML file. + + Returns: + A list of LineCoverage objects. + """ + line_tr_elements = self._FindElements( + emma_file_path, self._XPATH_SELECT_LOC) + line_coverage = [] + for tr in line_tr_elements: + # Get the coverage status. + coverage_status = self._CSS_TO_STATUS.get(tr.get('CLASS'), NOT_EXECUTABLE) + # Get the fractional coverage value. + if coverage_status == PARTIALLY_COVERED: + title_attribute = (tr[self._ELEMENT_PERCENT_COVERED].get('TITLE')) + # Parse string that contains percent covered: "83% line coverage ...". + percent_covered = title_attribute.split('%')[0] + fractional_coverage = int(percent_covered) / 100.0 + else: + fractional_coverage = 1.0 + + # Get the line number. + lineno_element = tr[self._ELEMENT_CONTAINING_LINENO] + # Handles oddly formatted HTML (where there is an extra <a> tag). + lineno = int(lineno_element.text or + lineno_element[self._ELEMENT_CONTAINING_LINENO].text) + # Get the original line of Java source code. + raw_source = tr[self._ELEMENT_CONTAINING_SOURCE_CODE].text + source = raw_source.replace(self._NO_BREAK_SPACE, ' ') + + line = LineCoverage(lineno, source, coverage_status, fractional_coverage) + line_coverage.append(line) + + return line_coverage + + def GetPackageNameToEmmaFileDict(self): + """Returns a dict mapping Java packages to EMMA HTML coverage files. + + Parses the EMMA index.html file to get a list of packages, then parses each + package HTML file to get a list of classes for that package, and creates + a dict with this info. + + Returns: + A dict mapping string representation of Java packages (with class + names appended) to the corresponding file paths of EMMA HTML files. + """ + # These <a> elements contain each package name and the path of the file + # where all classes within said package are listed. + package_link_elements = self._FindElements( + self._index_path, self._XPATH_SELECT_PACKAGE_ELEMENTS) + # Maps file path of package directory (EMMA generated) to package name. + # Example: emma_dir/f.html: org.chromium.chrome. + package_links = { + os.path.join(self._base_dir, link.attrib['HREF']): link.text + for link in package_link_elements if 'HREF' in link.attrib + } + + package_to_emma = {} + for package_emma_file_path, package_name in package_links.items(): + # These <a> elements contain each class name in the current package and + # the path of the file where the coverage info is stored for each class. + coverage_file_link_elements = self._FindElements( + package_emma_file_path, self._XPATH_SELECT_CLASS_ELEMENTS) + + for class_name_element in coverage_file_link_elements: + emma_coverage_file_path = os.path.join( + self._emma_files_path, class_name_element.attrib['HREF']) + full_package_name = '%s.%s' % (package_name, class_name_element.text) + package_to_emma[full_package_name] = emma_coverage_file_path + + return package_to_emma + + # pylint: disable=no-self-use + def _FindElements(self, file_path, xpath_selector): + """Reads a HTML file and performs an XPath match. + + Args: + file_path: String representing the path to the HTML file. + xpath_selector: String representing xpath search pattern. + + Returns: + A list of ElementTree.Elements matching the given XPath selector. + Returns an empty list if there is no match. + """ + with open(file_path) as f: + file_contents = f.read() + root = ElementTree.fromstring(file_contents) + return root.findall(xpath_selector) + + +class _EmmaCoverageStats(object): + """Computes code coverage stats for Java code using the coverage tool EMMA. + + This class provides an API that allows users to capture absolute code coverage + and code coverage on a subset of lines for each Java source file. Coverage + reports are generated in JSON format. + """ + # Regular expression to get package name from Java package statement. + RE_PACKAGE_MATCH_GROUP = 'package' + RE_PACKAGE = re.compile(r'package (?P<%s>[\w.]*);' % RE_PACKAGE_MATCH_GROUP) + + def __init__(self, emma_file_base_dir, files_for_coverage): + """Initialize _EmmaCoverageStats. + + Args: + emma_file_base_dir: String representing the path to the base directory + where EMMA HTML coverage files are stored, i.e. parent of index.html. + files_for_coverage: A list of Java source code file paths to get EMMA + coverage for. + """ + self._emma_parser = _EmmaHtmlParser(emma_file_base_dir) + self._source_to_emma = self._GetSourceFileToEmmaFileDict(files_for_coverage) + + def GetCoverageDict(self, lines_for_coverage): + """Returns a dict containing detailed coverage information. + + Gets detailed coverage stats for each file specified in the + |lines_for_coverage| dict and the total incremental number of lines covered + and executable for all files in |lines_for_coverage|. + + Args: + lines_for_coverage: A dict mapping Java source file paths to lists of line + numbers. + + Returns: + A dict containing coverage stats for the given dict of files and lines. + Contains absolute coverage stats for each file, coverage stats for each + file's lines specified in |lines_for_coverage|, line by line coverage + for each file, and overall coverage stats for the lines specified in + |lines_for_coverage|. + """ + file_coverage = {} + for file_path, line_numbers in lines_for_coverage.items(): + file_coverage_dict = self.GetCoverageDictForFile(file_path, line_numbers) + if file_coverage_dict: + file_coverage[file_path] = file_coverage_dict + else: + logging.warning( + 'No code coverage data for %s, skipping.', file_path) + + covered_statuses = [s['incremental'] for s in file_coverage.values()] + num_covered_lines = sum(s['covered'] for s in covered_statuses) + num_total_lines = sum(s['total'] for s in covered_statuses) + return { + 'files': file_coverage, + 'patch': { + 'incremental': { + 'covered': num_covered_lines, + 'total': num_total_lines + } + } + } + + def GetCoverageDictForFile(self, file_path, line_numbers): + """Returns a dict containing detailed coverage info for the given file. + + Args: + file_path: The path to the Java source file that we want to create the + coverage dict for. + line_numbers: A list of integer line numbers to retrieve additional stats + for. + + Returns: + A dict containing absolute, incremental, and line by line coverage for + a file. + """ + if file_path not in self._source_to_emma: + return None + emma_file = self._source_to_emma[file_path] + total_line_coverage = self._emma_parser.GetLineCoverage(emma_file) + incremental_line_coverage = [line for line in total_line_coverage + if line.lineno in line_numbers] + line_by_line_coverage = [ + { + 'line': line.source, + 'coverage': line.covered_status, + 'changed': line.lineno in line_numbers, + 'fractional_coverage': line.fractional_line_coverage, + } + for line in total_line_coverage + ] + total_covered_lines, total_lines = ( + self.GetSummaryStatsForLines(total_line_coverage)) + incremental_covered_lines, incremental_total_lines = ( + self.GetSummaryStatsForLines(incremental_line_coverage)) + + file_coverage_stats = { + 'absolute': { + 'covered': total_covered_lines, + 'total': total_lines + }, + 'incremental': { + 'covered': incremental_covered_lines, + 'total': incremental_total_lines + }, + 'source': line_by_line_coverage, + } + return file_coverage_stats + + # pylint: disable=no-self-use + def GetSummaryStatsForLines(self, line_coverage): + """Gets summary stats for a given list of LineCoverage objects. + + Args: + line_coverage: A list of LineCoverage objects. + + Returns: + A tuple containing the number of lines that are covered and the total + number of lines that are executable, respectively + """ + partially_covered_sum = 0 + covered_status_totals = {COVERED: 0, NOT_COVERED: 0, PARTIALLY_COVERED: 0} + for line in line_coverage: + status = line.covered_status + if status == NOT_EXECUTABLE: + continue + covered_status_totals[status] += 1 + if status == PARTIALLY_COVERED: + partially_covered_sum += line.fractional_line_coverage + + total_covered = covered_status_totals[COVERED] + partially_covered_sum + total_lines = sum(covered_status_totals.values()) + return total_covered, total_lines + + def _GetSourceFileToEmmaFileDict(self, files): + """Gets a dict used to correlate Java source files with EMMA HTML files. + + This method gathers the information needed to correlate EMMA HTML + files with Java source files. EMMA XML and plain text reports do not provide + line by line coverage data, so HTML reports must be used instead. + Unfortunately, the HTML files that are created are given garbage names + (i.e 1.html) so we need to manually correlate EMMA HTML files + with the original Java source files. + + Args: + files: A list of file names for which coverage information is desired. + + Returns: + A dict mapping Java source file paths to EMMA HTML file paths. + """ + # Maps Java source file paths to package names. + # Example: /usr/code/file.java -> org.chromium.file.java. + source_to_package = {} + for file_path in files: + package = self.GetPackageNameFromFile(file_path) + if package: + source_to_package[file_path] = package + else: + logging.warning("Skipping %s because it doesn\'t have a package " + "statement.", file_path) + + # Maps package names to EMMA report HTML files. + # Example: org.chromium.file.java -> out/coverage/1a.html. + package_to_emma = self._emma_parser.GetPackageNameToEmmaFileDict() + # Finally, we have a dict mapping Java file paths to EMMA report files. + # Example: /usr/code/file.java -> out/coverage/1a.html. + source_to_emma = { + source: package_to_emma[package] + for source, package in source_to_package.items() + if package in package_to_emma + } + return source_to_emma + + @staticmethod + def NeedsCoverage(file_path): + """Checks to see if the file needs to be analyzed for code coverage. + + Args: + file_path: A string representing path to the file. + + Returns: + True for Java files that exist, False for all others. + """ + if os.path.splitext(file_path)[1] == '.java' and os.path.exists(file_path): + return True + else: + logging.info('Skipping file %s, cannot compute code coverage.', file_path) + return False + + @staticmethod + def GetPackageNameFromFile(file_path): + """Gets the full package name including the file name for a given file path. + + Args: + file_path: String representing the path to the Java source file. + + Returns: + A string representing the full package name with file name appended or + None if there is no package statement in the file. + """ + with open(file_path) as f: + file_content = f.read() + package_match = re.search(_EmmaCoverageStats.RE_PACKAGE, file_content) + if package_match: + package = package_match.group(_EmmaCoverageStats.RE_PACKAGE_MATCH_GROUP) + file_name = os.path.basename(file_path) + return '%s.%s' % (package, file_name) + else: + return None + + +def GenerateCoverageReport(line_coverage_file, out_file_path, coverage_dir): + """Generates a coverage report for a given set of lines. + + Writes the results of the coverage analysis to the file specified by + |out_file_path|. + + Args: + line_coverage_file: The path to a file which contains a dict mapping file + names to lists of line numbers. Example: {file1: [1, 2, 3], ...} means + that we should compute coverage information on lines 1 - 3 for file1. + out_file_path: A string representing the location to write the JSON report. + coverage_dir: A string representing the file path where the EMMA + HTML coverage files are located (i.e. folder where index.html is located). + """ + with open(line_coverage_file) as f: + potential_files_for_coverage = json.load(f) + + files_for_coverage = { + f: lines + for f, lines in potential_files_for_coverage.items() + if _EmmaCoverageStats.NeedsCoverage(f) + } + + coverage_results = {} + if files_for_coverage: + code_coverage = _EmmaCoverageStats(coverage_dir, + list(files_for_coverage.keys())) + coverage_results = code_coverage.GetCoverageDict(files_for_coverage) + else: + logging.info('No Java files requiring coverage were included in %s.', + line_coverage_file) + + with open(out_file_path, 'w+') as out_status_file: + json.dump(coverage_results, out_status_file) + + +def main(): + argparser = argparse.ArgumentParser() + argparser.add_argument('--out', required=True, type=str, + help='Report output file path.') + argparser.add_argument('--emma-dir', required=True, type=str, + help='EMMA HTML report directory.') + argparser.add_argument('--lines-for-coverage-file', required=True, type=str, + help='File containing a JSON object. Should contain a ' + 'dict mapping file names to lists of line numbers of ' + 'code for which coverage information is desired.') + argparser.add_argument('-v', '--verbose', action='count', + help='Print verbose log information.') + args = argparser.parse_args() + run_tests_helper.SetLogLevel(args.verbose) + devil_chromium.Initialize() + GenerateCoverageReport(args.lines_for_coverage_file, args.out, args.emma_dir) + + +if __name__ == '__main__': + sys.exit(main()) |