mirror of
https://github.com/Swiftgram/Telegram-iOS.git
synced 2025-08-17 19:09:56 +00:00
5800 lines
135 KiB
Perl
Executable File
5800 lines
135 KiB
Perl
Executable File
#!/usr/bin/perl -w
|
|
#
|
|
# Copyright (c) International Business Machines Corp., 2002,2012
|
|
#
|
|
# This program is free software; you can redistribute it and/or modify
|
|
# it under the terms of the GNU General Public License as published by
|
|
# the Free Software Foundation; either version 2 of the License, or (at
|
|
# your option) any later version.
|
|
#
|
|
# This program 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
|
|
# General Public License for more details.
|
|
#
|
|
# You should have received a copy of the GNU General Public License
|
|
# along with this program; if not, write to the Free Software
|
|
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
#
|
|
#
|
|
# genhtml
|
|
#
|
|
# This script generates HTML output from .info files as created by the
|
|
# geninfo script. Call it with --help and refer to the genhtml man page
|
|
# to get information on usage and available options.
|
|
#
|
|
#
|
|
# History:
|
|
# 2002-08-23 created by Peter Oberparleiter <Peter.Oberparleiter@de.ibm.com>
|
|
# IBM Lab Boeblingen
|
|
# based on code by Manoj Iyer <manjo@mail.utexas.edu> and
|
|
# Megan Bock <mbock@us.ibm.com>
|
|
# IBM Austin
|
|
# 2002-08-27 / Peter Oberparleiter: implemented frame view
|
|
# 2002-08-29 / Peter Oberparleiter: implemented test description filtering
|
|
# so that by default only descriptions for test cases which
|
|
# actually hit some source lines are kept
|
|
# 2002-09-05 / Peter Oberparleiter: implemented --no-sourceview
|
|
# 2002-09-05 / Mike Kobler: One of my source file paths includes a "+" in
|
|
# the directory name. I found that genhtml.pl died when it
|
|
# encountered it. I was able to fix the problem by modifying
|
|
# the string with the escape character before parsing it.
|
|
# 2002-10-26 / Peter Oberparleiter: implemented --num-spaces
|
|
# 2003-04-07 / Peter Oberparleiter: fixed bug which resulted in an error
|
|
# when trying to combine .info files containing data without
|
|
# a test name
|
|
# 2003-04-10 / Peter Oberparleiter: extended fix by Mike to also cover
|
|
# other special characters
|
|
# 2003-04-30 / Peter Oberparleiter: made info write to STDERR, not STDOUT
|
|
# 2003-07-10 / Peter Oberparleiter: added line checksum support
|
|
# 2004-08-09 / Peter Oberparleiter: added configuration file support
|
|
# 2005-03-04 / Cal Pierog: added legend to HTML output, fixed coloring of
|
|
# "good coverage" background
|
|
# 2006-03-18 / Marcus Boerger: added --custom-intro, --custom-outro and
|
|
# overwrite --no-prefix if --prefix is present
|
|
# 2006-03-20 / Peter Oberparleiter: changes to custom_* function (rename
|
|
# to html_prolog/_epilog, minor modifications to implementation),
|
|
# changed prefix/noprefix handling to be consistent with current
|
|
# logic
|
|
# 2006-03-20 / Peter Oberparleiter: added --html-extension option
|
|
# 2008-07-14 / Tom Zoerner: added --function-coverage command line option;
|
|
# added function table to source file page
|
|
# 2008-08-13 / Peter Oberparleiter: modified function coverage
|
|
# implementation (now enabled per default),
|
|
# introduced sorting option (enabled per default)
|
|
#
|
|
|
|
use strict;
|
|
use File::Basename;
|
|
use Getopt::Long;
|
|
use Digest::MD5 qw(md5_base64);
|
|
|
|
|
|
# Global constants
|
|
our $title = "LCOV - code coverage report";
|
|
our $lcov_version = 'LCOV version 1.10';
|
|
our $lcov_url = "http://ltp.sourceforge.net/coverage/lcov.php";
|
|
our $tool_name = basename($0);
|
|
|
|
# Specify coverage rate limits (in %) for classifying file entries
|
|
# HI: $hi_limit <= rate <= 100 graph color: green
|
|
# MED: $med_limit <= rate < $hi_limit graph color: orange
|
|
# LO: 0 <= rate < $med_limit graph color: red
|
|
|
|
# For line coverage/all coverage types if not specified
|
|
our $hi_limit = 90;
|
|
our $med_limit = 75;
|
|
|
|
# For function coverage
|
|
our $fn_hi_limit;
|
|
our $fn_med_limit;
|
|
|
|
# For branch coverage
|
|
our $br_hi_limit;
|
|
our $br_med_limit;
|
|
|
|
# Width of overview image
|
|
our $overview_width = 80;
|
|
|
|
# Resolution of overview navigation: this number specifies the maximum
|
|
# difference in lines between the position a user selected from the overview
|
|
# and the position the source code window is scrolled to.
|
|
our $nav_resolution = 4;
|
|
|
|
# Clicking a line in the overview image should show the source code view at
|
|
# a position a bit further up so that the requested line is not the first
|
|
# line in the window. This number specifies that offset in lines.
|
|
our $nav_offset = 10;
|
|
|
|
# Clicking on a function name should show the source code at a position a
|
|
# few lines before the first line of code of that function. This number
|
|
# specifies that offset in lines.
|
|
our $func_offset = 2;
|
|
|
|
our $overview_title = "top level";
|
|
|
|
# Width for line coverage information in the source code view
|
|
our $line_field_width = 12;
|
|
|
|
# Width for branch coverage information in the source code view
|
|
our $br_field_width = 16;
|
|
|
|
# Internal Constants
|
|
|
|
# Header types
|
|
our $HDR_DIR = 0;
|
|
our $HDR_FILE = 1;
|
|
our $HDR_SOURCE = 2;
|
|
our $HDR_TESTDESC = 3;
|
|
our $HDR_FUNC = 4;
|
|
|
|
# Sort types
|
|
our $SORT_FILE = 0;
|
|
our $SORT_LINE = 1;
|
|
our $SORT_FUNC = 2;
|
|
our $SORT_BRANCH = 3;
|
|
|
|
# Fileview heading types
|
|
our $HEAD_NO_DETAIL = 1;
|
|
our $HEAD_DETAIL_HIDDEN = 2;
|
|
our $HEAD_DETAIL_SHOWN = 3;
|
|
|
|
# Offsets for storing branch coverage data in vectors
|
|
our $BR_BLOCK = 0;
|
|
our $BR_BRANCH = 1;
|
|
our $BR_TAKEN = 2;
|
|
our $BR_VEC_ENTRIES = 3;
|
|
our $BR_VEC_WIDTH = 32;
|
|
|
|
# Additional offsets used when converting branch coverage data to HTML
|
|
our $BR_LEN = 3;
|
|
our $BR_OPEN = 4;
|
|
our $BR_CLOSE = 5;
|
|
|
|
# Branch data combination types
|
|
our $BR_SUB = 0;
|
|
our $BR_ADD = 1;
|
|
|
|
# Error classes which users may specify to ignore during processing
|
|
our $ERROR_SOURCE = 0;
|
|
our %ERROR_ID = (
|
|
"source" => $ERROR_SOURCE,
|
|
);
|
|
|
|
# Data related prototypes
|
|
sub print_usage(*);
|
|
sub gen_html();
|
|
sub html_create($$);
|
|
sub process_dir($);
|
|
sub process_file($$$);
|
|
sub info(@);
|
|
sub read_info_file($);
|
|
sub get_info_entry($);
|
|
sub set_info_entry($$$$$$$$$;$$$$$$);
|
|
sub get_prefix($@);
|
|
sub shorten_prefix($);
|
|
sub get_dir_list(@);
|
|
sub get_relative_base_path($);
|
|
sub read_testfile($);
|
|
sub get_date_string();
|
|
sub create_sub_dir($);
|
|
sub subtract_counts($$);
|
|
sub add_counts($$);
|
|
sub apply_baseline($$);
|
|
sub remove_unused_descriptions();
|
|
sub get_found_and_hit($);
|
|
sub get_affecting_tests($$$);
|
|
sub combine_info_files($$);
|
|
sub merge_checksums($$$);
|
|
sub combine_info_entries($$$);
|
|
sub apply_prefix($$);
|
|
sub system_no_output($@);
|
|
sub read_config($);
|
|
sub apply_config($);
|
|
sub get_html_prolog($);
|
|
sub get_html_epilog($);
|
|
sub write_dir_page($$$$$$$$$$$$$$$$$);
|
|
sub classify_rate($$$$);
|
|
sub br_taken_add($$);
|
|
sub br_taken_sub($$);
|
|
sub br_ivec_len($);
|
|
sub br_ivec_get($$);
|
|
sub br_ivec_push($$$$);
|
|
sub combine_brcount($$$);
|
|
sub get_br_found_and_hit($);
|
|
sub warn_handler($);
|
|
sub die_handler($);
|
|
sub parse_ignore_errors(@);
|
|
sub rate($$;$$$);
|
|
|
|
|
|
# HTML related prototypes
|
|
sub escape_html($);
|
|
sub get_bar_graph_code($$$);
|
|
|
|
sub write_png_files();
|
|
sub write_htaccess_file();
|
|
sub write_css_file();
|
|
sub write_description_file($$$$$$$);
|
|
sub write_function_table(*$$$$$$$$$$);
|
|
|
|
sub write_html(*$);
|
|
sub write_html_prolog(*$$);
|
|
sub write_html_epilog(*$;$);
|
|
|
|
sub write_header(*$$$$$$$$$$);
|
|
sub write_header_prolog(*$);
|
|
sub write_header_line(*@);
|
|
sub write_header_epilog(*$);
|
|
|
|
sub write_file_table(*$$$$$$$);
|
|
sub write_file_table_prolog(*$@);
|
|
sub write_file_table_entry(*$$$@);
|
|
sub write_file_table_detail_entry(*$@);
|
|
sub write_file_table_epilog(*);
|
|
|
|
sub write_test_table_prolog(*$);
|
|
sub write_test_table_entry(*$$);
|
|
sub write_test_table_epilog(*);
|
|
|
|
sub write_source($$$$$$$);
|
|
sub write_source_prolog(*);
|
|
sub write_source_line(*$$$$$$);
|
|
sub write_source_epilog(*);
|
|
|
|
sub write_frameset(*$$$);
|
|
sub write_overview_line(*$$$);
|
|
sub write_overview(*$$$$);
|
|
|
|
# External prototype (defined in genpng)
|
|
sub gen_png($$$@);
|
|
|
|
|
|
# Global variables & initialization
|
|
our %info_data; # Hash containing all data from .info file
|
|
our $dir_prefix; # Prefix to remove from all sub directories
|
|
our %test_description; # Hash containing test descriptions if available
|
|
our $date = get_date_string();
|
|
|
|
our @info_filenames; # List of .info files to use as data source
|
|
our $test_title; # Title for output as written to each page header
|
|
our $output_directory; # Name of directory in which to store output
|
|
our $base_filename; # Optional name of file containing baseline data
|
|
our $desc_filename; # Name of file containing test descriptions
|
|
our $css_filename; # Optional name of external stylesheet file to use
|
|
our $quiet; # If set, suppress information messages
|
|
our $help; # Help option flag
|
|
our $version; # Version option flag
|
|
our $show_details; # If set, generate detailed directory view
|
|
our $no_prefix; # If set, do not remove filename prefix
|
|
our $func_coverage; # If set, generate function coverage statistics
|
|
our $no_func_coverage; # Disable func_coverage
|
|
our $br_coverage; # If set, generate branch coverage statistics
|
|
our $no_br_coverage; # Disable br_coverage
|
|
our $sort = 1; # If set, provide directory listings with sorted entries
|
|
our $no_sort; # Disable sort
|
|
our $frames; # If set, use frames for source code view
|
|
our $keep_descriptions; # If set, do not remove unused test case descriptions
|
|
our $no_sourceview; # If set, do not create a source code view for each file
|
|
our $highlight; # If set, highlight lines covered by converted data only
|
|
our $legend; # If set, include legend in output
|
|
our $tab_size = 8; # Number of spaces to use in place of tab
|
|
our $config; # Configuration file contents
|
|
our $html_prolog_file; # Custom HTML prolog file (up to and including <body>)
|
|
our $html_epilog_file; # Custom HTML epilog file (from </body> onwards)
|
|
our $html_prolog; # Actual HTML prolog
|
|
our $html_epilog; # Actual HTML epilog
|
|
our $html_ext = "html"; # Extension for generated HTML files
|
|
our $html_gzip = 0; # Compress with gzip
|
|
our $demangle_cpp = 0; # Demangle C++ function names
|
|
our @opt_ignore_errors; # Ignore certain error classes during processing
|
|
our @ignore;
|
|
our $opt_config_file; # User-specified configuration file location
|
|
our %opt_rc;
|
|
our $charset = "UTF-8"; # Default charset for HTML pages
|
|
our @fileview_sortlist;
|
|
our @fileview_sortname = ("", "-sort-l", "-sort-f", "-sort-b");
|
|
our @funcview_sortlist;
|
|
our @rate_name = ("Lo", "Med", "Hi");
|
|
our @rate_png = ("ruby.png", "amber.png", "emerald.png");
|
|
our $lcov_func_coverage = 1;
|
|
our $lcov_branch_coverage = 0;
|
|
|
|
our $cwd = `pwd`; # Current working directory
|
|
chomp($cwd);
|
|
our $tool_dir = dirname($0); # Directory where genhtml tool is installed
|
|
|
|
|
|
#
|
|
# Code entry point
|
|
#
|
|
|
|
$SIG{__WARN__} = \&warn_handler;
|
|
$SIG{__DIE__} = \&die_handler;
|
|
|
|
# Prettify version string
|
|
$lcov_version =~ s/\$\s*Revision\s*:?\s*(\S+)\s*\$/$1/;
|
|
|
|
# Add current working directory if $tool_dir is not already an absolute path
|
|
if (! ($tool_dir =~ /^\/(.*)$/))
|
|
{
|
|
$tool_dir = "$cwd/$tool_dir";
|
|
}
|
|
|
|
# Check command line for a configuration file name
|
|
Getopt::Long::Configure("pass_through", "no_auto_abbrev");
|
|
GetOptions("config-file=s" => \$opt_config_file,
|
|
"rc=s%" => \%opt_rc);
|
|
Getopt::Long::Configure("default");
|
|
|
|
# Read configuration file if available
|
|
if (defined($opt_config_file)) {
|
|
$config = read_config($opt_config_file);
|
|
} elsif (defined($ENV{"HOME"}) && (-r $ENV{"HOME"}."/.lcovrc"))
|
|
{
|
|
$config = read_config($ENV{"HOME"}."/.lcovrc");
|
|
}
|
|
elsif (-r "/etc/lcovrc")
|
|
{
|
|
$config = read_config("/etc/lcovrc");
|
|
}
|
|
|
|
if ($config || %opt_rc)
|
|
{
|
|
# Copy configuration file and --rc values to variables
|
|
apply_config({
|
|
"genhtml_css_file" => \$css_filename,
|
|
"genhtml_hi_limit" => \$hi_limit,
|
|
"genhtml_med_limit" => \$med_limit,
|
|
"genhtml_line_field_width" => \$line_field_width,
|
|
"genhtml_overview_width" => \$overview_width,
|
|
"genhtml_nav_resolution" => \$nav_resolution,
|
|
"genhtml_nav_offset" => \$nav_offset,
|
|
"genhtml_keep_descriptions" => \$keep_descriptions,
|
|
"genhtml_no_prefix" => \$no_prefix,
|
|
"genhtml_no_source" => \$no_sourceview,
|
|
"genhtml_num_spaces" => \$tab_size,
|
|
"genhtml_highlight" => \$highlight,
|
|
"genhtml_legend" => \$legend,
|
|
"genhtml_html_prolog" => \$html_prolog_file,
|
|
"genhtml_html_epilog" => \$html_epilog_file,
|
|
"genhtml_html_extension" => \$html_ext,
|
|
"genhtml_html_gzip" => \$html_gzip,
|
|
"genhtml_function_hi_limit" => \$fn_hi_limit,
|
|
"genhtml_function_med_limit" => \$fn_med_limit,
|
|
"genhtml_function_coverage" => \$func_coverage,
|
|
"genhtml_branch_hi_limit" => \$br_hi_limit,
|
|
"genhtml_branch_med_limit" => \$br_med_limit,
|
|
"genhtml_branch_coverage" => \$br_coverage,
|
|
"genhtml_branch_field_width" => \$br_field_width,
|
|
"genhtml_sort" => \$sort,
|
|
"genhtml_charset" => \$charset,
|
|
"lcov_function_coverage" => \$lcov_func_coverage,
|
|
"lcov_branch_coverage" => \$lcov_branch_coverage,
|
|
});
|
|
}
|
|
|
|
# Copy related values if not specified
|
|
$fn_hi_limit = $hi_limit if (!defined($fn_hi_limit));
|
|
$fn_med_limit = $med_limit if (!defined($fn_med_limit));
|
|
$br_hi_limit = $hi_limit if (!defined($br_hi_limit));
|
|
$br_med_limit = $med_limit if (!defined($br_med_limit));
|
|
$func_coverage = $lcov_func_coverage if (!defined($func_coverage));
|
|
$br_coverage = $lcov_branch_coverage if (!defined($br_coverage));
|
|
|
|
# Parse command line options
|
|
if (!GetOptions("output-directory|o=s" => \$output_directory,
|
|
"title|t=s" => \$test_title,
|
|
"description-file|d=s" => \$desc_filename,
|
|
"keep-descriptions|k" => \$keep_descriptions,
|
|
"css-file|c=s" => \$css_filename,
|
|
"baseline-file|b=s" => \$base_filename,
|
|
"prefix|p=s" => \$dir_prefix,
|
|
"num-spaces=i" => \$tab_size,
|
|
"no-prefix" => \$no_prefix,
|
|
"no-sourceview" => \$no_sourceview,
|
|
"show-details|s" => \$show_details,
|
|
"frames|f" => \$frames,
|
|
"highlight" => \$highlight,
|
|
"legend" => \$legend,
|
|
"quiet|q" => \$quiet,
|
|
"help|h|?" => \$help,
|
|
"version|v" => \$version,
|
|
"html-prolog=s" => \$html_prolog_file,
|
|
"html-epilog=s" => \$html_epilog_file,
|
|
"html-extension=s" => \$html_ext,
|
|
"html-gzip" => \$html_gzip,
|
|
"function-coverage" => \$func_coverage,
|
|
"no-function-coverage" => \$no_func_coverage,
|
|
"branch-coverage" => \$br_coverage,
|
|
"no-branch-coverage" => \$no_br_coverage,
|
|
"sort" => \$sort,
|
|
"no-sort" => \$no_sort,
|
|
"demangle-cpp" => \$demangle_cpp,
|
|
"ignore-errors=s" => \@opt_ignore_errors,
|
|
"config-file=s" => \$opt_config_file,
|
|
"rc=s%" => \%opt_rc,
|
|
))
|
|
{
|
|
print(STDERR "Use $tool_name --help to get usage information\n");
|
|
exit(1);
|
|
} else {
|
|
# Merge options
|
|
if ($no_func_coverage) {
|
|
$func_coverage = 0;
|
|
}
|
|
if ($no_br_coverage) {
|
|
$br_coverage = 0;
|
|
}
|
|
|
|
# Merge sort options
|
|
if ($no_sort) {
|
|
$sort = 0;
|
|
}
|
|
}
|
|
|
|
@info_filenames = @ARGV;
|
|
|
|
# Check for help option
|
|
if ($help)
|
|
{
|
|
print_usage(*STDOUT);
|
|
exit(0);
|
|
}
|
|
|
|
# Check for version option
|
|
if ($version)
|
|
{
|
|
print("$tool_name: $lcov_version\n");
|
|
exit(0);
|
|
}
|
|
|
|
# Determine which errors the user wants us to ignore
|
|
parse_ignore_errors(@opt_ignore_errors);
|
|
|
|
# Check for info filename
|
|
if (!@info_filenames)
|
|
{
|
|
die("No filename specified\n".
|
|
"Use $tool_name --help to get usage information\n");
|
|
}
|
|
|
|
# Generate a title if none is specified
|
|
if (!$test_title)
|
|
{
|
|
if (scalar(@info_filenames) == 1)
|
|
{
|
|
# Only one filename specified, use it as title
|
|
$test_title = basename($info_filenames[0]);
|
|
}
|
|
else
|
|
{
|
|
# More than one filename specified, used default title
|
|
$test_title = "unnamed";
|
|
}
|
|
}
|
|
|
|
# Make sure css_filename is an absolute path (in case we're changing
|
|
# directories)
|
|
if ($css_filename)
|
|
{
|
|
if (!($css_filename =~ /^\/(.*)$/))
|
|
{
|
|
$css_filename = $cwd."/".$css_filename;
|
|
}
|
|
}
|
|
|
|
# Make sure tab_size is within valid range
|
|
if ($tab_size < 1)
|
|
{
|
|
print(STDERR "ERROR: invalid number of spaces specified: ".
|
|
"$tab_size!\n");
|
|
exit(1);
|
|
}
|
|
|
|
# Get HTML prolog and epilog
|
|
$html_prolog = get_html_prolog($html_prolog_file);
|
|
$html_epilog = get_html_epilog($html_epilog_file);
|
|
|
|
# Issue a warning if --no-sourceview is enabled together with --frames
|
|
if ($no_sourceview && defined($frames))
|
|
{
|
|
warn("WARNING: option --frames disabled because --no-sourceview ".
|
|
"was specified!\n");
|
|
$frames = undef;
|
|
}
|
|
|
|
# Issue a warning if --no-prefix is enabled together with --prefix
|
|
if ($no_prefix && defined($dir_prefix))
|
|
{
|
|
warn("WARNING: option --prefix disabled because --no-prefix was ".
|
|
"specified!\n");
|
|
$dir_prefix = undef;
|
|
}
|
|
|
|
@fileview_sortlist = ($SORT_FILE);
|
|
@funcview_sortlist = ($SORT_FILE);
|
|
|
|
if ($sort) {
|
|
push(@fileview_sortlist, $SORT_LINE);
|
|
push(@fileview_sortlist, $SORT_FUNC) if ($func_coverage);
|
|
push(@fileview_sortlist, $SORT_BRANCH) if ($br_coverage);
|
|
push(@funcview_sortlist, $SORT_LINE);
|
|
}
|
|
|
|
if ($frames)
|
|
{
|
|
# Include genpng code needed for overview image generation
|
|
do("$tool_dir/genpng");
|
|
}
|
|
|
|
# Ensure that the c++filt tool is available when using --demangle-cpp
|
|
if ($demangle_cpp)
|
|
{
|
|
if (system_no_output(3, "c++filt", "--version")) {
|
|
die("ERROR: could not find c++filt tool needed for ".
|
|
"--demangle-cpp\n");
|
|
}
|
|
}
|
|
|
|
# Make sure output_directory exists, create it if necessary
|
|
if ($output_directory)
|
|
{
|
|
stat($output_directory);
|
|
|
|
if (! -e _)
|
|
{
|
|
create_sub_dir($output_directory);
|
|
}
|
|
}
|
|
|
|
# Do something
|
|
gen_html();
|
|
|
|
exit(0);
|
|
|
|
|
|
|
|
#
|
|
# print_usage(handle)
|
|
#
|
|
# Print usage information.
|
|
#
|
|
|
|
sub print_usage(*)
|
|
{
|
|
local *HANDLE = $_[0];
|
|
|
|
print(HANDLE <<END_OF_USAGE);
|
|
Usage: $tool_name [OPTIONS] INFOFILE(S)
|
|
|
|
Create HTML output for coverage data found in INFOFILE. Note that INFOFILE
|
|
may also be a list of filenames.
|
|
|
|
Misc:
|
|
-h, --help Print this help, then exit
|
|
-v, --version Print version number, then exit
|
|
-q, --quiet Do not print progress messages
|
|
--config-file FILENAME Specify configuration file location
|
|
--rc SETTING=VALUE Override configuration file setting
|
|
--ignore-errors ERRORS Continue after ERRORS (source)
|
|
|
|
Operation:
|
|
-o, --output-directory OUTDIR Write HTML output to OUTDIR
|
|
-s, --show-details Generate detailed directory view
|
|
-d, --description-file DESCFILE Read test case descriptions from DESCFILE
|
|
-k, --keep-descriptions Do not remove unused test descriptions
|
|
-b, --baseline-file BASEFILE Use BASEFILE as baseline file
|
|
-p, --prefix PREFIX Remove PREFIX from all directory names
|
|
--no-prefix Do not remove prefix from directory names
|
|
--(no-)function-coverage Enable (disable) function coverage display
|
|
--(no-)branch-coverage Enable (disable) branch coverage display
|
|
|
|
HTML output:
|
|
-f, --frames Use HTML frames for source code view
|
|
-t, --title TITLE Display TITLE in header of all pages
|
|
-c, --css-file CSSFILE Use external style sheet file CSSFILE
|
|
--no-source Do not create source code view
|
|
--num-spaces NUM Replace tabs with NUM spaces in source view
|
|
--highlight Highlight lines with converted-only data
|
|
--legend Include color legend in HTML output
|
|
--html-prolog FILE Use FILE as HTML prolog for generated pages
|
|
--html-epilog FILE Use FILE as HTML epilog for generated pages
|
|
--html-extension EXT Use EXT as filename extension for pages
|
|
--html-gzip Use gzip to compress HTML
|
|
--(no-)sort Enable (disable) sorted coverage views
|
|
--demangle-cpp Demangle C++ function names
|
|
|
|
For more information see: $lcov_url
|
|
END_OF_USAGE
|
|
;
|
|
}
|
|
|
|
|
|
#
|
|
# get_rate(found, hit)
|
|
#
|
|
# Return a relative value for the specified found&hit values
|
|
# which is used for sorting the corresponding entries in a
|
|
# file list.
|
|
#
|
|
|
|
sub get_rate($$)
|
|
{
|
|
my ($found, $hit) = @_;
|
|
|
|
if ($found == 0) {
|
|
return 10000;
|
|
}
|
|
return int($hit * 1000 / $found) * 10 + 2 - (1 / $found);
|
|
}
|
|
|
|
|
|
#
|
|
# get_overall_line(found, hit, name_singular, name_plural)
|
|
#
|
|
# Return a string containing overall information for the specified
|
|
# found/hit data.
|
|
#
|
|
|
|
sub get_overall_line($$$$)
|
|
{
|
|
my ($found, $hit, $name_sn, $name_pl) = @_;
|
|
my $name;
|
|
|
|
return "no data found" if (!defined($found) || $found == 0);
|
|
$name = ($found == 1) ? $name_sn : $name_pl;
|
|
return rate($hit, $found, "% ($hit of $found $name)");
|
|
}
|
|
|
|
|
|
#
|
|
# print_overall_rate(ln_do, ln_found, ln_hit, fn_do, fn_found, fn_hit, br_do
|
|
# br_found, br_hit)
|
|
#
|
|
# Print overall coverage rates for the specified coverage types.
|
|
#
|
|
|
|
sub print_overall_rate($$$$$$$$$)
|
|
{
|
|
my ($ln_do, $ln_found, $ln_hit, $fn_do, $fn_found, $fn_hit,
|
|
$br_do, $br_found, $br_hit) = @_;
|
|
|
|
info("Overall coverage rate:\n");
|
|
info(" lines......: %s\n",
|
|
get_overall_line($ln_found, $ln_hit, "line", "lines"))
|
|
if ($ln_do);
|
|
info(" functions..: %s\n",
|
|
get_overall_line($fn_found, $fn_hit, "function", "functions"))
|
|
if ($fn_do);
|
|
info(" branches...: %s\n",
|
|
get_overall_line($br_found, $br_hit, "branch", "branches"))
|
|
if ($br_do);
|
|
}
|
|
|
|
|
|
#
|
|
# gen_html()
|
|
#
|
|
# Generate a set of HTML pages from contents of .info file INFO_FILENAME.
|
|
# Files will be written to the current directory. If provided, test case
|
|
# descriptions will be read from .tests file TEST_FILENAME and included
|
|
# in ouput.
|
|
#
|
|
# Die on error.
|
|
#
|
|
|
|
sub gen_html()
|
|
{
|
|
local *HTML_HANDLE;
|
|
my %overview;
|
|
my %base_data;
|
|
my $lines_found;
|
|
my $lines_hit;
|
|
my $fn_found;
|
|
my $fn_hit;
|
|
my $br_found;
|
|
my $br_hit;
|
|
my $overall_found = 0;
|
|
my $overall_hit = 0;
|
|
my $total_fn_found = 0;
|
|
my $total_fn_hit = 0;
|
|
my $total_br_found = 0;
|
|
my $total_br_hit = 0;
|
|
my $dir_name;
|
|
my $link_name;
|
|
my @dir_list;
|
|
my %new_info;
|
|
|
|
# Read in all specified .info files
|
|
foreach (@info_filenames)
|
|
{
|
|
%new_info = %{read_info_file($_)};
|
|
|
|
# Combine %new_info with %info_data
|
|
%info_data = %{combine_info_files(\%info_data, \%new_info)};
|
|
}
|
|
|
|
info("Found %d entries.\n", scalar(keys(%info_data)));
|
|
|
|
# Read and apply baseline data if specified
|
|
if ($base_filename)
|
|
{
|
|
# Read baseline file
|
|
info("Reading baseline file $base_filename\n");
|
|
%base_data = %{read_info_file($base_filename)};
|
|
info("Found %d entries.\n", scalar(keys(%base_data)));
|
|
|
|
# Apply baseline
|
|
info("Subtracting baseline data.\n");
|
|
%info_data = %{apply_baseline(\%info_data, \%base_data)};
|
|
}
|
|
|
|
@dir_list = get_dir_list(keys(%info_data));
|
|
|
|
if ($no_prefix)
|
|
{
|
|
# User requested that we leave filenames alone
|
|
info("User asked not to remove filename prefix\n");
|
|
}
|
|
elsif (!defined($dir_prefix))
|
|
{
|
|
# Get prefix common to most directories in list
|
|
$dir_prefix = get_prefix(1, keys(%info_data));
|
|
|
|
if ($dir_prefix)
|
|
{
|
|
info("Found common filename prefix \"$dir_prefix\"\n");
|
|
}
|
|
else
|
|
{
|
|
info("No common filename prefix found!\n");
|
|
$no_prefix=1;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
info("Using user-specified filename prefix \"".
|
|
"$dir_prefix\"\n");
|
|
}
|
|
|
|
# Read in test description file if specified
|
|
if ($desc_filename)
|
|
{
|
|
info("Reading test description file $desc_filename\n");
|
|
%test_description = %{read_testfile($desc_filename)};
|
|
|
|
# Remove test descriptions which are not referenced
|
|
# from %info_data if user didn't tell us otherwise
|
|
if (!$keep_descriptions)
|
|
{
|
|
remove_unused_descriptions();
|
|
}
|
|
}
|
|
|
|
# Change to output directory if specified
|
|
if ($output_directory)
|
|
{
|
|
chdir($output_directory)
|
|
or die("ERROR: cannot change to directory ".
|
|
"$output_directory!\n");
|
|
}
|
|
|
|
info("Writing .css and .png files.\n");
|
|
write_css_file();
|
|
write_png_files();
|
|
|
|
if ($html_gzip)
|
|
{
|
|
info("Writing .htaccess file.\n");
|
|
write_htaccess_file();
|
|
}
|
|
|
|
info("Generating output.\n");
|
|
|
|
# Process each subdirectory and collect overview information
|
|
foreach $dir_name (@dir_list)
|
|
{
|
|
($lines_found, $lines_hit, $fn_found, $fn_hit,
|
|
$br_found, $br_hit)
|
|
= process_dir($dir_name);
|
|
|
|
# Handle files in root directory gracefully
|
|
$dir_name = "root" if ($dir_name eq "");
|
|
|
|
# Remove prefix if applicable
|
|
if (!$no_prefix && $dir_prefix)
|
|
{
|
|
# Match directory names beginning with $dir_prefix
|
|
$dir_name = apply_prefix($dir_name, $dir_prefix);
|
|
}
|
|
|
|
# Generate name for directory overview HTML page
|
|
if ($dir_name =~ /^\/(.*)$/)
|
|
{
|
|
$link_name = substr($dir_name, 1)."/index.$html_ext";
|
|
}
|
|
else
|
|
{
|
|
$link_name = $dir_name."/index.$html_ext";
|
|
}
|
|
|
|
$overview{$dir_name} = [$lines_found, $lines_hit, $fn_found,
|
|
$fn_hit, $br_found, $br_hit, $link_name,
|
|
get_rate($lines_found, $lines_hit),
|
|
get_rate($fn_found, $fn_hit),
|
|
get_rate($br_found, $br_hit)];
|
|
$overall_found += $lines_found;
|
|
$overall_hit += $lines_hit;
|
|
$total_fn_found += $fn_found;
|
|
$total_fn_hit += $fn_hit;
|
|
$total_br_found += $br_found;
|
|
$total_br_hit += $br_hit;
|
|
}
|
|
|
|
# Generate overview page
|
|
info("Writing directory view page.\n");
|
|
|
|
# Create sorted pages
|
|
foreach (@fileview_sortlist) {
|
|
write_dir_page($fileview_sortname[$_], ".", "", $test_title,
|
|
undef, $overall_found, $overall_hit,
|
|
$total_fn_found, $total_fn_hit, $total_br_found,
|
|
$total_br_hit, \%overview, {}, {}, {}, 0, $_);
|
|
}
|
|
|
|
# Check if there are any test case descriptions to write out
|
|
if (%test_description)
|
|
{
|
|
info("Writing test case description file.\n");
|
|
write_description_file( \%test_description,
|
|
$overall_found, $overall_hit,
|
|
$total_fn_found, $total_fn_hit,
|
|
$total_br_found, $total_br_hit);
|
|
}
|
|
|
|
print_overall_rate(1, $overall_found, $overall_hit,
|
|
$func_coverage, $total_fn_found, $total_fn_hit,
|
|
$br_coverage, $total_br_found, $total_br_hit);
|
|
|
|
chdir($cwd);
|
|
}
|
|
|
|
#
|
|
# html_create(handle, filename)
|
|
#
|
|
|
|
sub html_create($$)
|
|
{
|
|
my $handle = $_[0];
|
|
my $filename = $_[1];
|
|
|
|
if ($html_gzip)
|
|
{
|
|
open($handle, "|-", "gzip -c >'$filename'")
|
|
or die("ERROR: cannot open $filename for writing ".
|
|
"(gzip)!\n");
|
|
}
|
|
else
|
|
{
|
|
open($handle, ">", $filename)
|
|
or die("ERROR: cannot open $filename for writing!\n");
|
|
}
|
|
}
|
|
|
|
sub write_dir_page($$$$$$$$$$$$$$$$$)
|
|
{
|
|
my ($name, $rel_dir, $base_dir, $title, $trunc_dir, $overall_found,
|
|
$overall_hit, $total_fn_found, $total_fn_hit, $total_br_found,
|
|
$total_br_hit, $overview, $testhash, $testfnchash, $testbrhash,
|
|
$view_type, $sort_type) = @_;
|
|
|
|
# Generate directory overview page including details
|
|
html_create(*HTML_HANDLE, "$rel_dir/index$name.$html_ext");
|
|
if (!defined($trunc_dir)) {
|
|
$trunc_dir = "";
|
|
}
|
|
$title .= " - " if ($trunc_dir ne "");
|
|
write_html_prolog(*HTML_HANDLE, $base_dir, "LCOV - $title$trunc_dir");
|
|
write_header(*HTML_HANDLE, $view_type, $trunc_dir, $rel_dir,
|
|
$overall_found, $overall_hit, $total_fn_found,
|
|
$total_fn_hit, $total_br_found, $total_br_hit, $sort_type);
|
|
write_file_table(*HTML_HANDLE, $base_dir, $overview, $testhash,
|
|
$testfnchash, $testbrhash, $view_type, $sort_type);
|
|
write_html_epilog(*HTML_HANDLE, $base_dir);
|
|
close(*HTML_HANDLE);
|
|
}
|
|
|
|
|
|
#
|
|
# process_dir(dir_name)
|
|
#
|
|
|
|
sub process_dir($)
|
|
{
|
|
my $abs_dir = $_[0];
|
|
my $trunc_dir;
|
|
my $rel_dir = $abs_dir;
|
|
my $base_dir;
|
|
my $filename;
|
|
my %overview;
|
|
my $lines_found;
|
|
my $lines_hit;
|
|
my $fn_found;
|
|
my $fn_hit;
|
|
my $br_found;
|
|
my $br_hit;
|
|
my $overall_found=0;
|
|
my $overall_hit=0;
|
|
my $total_fn_found=0;
|
|
my $total_fn_hit=0;
|
|
my $total_br_found = 0;
|
|
my $total_br_hit = 0;
|
|
my $base_name;
|
|
my $extension;
|
|
my $testdata;
|
|
my %testhash;
|
|
my $testfncdata;
|
|
my %testfnchash;
|
|
my $testbrdata;
|
|
my %testbrhash;
|
|
my @sort_list;
|
|
local *HTML_HANDLE;
|
|
|
|
# Remove prefix if applicable
|
|
if (!$no_prefix)
|
|
{
|
|
# Match directory name beginning with $dir_prefix
|
|
$rel_dir = apply_prefix($rel_dir, $dir_prefix);
|
|
}
|
|
|
|
$trunc_dir = $rel_dir;
|
|
|
|
# Remove leading /
|
|
if ($rel_dir =~ /^\/(.*)$/)
|
|
{
|
|
$rel_dir = substr($rel_dir, 1);
|
|
}
|
|
|
|
# Handle files in root directory gracefully
|
|
$rel_dir = "root" if ($rel_dir eq "");
|
|
$trunc_dir = "root" if ($trunc_dir eq "");
|
|
|
|
$base_dir = get_relative_base_path($rel_dir);
|
|
|
|
create_sub_dir($rel_dir);
|
|
|
|
# Match filenames which specify files in this directory, not including
|
|
# sub-directories
|
|
foreach $filename (grep(/^\Q$abs_dir\E\/[^\/]*$/,keys(%info_data)))
|
|
{
|
|
my $page_link;
|
|
my $func_link;
|
|
|
|
($lines_found, $lines_hit, $fn_found, $fn_hit, $br_found,
|
|
$br_hit, $testdata, $testfncdata, $testbrdata) =
|
|
process_file($trunc_dir, $rel_dir, $filename);
|
|
|
|
$base_name = basename($filename);
|
|
|
|
if ($no_sourceview) {
|
|
$page_link = "";
|
|
} elsif ($frames) {
|
|
# Link to frameset page
|
|
$page_link = "$base_name.gcov.frameset.$html_ext";
|
|
} else {
|
|
# Link directory to source code view page
|
|
$page_link = "$base_name.gcov.$html_ext";
|
|
}
|
|
$overview{$base_name} = [$lines_found, $lines_hit, $fn_found,
|
|
$fn_hit, $br_found, $br_hit,
|
|
$page_link,
|
|
get_rate($lines_found, $lines_hit),
|
|
get_rate($fn_found, $fn_hit),
|
|
get_rate($br_found, $br_hit)];
|
|
|
|
$testhash{$base_name} = $testdata;
|
|
$testfnchash{$base_name} = $testfncdata;
|
|
$testbrhash{$base_name} = $testbrdata;
|
|
|
|
$overall_found += $lines_found;
|
|
$overall_hit += $lines_hit;
|
|
|
|
$total_fn_found += $fn_found;
|
|
$total_fn_hit += $fn_hit;
|
|
|
|
$total_br_found += $br_found;
|
|
$total_br_hit += $br_hit;
|
|
}
|
|
|
|
# Create sorted pages
|
|
foreach (@fileview_sortlist) {
|
|
# Generate directory overview page (without details)
|
|
write_dir_page($fileview_sortname[$_], $rel_dir, $base_dir,
|
|
$test_title, $trunc_dir, $overall_found,
|
|
$overall_hit, $total_fn_found, $total_fn_hit,
|
|
$total_br_found, $total_br_hit, \%overview, {},
|
|
{}, {}, 1, $_);
|
|
if (!$show_details) {
|
|
next;
|
|
}
|
|
# Generate directory overview page including details
|
|
write_dir_page("-detail".$fileview_sortname[$_], $rel_dir,
|
|
$base_dir, $test_title, $trunc_dir,
|
|
$overall_found, $overall_hit, $total_fn_found,
|
|
$total_fn_hit, $total_br_found, $total_br_hit,
|
|
\%overview, \%testhash, \%testfnchash,
|
|
\%testbrhash, 1, $_);
|
|
}
|
|
|
|
# Calculate resulting line counts
|
|
return ($overall_found, $overall_hit, $total_fn_found, $total_fn_hit,
|
|
$total_br_found, $total_br_hit);
|
|
}
|
|
|
|
|
|
#
|
|
# get_converted_lines(testdata)
|
|
#
|
|
# Return hash of line numbers of those lines which were only covered in
|
|
# converted data sets.
|
|
#
|
|
|
|
sub get_converted_lines($)
|
|
{
|
|
my $testdata = $_[0];
|
|
my $testcount;
|
|
my %converted;
|
|
my %nonconverted;
|
|
my $hash;
|
|
my $testcase;
|
|
my $line;
|
|
my %result;
|
|
|
|
|
|
# Get a hash containing line numbers with positive counts both for
|
|
# converted and original data sets
|
|
foreach $testcase (keys(%{$testdata}))
|
|
{
|
|
# Check to see if this is a converted data set
|
|
if ($testcase =~ /,diff$/)
|
|
{
|
|
$hash = \%converted;
|
|
}
|
|
else
|
|
{
|
|
$hash = \%nonconverted;
|
|
}
|
|
|
|
$testcount = $testdata->{$testcase};
|
|
# Add lines with a positive count to hash
|
|
foreach $line (keys%{$testcount})
|
|
{
|
|
if ($testcount->{$line} > 0)
|
|
{
|
|
$hash->{$line} = 1;
|
|
}
|
|
}
|
|
}
|
|
|
|
# Combine both hashes to resulting list
|
|
foreach $line (keys(%converted))
|
|
{
|
|
if (!defined($nonconverted{$line}))
|
|
{
|
|
$result{$line} = 1;
|
|
}
|
|
}
|
|
|
|
return \%result;
|
|
}
|
|
|
|
|
|
sub write_function_page($$$$$$$$$$$$$$$$$$)
|
|
{
|
|
my ($base_dir, $rel_dir, $trunc_dir, $base_name, $title,
|
|
$lines_found, $lines_hit, $fn_found, $fn_hit, $br_found, $br_hit,
|
|
$sumcount, $funcdata, $sumfnccount, $testfncdata, $sumbrcount,
|
|
$testbrdata, $sort_type) = @_;
|
|
my $pagetitle;
|
|
my $filename;
|
|
|
|
# Generate function table for this file
|
|
if ($sort_type == 0) {
|
|
$filename = "$rel_dir/$base_name.func.$html_ext";
|
|
} else {
|
|
$filename = "$rel_dir/$base_name.func-sort-c.$html_ext";
|
|
}
|
|
html_create(*HTML_HANDLE, $filename);
|
|
$pagetitle = "LCOV - $title - $trunc_dir/$base_name - functions";
|
|
write_html_prolog(*HTML_HANDLE, $base_dir, $pagetitle);
|
|
write_header(*HTML_HANDLE, 4, "$trunc_dir/$base_name",
|
|
"$rel_dir/$base_name", $lines_found, $lines_hit,
|
|
$fn_found, $fn_hit, $br_found, $br_hit, $sort_type);
|
|
write_function_table(*HTML_HANDLE, "$base_name.gcov.$html_ext",
|
|
$sumcount, $funcdata,
|
|
$sumfnccount, $testfncdata, $sumbrcount,
|
|
$testbrdata, $base_name,
|
|
$base_dir, $sort_type);
|
|
write_html_epilog(*HTML_HANDLE, $base_dir, 1);
|
|
close(*HTML_HANDLE);
|
|
}
|
|
|
|
|
|
#
|
|
# process_file(trunc_dir, rel_dir, filename)
|
|
#
|
|
|
|
sub process_file($$$)
|
|
{
|
|
info("Processing file ".apply_prefix($_[2], $dir_prefix)."\n");
|
|
|
|
my $trunc_dir = $_[0];
|
|
my $rel_dir = $_[1];
|
|
my $filename = $_[2];
|
|
my $base_name = basename($filename);
|
|
my $base_dir = get_relative_base_path($rel_dir);
|
|
my $testdata;
|
|
my $testcount;
|
|
my $sumcount;
|
|
my $funcdata;
|
|
my $checkdata;
|
|
my $testfncdata;
|
|
my $sumfnccount;
|
|
my $testbrdata;
|
|
my $sumbrcount;
|
|
my $lines_found;
|
|
my $lines_hit;
|
|
my $fn_found;
|
|
my $fn_hit;
|
|
my $br_found;
|
|
my $br_hit;
|
|
my $converted;
|
|
my @source;
|
|
my $pagetitle;
|
|
local *HTML_HANDLE;
|
|
|
|
($testdata, $sumcount, $funcdata, $checkdata, $testfncdata,
|
|
$sumfnccount, $testbrdata, $sumbrcount, $lines_found, $lines_hit,
|
|
$fn_found, $fn_hit, $br_found, $br_hit)
|
|
= get_info_entry($info_data{$filename});
|
|
|
|
# Return after this point in case user asked us not to generate
|
|
# source code view
|
|
if ($no_sourceview)
|
|
{
|
|
return ($lines_found, $lines_hit, $fn_found, $fn_hit,
|
|
$br_found, $br_hit, $testdata, $testfncdata,
|
|
$testbrdata);
|
|
}
|
|
|
|
$converted = get_converted_lines($testdata);
|
|
# Generate source code view for this file
|
|
html_create(*HTML_HANDLE, "$rel_dir/$base_name.gcov.$html_ext");
|
|
$pagetitle = "LCOV - $test_title - $trunc_dir/$base_name";
|
|
write_html_prolog(*HTML_HANDLE, $base_dir, $pagetitle);
|
|
write_header(*HTML_HANDLE, 2, "$trunc_dir/$base_name",
|
|
"$rel_dir/$base_name", $lines_found, $lines_hit,
|
|
$fn_found, $fn_hit, $br_found, $br_hit, 0);
|
|
@source = write_source(*HTML_HANDLE, $filename, $sumcount, $checkdata,
|
|
$converted, $funcdata, $sumbrcount);
|
|
|
|
write_html_epilog(*HTML_HANDLE, $base_dir, 1);
|
|
close(*HTML_HANDLE);
|
|
|
|
if ($func_coverage) {
|
|
# Create function tables
|
|
foreach (@funcview_sortlist) {
|
|
write_function_page($base_dir, $rel_dir, $trunc_dir,
|
|
$base_name, $test_title,
|
|
$lines_found, $lines_hit,
|
|
$fn_found, $fn_hit, $br_found,
|
|
$br_hit, $sumcount,
|
|
$funcdata, $sumfnccount,
|
|
$testfncdata, $sumbrcount,
|
|
$testbrdata, $_);
|
|
}
|
|
}
|
|
|
|
# Additional files are needed in case of frame output
|
|
if (!$frames)
|
|
{
|
|
return ($lines_found, $lines_hit, $fn_found, $fn_hit,
|
|
$br_found, $br_hit, $testdata, $testfncdata,
|
|
$testbrdata);
|
|
}
|
|
|
|
# Create overview png file
|
|
gen_png("$rel_dir/$base_name.gcov.png", $overview_width, $tab_size,
|
|
@source);
|
|
|
|
# Create frameset page
|
|
html_create(*HTML_HANDLE,
|
|
"$rel_dir/$base_name.gcov.frameset.$html_ext");
|
|
write_frameset(*HTML_HANDLE, $base_dir, $base_name, $pagetitle);
|
|
close(*HTML_HANDLE);
|
|
|
|
# Write overview frame
|
|
html_create(*HTML_HANDLE,
|
|
"$rel_dir/$base_name.gcov.overview.$html_ext");
|
|
write_overview(*HTML_HANDLE, $base_dir, $base_name, $pagetitle,
|
|
scalar(@source));
|
|
close(*HTML_HANDLE);
|
|
|
|
return ($lines_found, $lines_hit, $fn_found, $fn_hit, $br_found,
|
|
$br_hit, $testdata, $testfncdata, $testbrdata);
|
|
}
|
|
|
|
|
|
#
|
|
# read_info_file(info_filename)
|
|
#
|
|
# Read in the contents of the .info file specified by INFO_FILENAME. Data will
|
|
# be returned as a reference to a hash containing the following mappings:
|
|
#
|
|
# %result: for each filename found in file -> \%data
|
|
#
|
|
# %data: "test" -> \%testdata
|
|
# "sum" -> \%sumcount
|
|
# "func" -> \%funcdata
|
|
# "found" -> $lines_found (number of instrumented lines found in file)
|
|
# "hit" -> $lines_hit (number of executed lines in file)
|
|
# "check" -> \%checkdata
|
|
# "testfnc" -> \%testfncdata
|
|
# "sumfnc" -> \%sumfnccount
|
|
# "testbr" -> \%testbrdata
|
|
# "sumbr" -> \%sumbrcount
|
|
#
|
|
# %testdata : name of test affecting this file -> \%testcount
|
|
# %testfncdata: name of test affecting this file -> \%testfnccount
|
|
# %testbrdata: name of test affecting this file -> \%testbrcount
|
|
#
|
|
# %testcount : line number -> execution count for a single test
|
|
# %testfnccount: function name -> execution count for a single test
|
|
# %testbrcount : line number -> branch coverage data for a single test
|
|
# %sumcount : line number -> execution count for all tests
|
|
# %sumfnccount : function name -> execution count for all tests
|
|
# %sumbrcount : line number -> branch coverage data for all tests
|
|
# %funcdata : function name -> line number
|
|
# %checkdata : line number -> checksum of source code line
|
|
# $brdata : vector of items: block, branch, taken
|
|
#
|
|
# Note that .info file sections referring to the same file and test name
|
|
# will automatically be combined by adding all execution counts.
|
|
#
|
|
# Note that if INFO_FILENAME ends with ".gz", it is assumed that the file
|
|
# is compressed using GZIP. If available, GUNZIP will be used to decompress
|
|
# this file.
|
|
#
|
|
# Die on error.
|
|
#
|
|
|
|
sub read_info_file($)
|
|
{
|
|
my $tracefile = $_[0]; # Name of tracefile
|
|
my %result; # Resulting hash: file -> data
|
|
my $data; # Data handle for current entry
|
|
my $testdata; # " "
|
|
my $testcount; # " "
|
|
my $sumcount; # " "
|
|
my $funcdata; # " "
|
|
my $checkdata; # " "
|
|
my $testfncdata;
|
|
my $testfnccount;
|
|
my $sumfnccount;
|
|
my $testbrdata;
|
|
my $testbrcount;
|
|
my $sumbrcount;
|
|
my $line; # Current line read from .info file
|
|
my $testname; # Current test name
|
|
my $filename; # Current filename
|
|
my $hitcount; # Count for lines hit
|
|
my $count; # Execution count of current line
|
|
my $negative; # If set, warn about negative counts
|
|
my $changed_testname; # If set, warn about changed testname
|
|
my $line_checksum; # Checksum of current line
|
|
my $br_found;
|
|
my $br_hit;
|
|
local *INFO_HANDLE; # Filehandle for .info file
|
|
|
|
info("Reading data file $tracefile\n");
|
|
|
|
# Check if file exists and is readable
|
|
stat($_[0]);
|
|
if (!(-r _))
|
|
{
|
|
die("ERROR: cannot read file $_[0]!\n");
|
|
}
|
|
|
|
# Check if this is really a plain file
|
|
if (!(-f _))
|
|
{
|
|
die("ERROR: not a plain file: $_[0]!\n");
|
|
}
|
|
|
|
# Check for .gz extension
|
|
if ($_[0] =~ /\.gz$/)
|
|
{
|
|
# Check for availability of GZIP tool
|
|
system_no_output(1, "gunzip" ,"-h")
|
|
and die("ERROR: gunzip command not available!\n");
|
|
|
|
# Check integrity of compressed file
|
|
system_no_output(1, "gunzip", "-t", $_[0])
|
|
and die("ERROR: integrity check failed for ".
|
|
"compressed file $_[0]!\n");
|
|
|
|
# Open compressed file
|
|
open(INFO_HANDLE, "-|", "gunzip -c '$_[0]'")
|
|
or die("ERROR: cannot start gunzip to decompress ".
|
|
"file $_[0]!\n");
|
|
}
|
|
else
|
|
{
|
|
# Open decompressed file
|
|
open(INFO_HANDLE, "<", $_[0])
|
|
or die("ERROR: cannot read file $_[0]!\n");
|
|
}
|
|
|
|
$testname = "";
|
|
while (<INFO_HANDLE>)
|
|
{
|
|
chomp($_);
|
|
$line = $_;
|
|
|
|
# Switch statement
|
|
foreach ($line)
|
|
{
|
|
/^TN:([^,]*)(,diff)?/ && do
|
|
{
|
|
# Test name information found
|
|
$testname = defined($1) ? $1 : "";
|
|
if ($testname =~ s/\W/_/g)
|
|
{
|
|
$changed_testname = 1;
|
|
}
|
|
$testname .= $2 if (defined($2));
|
|
last;
|
|
};
|
|
|
|
/^[SK]F:(.*)/ && do
|
|
{
|
|
# Filename information found
|
|
# Retrieve data for new entry
|
|
$filename = $1;
|
|
|
|
$data = $result{$filename};
|
|
($testdata, $sumcount, $funcdata, $checkdata,
|
|
$testfncdata, $sumfnccount, $testbrdata,
|
|
$sumbrcount) =
|
|
get_info_entry($data);
|
|
|
|
if (defined($testname))
|
|
{
|
|
$testcount = $testdata->{$testname};
|
|
$testfnccount = $testfncdata->{$testname};
|
|
$testbrcount = $testbrdata->{$testname};
|
|
}
|
|
else
|
|
{
|
|
$testcount = {};
|
|
$testfnccount = {};
|
|
$testbrcount = {};
|
|
}
|
|
last;
|
|
};
|
|
|
|
/^DA:(\d+),(-?\d+)(,[^,\s]+)?/ && do
|
|
{
|
|
# Fix negative counts
|
|
$count = $2 < 0 ? 0 : $2;
|
|
if ($2 < 0)
|
|
{
|
|
$negative = 1;
|
|
}
|
|
# Execution count found, add to structure
|
|
# Add summary counts
|
|
$sumcount->{$1} += $count;
|
|
|
|
# Add test-specific counts
|
|
if (defined($testname))
|
|
{
|
|
$testcount->{$1} += $count;
|
|
}
|
|
|
|
# Store line checksum if available
|
|
if (defined($3))
|
|
{
|
|
$line_checksum = substr($3, 1);
|
|
|
|
# Does it match a previous definition
|
|
if (defined($checkdata->{$1}) &&
|
|
($checkdata->{$1} ne
|
|
$line_checksum))
|
|
{
|
|
die("ERROR: checksum mismatch ".
|
|
"at $filename:$1\n");
|
|
}
|
|
|
|
$checkdata->{$1} = $line_checksum;
|
|
}
|
|
last;
|
|
};
|
|
|
|
/^FN:(\d+),([^,]+)/ && do
|
|
{
|
|
last if (!$func_coverage);
|
|
|
|
# Function data found, add to structure
|
|
$funcdata->{$2} = $1;
|
|
|
|
# Also initialize function call data
|
|
if (!defined($sumfnccount->{$2})) {
|
|
$sumfnccount->{$2} = 0;
|
|
}
|
|
if (defined($testname))
|
|
{
|
|
if (!defined($testfnccount->{$2})) {
|
|
$testfnccount->{$2} = 0;
|
|
}
|
|
}
|
|
last;
|
|
};
|
|
|
|
/^FNDA:(\d+),([^,]+)/ && do
|
|
{
|
|
last if (!$func_coverage);
|
|
# Function call count found, add to structure
|
|
# Add summary counts
|
|
$sumfnccount->{$2} += $1;
|
|
|
|
# Add test-specific counts
|
|
if (defined($testname))
|
|
{
|
|
$testfnccount->{$2} += $1;
|
|
}
|
|
last;
|
|
};
|
|
|
|
/^BRDA:(\d+),(\d+),(\d+),(\d+|-)/ && do {
|
|
# Branch coverage data found
|
|
my ($line, $block, $branch, $taken) =
|
|
($1, $2, $3, $4);
|
|
|
|
last if (!$br_coverage);
|
|
$sumbrcount->{$line} =
|
|
br_ivec_push($sumbrcount->{$line},
|
|
$block, $branch, $taken);
|
|
|
|
# Add test-specific counts
|
|
if (defined($testname)) {
|
|
$testbrcount->{$line} =
|
|
br_ivec_push(
|
|
$testbrcount->{$line},
|
|
$block, $branch,
|
|
$taken);
|
|
}
|
|
last;
|
|
};
|
|
|
|
/^end_of_record/ && do
|
|
{
|
|
# Found end of section marker
|
|
if ($filename)
|
|
{
|
|
# Store current section data
|
|
if (defined($testname))
|
|
{
|
|
$testdata->{$testname} =
|
|
$testcount;
|
|
$testfncdata->{$testname} =
|
|
$testfnccount;
|
|
$testbrdata->{$testname} =
|
|
$testbrcount;
|
|
}
|
|
|
|
set_info_entry($data, $testdata,
|
|
$sumcount, $funcdata,
|
|
$checkdata, $testfncdata,
|
|
$sumfnccount,
|
|
$testbrdata,
|
|
$sumbrcount);
|
|
$result{$filename} = $data;
|
|
last;
|
|
}
|
|
};
|
|
|
|
# default
|
|
last;
|
|
}
|
|
}
|
|
close(INFO_HANDLE);
|
|
|
|
# Calculate lines_found and lines_hit for each file
|
|
foreach $filename (keys(%result))
|
|
{
|
|
$data = $result{$filename};
|
|
|
|
($testdata, $sumcount, undef, undef, $testfncdata,
|
|
$sumfnccount, $testbrdata, $sumbrcount) =
|
|
get_info_entry($data);
|
|
|
|
# Filter out empty files
|
|
if (scalar(keys(%{$sumcount})) == 0)
|
|
{
|
|
delete($result{$filename});
|
|
next;
|
|
}
|
|
# Filter out empty test cases
|
|
foreach $testname (keys(%{$testdata}))
|
|
{
|
|
if (!defined($testdata->{$testname}) ||
|
|
scalar(keys(%{$testdata->{$testname}})) == 0)
|
|
{
|
|
delete($testdata->{$testname});
|
|
delete($testfncdata->{$testname});
|
|
}
|
|
}
|
|
|
|
$data->{"found"} = scalar(keys(%{$sumcount}));
|
|
$hitcount = 0;
|
|
|
|
foreach (keys(%{$sumcount}))
|
|
{
|
|
if ($sumcount->{$_} > 0) { $hitcount++; }
|
|
}
|
|
|
|
$data->{"hit"} = $hitcount;
|
|
|
|
# Get found/hit values for function call data
|
|
$data->{"f_found"} = scalar(keys(%{$sumfnccount}));
|
|
$hitcount = 0;
|
|
|
|
foreach (keys(%{$sumfnccount})) {
|
|
if ($sumfnccount->{$_} > 0) {
|
|
$hitcount++;
|
|
}
|
|
}
|
|
$data->{"f_hit"} = $hitcount;
|
|
|
|
# Get found/hit values for branch data
|
|
($br_found, $br_hit) = get_br_found_and_hit($sumbrcount);
|
|
|
|
$data->{"b_found"} = $br_found;
|
|
$data->{"b_hit"} = $br_hit;
|
|
}
|
|
|
|
if (scalar(keys(%result)) == 0)
|
|
{
|
|
die("ERROR: no valid records found in tracefile $tracefile\n");
|
|
}
|
|
if ($negative)
|
|
{
|
|
warn("WARNING: negative counts found in tracefile ".
|
|
"$tracefile\n");
|
|
}
|
|
if ($changed_testname)
|
|
{
|
|
warn("WARNING: invalid characters removed from testname in ".
|
|
"tracefile $tracefile\n");
|
|
}
|
|
|
|
return(\%result);
|
|
}
|
|
|
|
|
|
#
|
|
# get_info_entry(hash_ref)
|
|
#
|
|
# Retrieve data from an entry of the structure generated by read_info_file().
|
|
# Return a list of references to hashes:
|
|
# (test data hash ref, sum count hash ref, funcdata hash ref, checkdata hash
|
|
# ref, testfncdata hash ref, sumfnccount hash ref, lines found, lines hit,
|
|
# functions found, functions hit)
|
|
#
|
|
|
|
sub get_info_entry($)
|
|
{
|
|
my $testdata_ref = $_[0]->{"test"};
|
|
my $sumcount_ref = $_[0]->{"sum"};
|
|
my $funcdata_ref = $_[0]->{"func"};
|
|
my $checkdata_ref = $_[0]->{"check"};
|
|
my $testfncdata = $_[0]->{"testfnc"};
|
|
my $sumfnccount = $_[0]->{"sumfnc"};
|
|
my $testbrdata = $_[0]->{"testbr"};
|
|
my $sumbrcount = $_[0]->{"sumbr"};
|
|
my $lines_found = $_[0]->{"found"};
|
|
my $lines_hit = $_[0]->{"hit"};
|
|
my $fn_found = $_[0]->{"f_found"};
|
|
my $fn_hit = $_[0]->{"f_hit"};
|
|
my $br_found = $_[0]->{"b_found"};
|
|
my $br_hit = $_[0]->{"b_hit"};
|
|
|
|
return ($testdata_ref, $sumcount_ref, $funcdata_ref, $checkdata_ref,
|
|
$testfncdata, $sumfnccount, $testbrdata, $sumbrcount,
|
|
$lines_found, $lines_hit, $fn_found, $fn_hit,
|
|
$br_found, $br_hit);
|
|
}
|
|
|
|
|
|
#
|
|
# set_info_entry(hash_ref, testdata_ref, sumcount_ref, funcdata_ref,
|
|
# checkdata_ref, testfncdata_ref, sumfcncount_ref,
|
|
# testbrdata_ref, sumbrcount_ref[,lines_found,
|
|
# lines_hit, f_found, f_hit, $b_found, $b_hit])
|
|
#
|
|
# Update the hash referenced by HASH_REF with the provided data references.
|
|
#
|
|
|
|
sub set_info_entry($$$$$$$$$;$$$$$$)
|
|
{
|
|
my $data_ref = $_[0];
|
|
|
|
$data_ref->{"test"} = $_[1];
|
|
$data_ref->{"sum"} = $_[2];
|
|
$data_ref->{"func"} = $_[3];
|
|
$data_ref->{"check"} = $_[4];
|
|
$data_ref->{"testfnc"} = $_[5];
|
|
$data_ref->{"sumfnc"} = $_[6];
|
|
$data_ref->{"testbr"} = $_[7];
|
|
$data_ref->{"sumbr"} = $_[8];
|
|
|
|
if (defined($_[9])) { $data_ref->{"found"} = $_[9]; }
|
|
if (defined($_[10])) { $data_ref->{"hit"} = $_[10]; }
|
|
if (defined($_[11])) { $data_ref->{"f_found"} = $_[11]; }
|
|
if (defined($_[12])) { $data_ref->{"f_hit"} = $_[12]; }
|
|
if (defined($_[13])) { $data_ref->{"b_found"} = $_[13]; }
|
|
if (defined($_[14])) { $data_ref->{"b_hit"} = $_[14]; }
|
|
}
|
|
|
|
|
|
#
|
|
# add_counts(data1_ref, data2_ref)
|
|
#
|
|
# DATA1_REF and DATA2_REF are references to hashes containing a mapping
|
|
#
|
|
# line number -> execution count
|
|
#
|
|
# Return a list (RESULT_REF, LINES_FOUND, LINES_HIT) where RESULT_REF
|
|
# is a reference to a hash containing the combined mapping in which
|
|
# execution counts are added.
|
|
#
|
|
|
|
sub add_counts($$)
|
|
{
|
|
my %data1 = %{$_[0]}; # Hash 1
|
|
my %data2 = %{$_[1]}; # Hash 2
|
|
my %result; # Resulting hash
|
|
my $line; # Current line iteration scalar
|
|
my $data1_count; # Count of line in hash1
|
|
my $data2_count; # Count of line in hash2
|
|
my $found = 0; # Total number of lines found
|
|
my $hit = 0; # Number of lines with a count > 0
|
|
|
|
foreach $line (keys(%data1))
|
|
{
|
|
$data1_count = $data1{$line};
|
|
$data2_count = $data2{$line};
|
|
|
|
# Add counts if present in both hashes
|
|
if (defined($data2_count)) { $data1_count += $data2_count; }
|
|
|
|
# Store sum in %result
|
|
$result{$line} = $data1_count;
|
|
|
|
$found++;
|
|
if ($data1_count > 0) { $hit++; }
|
|
}
|
|
|
|
# Add lines unique to data2
|
|
foreach $line (keys(%data2))
|
|
{
|
|
# Skip lines already in data1
|
|
if (defined($data1{$line})) { next; }
|
|
|
|
# Copy count from data2
|
|
$result{$line} = $data2{$line};
|
|
|
|
$found++;
|
|
if ($result{$line} > 0) { $hit++; }
|
|
}
|
|
|
|
return (\%result, $found, $hit);
|
|
}
|
|
|
|
|
|
#
|
|
# merge_checksums(ref1, ref2, filename)
|
|
#
|
|
# REF1 and REF2 are references to hashes containing a mapping
|
|
#
|
|
# line number -> checksum
|
|
#
|
|
# Merge checksum lists defined in REF1 and REF2 and return reference to
|
|
# resulting hash. Die if a checksum for a line is defined in both hashes
|
|
# but does not match.
|
|
#
|
|
|
|
sub merge_checksums($$$)
|
|
{
|
|
my $ref1 = $_[0];
|
|
my $ref2 = $_[1];
|
|
my $filename = $_[2];
|
|
my %result;
|
|
my $line;
|
|
|
|
foreach $line (keys(%{$ref1}))
|
|
{
|
|
if (defined($ref2->{$line}) &&
|
|
($ref1->{$line} ne $ref2->{$line}))
|
|
{
|
|
die("ERROR: checksum mismatch at $filename:$line\n");
|
|
}
|
|
$result{$line} = $ref1->{$line};
|
|
}
|
|
|
|
foreach $line (keys(%{$ref2}))
|
|
{
|
|
$result{$line} = $ref2->{$line};
|
|
}
|
|
|
|
return \%result;
|
|
}
|
|
|
|
|
|
#
|
|
# merge_func_data(funcdata1, funcdata2, filename)
|
|
#
|
|
|
|
sub merge_func_data($$$)
|
|
{
|
|
my ($funcdata1, $funcdata2, $filename) = @_;
|
|
my %result;
|
|
my $func;
|
|
|
|
if (defined($funcdata1)) {
|
|
%result = %{$funcdata1};
|
|
}
|
|
|
|
foreach $func (keys(%{$funcdata2})) {
|
|
my $line1 = $result{$func};
|
|
my $line2 = $funcdata2->{$func};
|
|
|
|
if (defined($line1) && ($line1 != $line2)) {
|
|
warn("WARNING: function data mismatch at ".
|
|
"$filename:$line2\n");
|
|
next;
|
|
}
|
|
$result{$func} = $line2;
|
|
}
|
|
|
|
return \%result;
|
|
}
|
|
|
|
|
|
#
|
|
# add_fnccount(fnccount1, fnccount2)
|
|
#
|
|
# Add function call count data. Return list (fnccount_added, f_found, f_hit)
|
|
#
|
|
|
|
sub add_fnccount($$)
|
|
{
|
|
my ($fnccount1, $fnccount2) = @_;
|
|
my %result;
|
|
my $fn_found;
|
|
my $fn_hit;
|
|
my $function;
|
|
|
|
if (defined($fnccount1)) {
|
|
%result = %{$fnccount1};
|
|
}
|
|
foreach $function (keys(%{$fnccount2})) {
|
|
$result{$function} += $fnccount2->{$function};
|
|
}
|
|
$fn_found = scalar(keys(%result));
|
|
$fn_hit = 0;
|
|
foreach $function (keys(%result)) {
|
|
if ($result{$function} > 0) {
|
|
$fn_hit++;
|
|
}
|
|
}
|
|
|
|
return (\%result, $fn_found, $fn_hit);
|
|
}
|
|
|
|
#
|
|
# add_testfncdata(testfncdata1, testfncdata2)
|
|
#
|
|
# Add function call count data for several tests. Return reference to
|
|
# added_testfncdata.
|
|
#
|
|
|
|
sub add_testfncdata($$)
|
|
{
|
|
my ($testfncdata1, $testfncdata2) = @_;
|
|
my %result;
|
|
my $testname;
|
|
|
|
foreach $testname (keys(%{$testfncdata1})) {
|
|
if (defined($testfncdata2->{$testname})) {
|
|
my $fnccount;
|
|
|
|
# Function call count data for this testname exists
|
|
# in both data sets: add
|
|
($fnccount) = add_fnccount(
|
|
$testfncdata1->{$testname},
|
|
$testfncdata2->{$testname});
|
|
$result{$testname} = $fnccount;
|
|
next;
|
|
}
|
|
# Function call count data for this testname is unique to
|
|
# data set 1: copy
|
|
$result{$testname} = $testfncdata1->{$testname};
|
|
}
|
|
|
|
# Add count data for testnames unique to data set 2
|
|
foreach $testname (keys(%{$testfncdata2})) {
|
|
if (!defined($result{$testname})) {
|
|
$result{$testname} = $testfncdata2->{$testname};
|
|
}
|
|
}
|
|
return \%result;
|
|
}
|
|
|
|
|
|
#
|
|
# brcount_to_db(brcount)
|
|
#
|
|
# Convert brcount data to the following format:
|
|
#
|
|
# db: line number -> block hash
|
|
# block hash: block number -> branch hash
|
|
# branch hash: branch number -> taken value
|
|
#
|
|
|
|
sub brcount_to_db($)
|
|
{
|
|
my ($brcount) = @_;
|
|
my $line;
|
|
my $db;
|
|
|
|
# Add branches from first count to database
|
|
foreach $line (keys(%{$brcount})) {
|
|
my $brdata = $brcount->{$line};
|
|
my $i;
|
|
my $num = br_ivec_len($brdata);
|
|
|
|
for ($i = 0; $i < $num; $i++) {
|
|
my ($block, $branch, $taken) = br_ivec_get($brdata, $i);
|
|
|
|
$db->{$line}->{$block}->{$branch} = $taken;
|
|
}
|
|
}
|
|
|
|
return $db;
|
|
}
|
|
|
|
|
|
#
|
|
# db_to_brcount(db)
|
|
#
|
|
# Convert branch coverage data back to brcount format.
|
|
#
|
|
|
|
sub db_to_brcount($)
|
|
{
|
|
my ($db) = @_;
|
|
my $line;
|
|
my $brcount = {};
|
|
my $br_found = 0;
|
|
my $br_hit = 0;
|
|
|
|
# Convert database back to brcount format
|
|
foreach $line (sort({$a <=> $b} keys(%{$db}))) {
|
|
my $ldata = $db->{$line};
|
|
my $brdata;
|
|
my $block;
|
|
|
|
foreach $block (sort({$a <=> $b} keys(%{$ldata}))) {
|
|
my $bdata = $ldata->{$block};
|
|
my $branch;
|
|
|
|
foreach $branch (sort({$a <=> $b} keys(%{$bdata}))) {
|
|
my $taken = $bdata->{$branch};
|
|
|
|
$br_found++;
|
|
$br_hit++ if ($taken ne "-" && $taken > 0);
|
|
$brdata = br_ivec_push($brdata, $block,
|
|
$branch, $taken);
|
|
}
|
|
}
|
|
$brcount->{$line} = $brdata;
|
|
}
|
|
|
|
return ($brcount, $br_found, $br_hit);
|
|
}
|
|
|
|
|
|
#
|
|
# combine_brcount(brcount1, brcount2, type)
|
|
#
|
|
# If add is BR_ADD, add branch coverage data and return list (brcount_added,
|
|
# br_found, br_hit). If add is BR_SUB, subtract the taken values of brcount2
|
|
# from brcount1 and return (brcount_sub, br_found, br_hit).
|
|
#
|
|
|
|
sub combine_brcount($$$)
|
|
{
|
|
my ($brcount1, $brcount2, $type) = @_;
|
|
my $line;
|
|
my $block;
|
|
my $branch;
|
|
my $taken;
|
|
my $db;
|
|
my $br_found = 0;
|
|
my $br_hit = 0;
|
|
my $result;
|
|
|
|
# Convert branches from first count to database
|
|
$db = brcount_to_db($brcount1);
|
|
# Combine values from database and second count
|
|
foreach $line (keys(%{$brcount2})) {
|
|
my $brdata = $brcount2->{$line};
|
|
my $num = br_ivec_len($brdata);
|
|
my $i;
|
|
|
|
for ($i = 0; $i < $num; $i++) {
|
|
($block, $branch, $taken) = br_ivec_get($brdata, $i);
|
|
my $new_taken = $db->{$line}->{$block}->{$branch};
|
|
|
|
if ($type == $BR_ADD) {
|
|
$new_taken = br_taken_add($new_taken, $taken);
|
|
} elsif ($type == $BR_SUB) {
|
|
$new_taken = br_taken_sub($new_taken, $taken);
|
|
}
|
|
$db->{$line}->{$block}->{$branch} = $new_taken
|
|
if (defined($new_taken));
|
|
}
|
|
}
|
|
# Convert database back to brcount format
|
|
($result, $br_found, $br_hit) = db_to_brcount($db);
|
|
|
|
return ($result, $br_found, $br_hit);
|
|
}
|
|
|
|
|
|
#
|
|
# add_testbrdata(testbrdata1, testbrdata2)
|
|
#
|
|
# Add branch coverage data for several tests. Return reference to
|
|
# added_testbrdata.
|
|
#
|
|
|
|
sub add_testbrdata($$)
|
|
{
|
|
my ($testbrdata1, $testbrdata2) = @_;
|
|
my %result;
|
|
my $testname;
|
|
|
|
foreach $testname (keys(%{$testbrdata1})) {
|
|
if (defined($testbrdata2->{$testname})) {
|
|
my $brcount;
|
|
|
|
# Branch coverage data for this testname exists
|
|
# in both data sets: add
|
|
($brcount) = combine_brcount($testbrdata1->{$testname},
|
|
$testbrdata2->{$testname}, $BR_ADD);
|
|
$result{$testname} = $brcount;
|
|
next;
|
|
}
|
|
# Branch coverage data for this testname is unique to
|
|
# data set 1: copy
|
|
$result{$testname} = $testbrdata1->{$testname};
|
|
}
|
|
|
|
# Add count data for testnames unique to data set 2
|
|
foreach $testname (keys(%{$testbrdata2})) {
|
|
if (!defined($result{$testname})) {
|
|
$result{$testname} = $testbrdata2->{$testname};
|
|
}
|
|
}
|
|
return \%result;
|
|
}
|
|
|
|
|
|
#
|
|
# combine_info_entries(entry_ref1, entry_ref2, filename)
|
|
#
|
|
# Combine .info data entry hashes referenced by ENTRY_REF1 and ENTRY_REF2.
|
|
# Return reference to resulting hash.
|
|
#
|
|
|
|
sub combine_info_entries($$$)
|
|
{
|
|
my $entry1 = $_[0]; # Reference to hash containing first entry
|
|
my $testdata1;
|
|
my $sumcount1;
|
|
my $funcdata1;
|
|
my $checkdata1;
|
|
my $testfncdata1;
|
|
my $sumfnccount1;
|
|
my $testbrdata1;
|
|
my $sumbrcount1;
|
|
|
|
my $entry2 = $_[1]; # Reference to hash containing second entry
|
|
my $testdata2;
|
|
my $sumcount2;
|
|
my $funcdata2;
|
|
my $checkdata2;
|
|
my $testfncdata2;
|
|
my $sumfnccount2;
|
|
my $testbrdata2;
|
|
my $sumbrcount2;
|
|
|
|
my %result; # Hash containing combined entry
|
|
my %result_testdata;
|
|
my $result_sumcount = {};
|
|
my $result_funcdata;
|
|
my $result_testfncdata;
|
|
my $result_sumfnccount;
|
|
my $result_testbrdata;
|
|
my $result_sumbrcount;
|
|
my $lines_found;
|
|
my $lines_hit;
|
|
my $fn_found;
|
|
my $fn_hit;
|
|
my $br_found;
|
|
my $br_hit;
|
|
|
|
my $testname;
|
|
my $filename = $_[2];
|
|
|
|
# Retrieve data
|
|
($testdata1, $sumcount1, $funcdata1, $checkdata1, $testfncdata1,
|
|
$sumfnccount1, $testbrdata1, $sumbrcount1) = get_info_entry($entry1);
|
|
($testdata2, $sumcount2, $funcdata2, $checkdata2, $testfncdata2,
|
|
$sumfnccount2, $testbrdata2, $sumbrcount2) = get_info_entry($entry2);
|
|
|
|
# Merge checksums
|
|
$checkdata1 = merge_checksums($checkdata1, $checkdata2, $filename);
|
|
|
|
# Combine funcdata
|
|
$result_funcdata = merge_func_data($funcdata1, $funcdata2, $filename);
|
|
|
|
# Combine function call count data
|
|
$result_testfncdata = add_testfncdata($testfncdata1, $testfncdata2);
|
|
($result_sumfnccount, $fn_found, $fn_hit) =
|
|
add_fnccount($sumfnccount1, $sumfnccount2);
|
|
|
|
# Combine branch coverage data
|
|
$result_testbrdata = add_testbrdata($testbrdata1, $testbrdata2);
|
|
($result_sumbrcount, $br_found, $br_hit) =
|
|
combine_brcount($sumbrcount1, $sumbrcount2, $BR_ADD);
|
|
|
|
# Combine testdata
|
|
foreach $testname (keys(%{$testdata1}))
|
|
{
|
|
if (defined($testdata2->{$testname}))
|
|
{
|
|
# testname is present in both entries, requires
|
|
# combination
|
|
($result_testdata{$testname}) =
|
|
add_counts($testdata1->{$testname},
|
|
$testdata2->{$testname});
|
|
}
|
|
else
|
|
{
|
|
# testname only present in entry1, add to result
|
|
$result_testdata{$testname} = $testdata1->{$testname};
|
|
}
|
|
|
|
# update sum count hash
|
|
($result_sumcount, $lines_found, $lines_hit) =
|
|
add_counts($result_sumcount,
|
|
$result_testdata{$testname});
|
|
}
|
|
|
|
foreach $testname (keys(%{$testdata2}))
|
|
{
|
|
# Skip testnames already covered by previous iteration
|
|
if (defined($testdata1->{$testname})) { next; }
|
|
|
|
# testname only present in entry2, add to result hash
|
|
$result_testdata{$testname} = $testdata2->{$testname};
|
|
|
|
# update sum count hash
|
|
($result_sumcount, $lines_found, $lines_hit) =
|
|
add_counts($result_sumcount,
|
|
$result_testdata{$testname});
|
|
}
|
|
|
|
# Calculate resulting sumcount
|
|
|
|
# Store result
|
|
set_info_entry(\%result, \%result_testdata, $result_sumcount,
|
|
$result_funcdata, $checkdata1, $result_testfncdata,
|
|
$result_sumfnccount, $result_testbrdata,
|
|
$result_sumbrcount, $lines_found, $lines_hit,
|
|
$fn_found, $fn_hit, $br_found, $br_hit);
|
|
|
|
return(\%result);
|
|
}
|
|
|
|
|
|
#
|
|
# combine_info_files(info_ref1, info_ref2)
|
|
#
|
|
# Combine .info data in hashes referenced by INFO_REF1 and INFO_REF2. Return
|
|
# reference to resulting hash.
|
|
#
|
|
|
|
sub combine_info_files($$)
|
|
{
|
|
my %hash1 = %{$_[0]};
|
|
my %hash2 = %{$_[1]};
|
|
my $filename;
|
|
|
|
foreach $filename (keys(%hash2))
|
|
{
|
|
if ($hash1{$filename})
|
|
{
|
|
# Entry already exists in hash1, combine them
|
|
$hash1{$filename} =
|
|
combine_info_entries($hash1{$filename},
|
|
$hash2{$filename},
|
|
$filename);
|
|
}
|
|
else
|
|
{
|
|
# Entry is unique in both hashes, simply add to
|
|
# resulting hash
|
|
$hash1{$filename} = $hash2{$filename};
|
|
}
|
|
}
|
|
|
|
return(\%hash1);
|
|
}
|
|
|
|
|
|
#
|
|
# get_prefix(min_dir, filename_list)
|
|
#
|
|
# Search FILENAME_LIST for a directory prefix which is common to as many
|
|
# list entries as possible, so that removing this prefix will minimize the
|
|
# sum of the lengths of all resulting shortened filenames while observing
|
|
# that no filename has less than MIN_DIR parent directories.
|
|
#
|
|
|
|
sub get_prefix($@)
|
|
{
|
|
my ($min_dir, @filename_list) = @_;
|
|
my %prefix; # mapping: prefix -> sum of lengths
|
|
my $current; # Temporary iteration variable
|
|
|
|
# Find list of prefixes
|
|
foreach (@filename_list)
|
|
{
|
|
# Need explicit assignment to get a copy of $_ so that
|
|
# shortening the contained prefix does not affect the list
|
|
$current = $_;
|
|
while ($current = shorten_prefix($current))
|
|
{
|
|
$current .= "/";
|
|
|
|
# Skip rest if the remaining prefix has already been
|
|
# added to hash
|
|
if (exists($prefix{$current})) { last; }
|
|
|
|
# Initialize with 0
|
|
$prefix{$current}="0";
|
|
}
|
|
|
|
}
|
|
|
|
# Remove all prefixes that would cause filenames to have less than
|
|
# the minimum number of parent directories
|
|
foreach my $filename (@filename_list) {
|
|
my $dir = dirname($filename);
|
|
|
|
for (my $i = 0; $i < $min_dir; $i++) {
|
|
delete($prefix{$dir."/"});
|
|
$dir = shorten_prefix($dir);
|
|
}
|
|
}
|
|
|
|
# Check if any prefix remains
|
|
return undef if (!%prefix);
|
|
|
|
# Calculate sum of lengths for all prefixes
|
|
foreach $current (keys(%prefix))
|
|
{
|
|
foreach (@filename_list)
|
|
{
|
|
# Add original length
|
|
$prefix{$current} += length($_);
|
|
|
|
# Check whether prefix matches
|
|
if (substr($_, 0, length($current)) eq $current)
|
|
{
|
|
# Subtract prefix length for this filename
|
|
$prefix{$current} -= length($current);
|
|
}
|
|
}
|
|
}
|
|
|
|
# Find and return prefix with minimal sum
|
|
$current = (keys(%prefix))[0];
|
|
|
|
foreach (keys(%prefix))
|
|
{
|
|
if ($prefix{$_} < $prefix{$current})
|
|
{
|
|
$current = $_;
|
|
}
|
|
}
|
|
|
|
$current =~ s/\/$//;
|
|
|
|
return($current);
|
|
}
|
|
|
|
|
|
#
|
|
# shorten_prefix(prefix)
|
|
#
|
|
# Return PREFIX shortened by last directory component.
|
|
#
|
|
|
|
sub shorten_prefix($)
|
|
{
|
|
my @list = split("/", $_[0]);
|
|
|
|
pop(@list);
|
|
return join("/", @list);
|
|
}
|
|
|
|
|
|
|
|
#
|
|
# get_dir_list(filename_list)
|
|
#
|
|
# Return sorted list of directories for each entry in given FILENAME_LIST.
|
|
#
|
|
|
|
sub get_dir_list(@)
|
|
{
|
|
my %result;
|
|
|
|
foreach (@_)
|
|
{
|
|
$result{shorten_prefix($_)} = "";
|
|
}
|
|
|
|
return(sort(keys(%result)));
|
|
}
|
|
|
|
|
|
#
|
|
# get_relative_base_path(subdirectory)
|
|
#
|
|
# Return a relative path string which references the base path when applied
|
|
# in SUBDIRECTORY.
|
|
#
|
|
# Example: get_relative_base_path("fs/mm") -> "../../"
|
|
#
|
|
|
|
sub get_relative_base_path($)
|
|
{
|
|
my $result = "";
|
|
my $index;
|
|
|
|
# Make an empty directory path a special case
|
|
if (!$_[0]) { return(""); }
|
|
|
|
# Count number of /s in path
|
|
$index = ($_[0] =~ s/\//\//g);
|
|
|
|
# Add a ../ to $result for each / in the directory path + 1
|
|
for (; $index>=0; $index--)
|
|
{
|
|
$result .= "../";
|
|
}
|
|
|
|
return $result;
|
|
}
|
|
|
|
|
|
#
|
|
# read_testfile(test_filename)
|
|
#
|
|
# Read in file TEST_FILENAME which contains test descriptions in the format:
|
|
#
|
|
# TN:<whitespace><test name>
|
|
# TD:<whitespace><test description>
|
|
#
|
|
# for each test case. Return a reference to a hash containing a mapping
|
|
#
|
|
# test name -> test description.
|
|
#
|
|
# Die on error.
|
|
#
|
|
|
|
sub read_testfile($)
|
|
{
|
|
my %result;
|
|
my $test_name;
|
|
my $changed_testname;
|
|
local *TEST_HANDLE;
|
|
|
|
open(TEST_HANDLE, "<", $_[0])
|
|
or die("ERROR: cannot open $_[0]!\n");
|
|
|
|
while (<TEST_HANDLE>)
|
|
{
|
|
chomp($_);
|
|
|
|
# Match lines beginning with TN:<whitespace(s)>
|
|
if (/^TN:\s+(.*?)\s*$/)
|
|
{
|
|
# Store name for later use
|
|
$test_name = $1;
|
|
if ($test_name =~ s/\W/_/g)
|
|
{
|
|
$changed_testname = 1;
|
|
}
|
|
}
|
|
|
|
# Match lines beginning with TD:<whitespace(s)>
|
|
if (/^TD:\s+(.*?)\s*$/)
|
|
{
|
|
# Check for empty line
|
|
if ($1)
|
|
{
|
|
# Add description to hash
|
|
$result{$test_name} .= " $1";
|
|
}
|
|
else
|
|
{
|
|
# Add empty line
|
|
$result{$test_name} .= "\n\n";
|
|
}
|
|
}
|
|
}
|
|
|
|
close(TEST_HANDLE);
|
|
|
|
if ($changed_testname)
|
|
{
|
|
warn("WARNING: invalid characters removed from testname in ".
|
|
"descriptions file $_[0]\n");
|
|
}
|
|
|
|
return \%result;
|
|
}
|
|
|
|
|
|
#
|
|
# escape_html(STRING)
|
|
#
|
|
# Return a copy of STRING in which all occurrences of HTML special characters
|
|
# are escaped.
|
|
#
|
|
|
|
sub escape_html($)
|
|
{
|
|
my $string = $_[0];
|
|
|
|
if (!$string) { return ""; }
|
|
|
|
$string =~ s/&/&/g; # & -> &
|
|
$string =~ s/</</g; # < -> <
|
|
$string =~ s/>/>/g; # > -> >
|
|
$string =~ s/\"/"/g; # " -> "
|
|
|
|
while ($string =~ /^([^\t]*)(\t)/)
|
|
{
|
|
my $replacement = " "x($tab_size - (length($1) % $tab_size));
|
|
$string =~ s/^([^\t]*)(\t)/$1$replacement/;
|
|
}
|
|
|
|
$string =~ s/\n/<br>/g; # \n -> <br>
|
|
|
|
return $string;
|
|
}
|
|
|
|
|
|
#
|
|
# get_date_string()
|
|
#
|
|
# Return the current date in the form: yyyy-mm-dd
|
|
#
|
|
|
|
sub get_date_string()
|
|
{
|
|
my $year;
|
|
my $month;
|
|
my $day;
|
|
|
|
($year, $month, $day) = (localtime())[5, 4, 3];
|
|
|
|
return sprintf("%d-%02d-%02d", $year+1900, $month+1, $day);
|
|
}
|
|
|
|
|
|
#
|
|
# create_sub_dir(dir_name)
|
|
#
|
|
# Create subdirectory DIR_NAME if it does not already exist, including all its
|
|
# parent directories.
|
|
#
|
|
# Die on error.
|
|
#
|
|
|
|
sub create_sub_dir($)
|
|
{
|
|
my ($dir) = @_;
|
|
|
|
system("mkdir", "-p" ,$dir)
|
|
and die("ERROR: cannot create directory $dir!\n");
|
|
}
|
|
|
|
|
|
#
|
|
# write_description_file(descriptions, overall_found, overall_hit,
|
|
# total_fn_found, total_fn_hit, total_br_found,
|
|
# total_br_hit)
|
|
#
|
|
# Write HTML file containing all test case descriptions. DESCRIPTIONS is a
|
|
# reference to a hash containing a mapping
|
|
#
|
|
# test case name -> test case description
|
|
#
|
|
# Die on error.
|
|
#
|
|
|
|
sub write_description_file($$$$$$$)
|
|
{
|
|
my %description = %{$_[0]};
|
|
my $found = $_[1];
|
|
my $hit = $_[2];
|
|
my $fn_found = $_[3];
|
|
my $fn_hit = $_[4];
|
|
my $br_found = $_[5];
|
|
my $br_hit = $_[6];
|
|
my $test_name;
|
|
local *HTML_HANDLE;
|
|
|
|
html_create(*HTML_HANDLE,"descriptions.$html_ext");
|
|
write_html_prolog(*HTML_HANDLE, "", "LCOV - test case descriptions");
|
|
write_header(*HTML_HANDLE, 3, "", "", $found, $hit, $fn_found,
|
|
$fn_hit, $br_found, $br_hit, 0);
|
|
|
|
write_test_table_prolog(*HTML_HANDLE,
|
|
"Test case descriptions - alphabetical list");
|
|
|
|
foreach $test_name (sort(keys(%description)))
|
|
{
|
|
write_test_table_entry(*HTML_HANDLE, $test_name,
|
|
escape_html($description{$test_name}));
|
|
}
|
|
|
|
write_test_table_epilog(*HTML_HANDLE);
|
|
write_html_epilog(*HTML_HANDLE, "");
|
|
|
|
close(*HTML_HANDLE);
|
|
}
|
|
|
|
|
|
|
|
#
|
|
# write_png_files()
|
|
#
|
|
# Create all necessary .png files for the HTML-output in the current
|
|
# directory. .png-files are used as bar graphs.
|
|
#
|
|
# Die on error.
|
|
#
|
|
|
|
sub write_png_files()
|
|
{
|
|
my %data;
|
|
local *PNG_HANDLE;
|
|
|
|
$data{"ruby.png"} =
|
|
[0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00,
|
|
0x00, 0x0d, 0x49, 0x48, 0x44, 0x52, 0x00, 0x00, 0x00, 0x01,
|
|
0x00, 0x00, 0x00, 0x01, 0x01, 0x03, 0x00, 0x00, 0x00, 0x25,
|
|
0xdb, 0x56, 0xca, 0x00, 0x00, 0x00, 0x07, 0x74, 0x49, 0x4d,
|
|
0x45, 0x07, 0xd2, 0x07, 0x11, 0x0f, 0x18, 0x10, 0x5d, 0x57,
|
|
0x34, 0x6e, 0x00, 0x00, 0x00, 0x09, 0x70, 0x48, 0x59, 0x73,
|
|
0x00, 0x00, 0x0b, 0x12, 0x00, 0x00, 0x0b, 0x12, 0x01, 0xd2,
|
|
0xdd, 0x7e, 0xfc, 0x00, 0x00, 0x00, 0x04, 0x67, 0x41, 0x4d,
|
|
0x41, 0x00, 0x00, 0xb1, 0x8f, 0x0b, 0xfc, 0x61, 0x05, 0x00,
|
|
0x00, 0x00, 0x06, 0x50, 0x4c, 0x54, 0x45, 0xff, 0x35, 0x2f,
|
|
0x00, 0x00, 0x00, 0xd0, 0x33, 0x9a, 0x9d, 0x00, 0x00, 0x00,
|
|
0x0a, 0x49, 0x44, 0x41, 0x54, 0x78, 0xda, 0x63, 0x60, 0x00,
|
|
0x00, 0x00, 0x02, 0x00, 0x01, 0xe5, 0x27, 0xde, 0xfc, 0x00,
|
|
0x00, 0x00, 0x00, 0x49, 0x45, 0x4e, 0x44, 0xae, 0x42, 0x60,
|
|
0x82];
|
|
$data{"amber.png"} =
|
|
[0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00,
|
|
0x00, 0x0d, 0x49, 0x48, 0x44, 0x52, 0x00, 0x00, 0x00, 0x01,
|
|
0x00, 0x00, 0x00, 0x01, 0x01, 0x03, 0x00, 0x00, 0x00, 0x25,
|
|
0xdb, 0x56, 0xca, 0x00, 0x00, 0x00, 0x07, 0x74, 0x49, 0x4d,
|
|
0x45, 0x07, 0xd2, 0x07, 0x11, 0x0f, 0x28, 0x04, 0x98, 0xcb,
|
|
0xd6, 0xe0, 0x00, 0x00, 0x00, 0x09, 0x70, 0x48, 0x59, 0x73,
|
|
0x00, 0x00, 0x0b, 0x12, 0x00, 0x00, 0x0b, 0x12, 0x01, 0xd2,
|
|
0xdd, 0x7e, 0xfc, 0x00, 0x00, 0x00, 0x04, 0x67, 0x41, 0x4d,
|
|
0x41, 0x00, 0x00, 0xb1, 0x8f, 0x0b, 0xfc, 0x61, 0x05, 0x00,
|
|
0x00, 0x00, 0x06, 0x50, 0x4c, 0x54, 0x45, 0xff, 0xe0, 0x50,
|
|
0x00, 0x00, 0x00, 0xa2, 0x7a, 0xda, 0x7e, 0x00, 0x00, 0x00,
|
|
0x0a, 0x49, 0x44, 0x41, 0x54, 0x78, 0xda, 0x63, 0x60, 0x00,
|
|
0x00, 0x00, 0x02, 0x00, 0x01, 0xe5, 0x27, 0xde, 0xfc, 0x00,
|
|
0x00, 0x00, 0x00, 0x49, 0x45, 0x4e, 0x44, 0xae, 0x42, 0x60,
|
|
0x82];
|
|
$data{"emerald.png"} =
|
|
[0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00,
|
|
0x00, 0x0d, 0x49, 0x48, 0x44, 0x52, 0x00, 0x00, 0x00, 0x01,
|
|
0x00, 0x00, 0x00, 0x01, 0x01, 0x03, 0x00, 0x00, 0x00, 0x25,
|
|
0xdb, 0x56, 0xca, 0x00, 0x00, 0x00, 0x07, 0x74, 0x49, 0x4d,
|
|
0x45, 0x07, 0xd2, 0x07, 0x11, 0x0f, 0x22, 0x2b, 0xc9, 0xf5,
|
|
0x03, 0x33, 0x00, 0x00, 0x00, 0x09, 0x70, 0x48, 0x59, 0x73,
|
|
0x00, 0x00, 0x0b, 0x12, 0x00, 0x00, 0x0b, 0x12, 0x01, 0xd2,
|
|
0xdd, 0x7e, 0xfc, 0x00, 0x00, 0x00, 0x04, 0x67, 0x41, 0x4d,
|
|
0x41, 0x00, 0x00, 0xb1, 0x8f, 0x0b, 0xfc, 0x61, 0x05, 0x00,
|
|
0x00, 0x00, 0x06, 0x50, 0x4c, 0x54, 0x45, 0x1b, 0xea, 0x59,
|
|
0x0a, 0x0a, 0x0a, 0x0f, 0xba, 0x50, 0x83, 0x00, 0x00, 0x00,
|
|
0x0a, 0x49, 0x44, 0x41, 0x54, 0x78, 0xda, 0x63, 0x60, 0x00,
|
|
0x00, 0x00, 0x02, 0x00, 0x01, 0xe5, 0x27, 0xde, 0xfc, 0x00,
|
|
0x00, 0x00, 0x00, 0x49, 0x45, 0x4e, 0x44, 0xae, 0x42, 0x60,
|
|
0x82];
|
|
$data{"snow.png"} =
|
|
[0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00,
|
|
0x00, 0x0d, 0x49, 0x48, 0x44, 0x52, 0x00, 0x00, 0x00, 0x01,
|
|
0x00, 0x00, 0x00, 0x01, 0x01, 0x03, 0x00, 0x00, 0x00, 0x25,
|
|
0xdb, 0x56, 0xca, 0x00, 0x00, 0x00, 0x07, 0x74, 0x49, 0x4d,
|
|
0x45, 0x07, 0xd2, 0x07, 0x11, 0x0f, 0x1e, 0x1d, 0x75, 0xbc,
|
|
0xef, 0x55, 0x00, 0x00, 0x00, 0x09, 0x70, 0x48, 0x59, 0x73,
|
|
0x00, 0x00, 0x0b, 0x12, 0x00, 0x00, 0x0b, 0x12, 0x01, 0xd2,
|
|
0xdd, 0x7e, 0xfc, 0x00, 0x00, 0x00, 0x04, 0x67, 0x41, 0x4d,
|
|
0x41, 0x00, 0x00, 0xb1, 0x8f, 0x0b, 0xfc, 0x61, 0x05, 0x00,
|
|
0x00, 0x00, 0x06, 0x50, 0x4c, 0x54, 0x45, 0xff, 0xff, 0xff,
|
|
0x00, 0x00, 0x00, 0x55, 0xc2, 0xd3, 0x7e, 0x00, 0x00, 0x00,
|
|
0x0a, 0x49, 0x44, 0x41, 0x54, 0x78, 0xda, 0x63, 0x60, 0x00,
|
|
0x00, 0x00, 0x02, 0x00, 0x01, 0xe5, 0x27, 0xde, 0xfc, 0x00,
|
|
0x00, 0x00, 0x00, 0x49, 0x45, 0x4e, 0x44, 0xae, 0x42, 0x60,
|
|
0x82];
|
|
$data{"glass.png"} =
|
|
[0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00,
|
|
0x00, 0x0d, 0x49, 0x48, 0x44, 0x52, 0x00, 0x00, 0x00, 0x01,
|
|
0x00, 0x00, 0x00, 0x01, 0x01, 0x03, 0x00, 0x00, 0x00, 0x25,
|
|
0xdb, 0x56, 0xca, 0x00, 0x00, 0x00, 0x04, 0x67, 0x41, 0x4d,
|
|
0x41, 0x00, 0x00, 0xb1, 0x8f, 0x0b, 0xfc, 0x61, 0x05, 0x00,
|
|
0x00, 0x00, 0x06, 0x50, 0x4c, 0x54, 0x45, 0xff, 0xff, 0xff,
|
|
0x00, 0x00, 0x00, 0x55, 0xc2, 0xd3, 0x7e, 0x00, 0x00, 0x00,
|
|
0x01, 0x74, 0x52, 0x4e, 0x53, 0x00, 0x40, 0xe6, 0xd8, 0x66,
|
|
0x00, 0x00, 0x00, 0x01, 0x62, 0x4b, 0x47, 0x44, 0x00, 0x88,
|
|
0x05, 0x1d, 0x48, 0x00, 0x00, 0x00, 0x09, 0x70, 0x48, 0x59,
|
|
0x73, 0x00, 0x00, 0x0b, 0x12, 0x00, 0x00, 0x0b, 0x12, 0x01,
|
|
0xd2, 0xdd, 0x7e, 0xfc, 0x00, 0x00, 0x00, 0x07, 0x74, 0x49,
|
|
0x4d, 0x45, 0x07, 0xd2, 0x07, 0x13, 0x0f, 0x08, 0x19, 0xc4,
|
|
0x40, 0x56, 0x10, 0x00, 0x00, 0x00, 0x0a, 0x49, 0x44, 0x41,
|
|
0x54, 0x78, 0x9c, 0x63, 0x60, 0x00, 0x00, 0x00, 0x02, 0x00,
|
|
0x01, 0x48, 0xaf, 0xa4, 0x71, 0x00, 0x00, 0x00, 0x00, 0x49,
|
|
0x45, 0x4e, 0x44, 0xae, 0x42, 0x60, 0x82];
|
|
$data{"updown.png"} =
|
|
[0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00,
|
|
0x00, 0x0d, 0x49, 0x48, 0x44, 0x52, 0x00, 0x00, 0x00, 0x0a,
|
|
0x00, 0x00, 0x00, 0x0e, 0x08, 0x06, 0x00, 0x00, 0x00, 0x16,
|
|
0xa3, 0x8d, 0xab, 0x00, 0x00, 0x00, 0x3c, 0x49, 0x44, 0x41,
|
|
0x54, 0x28, 0xcf, 0x63, 0x60, 0x40, 0x03, 0xff, 0xa1, 0x00,
|
|
0x5d, 0x9c, 0x11, 0x5d, 0x11, 0x8a, 0x24, 0x23, 0x23, 0x23,
|
|
0x86, 0x42, 0x6c, 0xa6, 0x20, 0x2b, 0x66, 0xc4, 0xa7, 0x08,
|
|
0x59, 0x31, 0x23, 0x21, 0x45, 0x30, 0xc0, 0xc4, 0x30, 0x60,
|
|
0x80, 0xfa, 0x6e, 0x24, 0x3e, 0x78, 0x48, 0x0a, 0x70, 0x62,
|
|
0xa2, 0x90, 0x81, 0xd8, 0x44, 0x01, 0x00, 0xe9, 0x5c, 0x2f,
|
|
0xf5, 0xe2, 0x9d, 0x0f, 0xf9, 0x00, 0x00, 0x00, 0x00, 0x49,
|
|
0x45, 0x4e, 0x44, 0xae, 0x42, 0x60, 0x82] if ($sort);
|
|
foreach (keys(%data))
|
|
{
|
|
open(PNG_HANDLE, ">", $_)
|
|
or die("ERROR: cannot create $_!\n");
|
|
binmode(PNG_HANDLE);
|
|
print(PNG_HANDLE map(chr,@{$data{$_}}));
|
|
close(PNG_HANDLE);
|
|
}
|
|
}
|
|
|
|
|
|
#
|
|
# write_htaccess_file()
|
|
#
|
|
|
|
sub write_htaccess_file()
|
|
{
|
|
local *HTACCESS_HANDLE;
|
|
my $htaccess_data;
|
|
|
|
open(*HTACCESS_HANDLE, ">", ".htaccess")
|
|
or die("ERROR: cannot open .htaccess for writing!\n");
|
|
|
|
$htaccess_data = (<<"END_OF_HTACCESS")
|
|
AddEncoding x-gzip .html
|
|
END_OF_HTACCESS
|
|
;
|
|
|
|
print(HTACCESS_HANDLE $htaccess_data);
|
|
close(*HTACCESS_HANDLE);
|
|
}
|
|
|
|
|
|
#
|
|
# write_css_file()
|
|
#
|
|
# Write the cascading style sheet file gcov.css to the current directory.
|
|
# This file defines basic layout attributes of all generated HTML pages.
|
|
#
|
|
|
|
sub write_css_file()
|
|
{
|
|
local *CSS_HANDLE;
|
|
|
|
# Check for a specified external style sheet file
|
|
if ($css_filename)
|
|
{
|
|
# Simply copy that file
|
|
system("cp", $css_filename, "gcov.css")
|
|
and die("ERROR: cannot copy file $css_filename!\n");
|
|
return;
|
|
}
|
|
|
|
open(CSS_HANDLE, ">", "gcov.css")
|
|
or die ("ERROR: cannot open gcov.css for writing!\n");
|
|
|
|
|
|
# *************************************************************
|
|
|
|
my $css_data = ($_=<<"END_OF_CSS")
|
|
/* All views: initial background and text color */
|
|
body
|
|
{
|
|
color: #000000;
|
|
background-color: #FFFFFF;
|
|
}
|
|
|
|
/* All views: standard link format*/
|
|
a:link
|
|
{
|
|
color: #284FA8;
|
|
text-decoration: underline;
|
|
}
|
|
|
|
/* All views: standard link - visited format */
|
|
a:visited
|
|
{
|
|
color: #00CB40;
|
|
text-decoration: underline;
|
|
}
|
|
|
|
/* All views: standard link - activated format */
|
|
a:active
|
|
{
|
|
color: #FF0040;
|
|
text-decoration: underline;
|
|
}
|
|
|
|
/* All views: main title format */
|
|
td.title
|
|
{
|
|
text-align: center;
|
|
padding-bottom: 10px;
|
|
font-family: sans-serif;
|
|
font-size: 20pt;
|
|
font-style: italic;
|
|
font-weight: bold;
|
|
}
|
|
|
|
/* All views: header item format */
|
|
td.headerItem
|
|
{
|
|
text-align: right;
|
|
padding-right: 6px;
|
|
font-family: sans-serif;
|
|
font-weight: bold;
|
|
vertical-align: top;
|
|
white-space: nowrap;
|
|
}
|
|
|
|
/* All views: header item value format */
|
|
td.headerValue
|
|
{
|
|
text-align: left;
|
|
color: #284FA8;
|
|
font-family: sans-serif;
|
|
font-weight: bold;
|
|
white-space: nowrap;
|
|
}
|
|
|
|
/* All views: header item coverage table heading */
|
|
td.headerCovTableHead
|
|
{
|
|
text-align: center;
|
|
padding-right: 6px;
|
|
padding-left: 6px;
|
|
padding-bottom: 0px;
|
|
font-family: sans-serif;
|
|
font-size: 80%;
|
|
white-space: nowrap;
|
|
}
|
|
|
|
/* All views: header item coverage table entry */
|
|
td.headerCovTableEntry
|
|
{
|
|
text-align: right;
|
|
color: #284FA8;
|
|
font-family: sans-serif;
|
|
font-weight: bold;
|
|
white-space: nowrap;
|
|
padding-left: 12px;
|
|
padding-right: 4px;
|
|
background-color: #DAE7FE;
|
|
}
|
|
|
|
/* All views: header item coverage table entry for high coverage rate */
|
|
td.headerCovTableEntryHi
|
|
{
|
|
text-align: right;
|
|
color: #000000;
|
|
font-family: sans-serif;
|
|
font-weight: bold;
|
|
white-space: nowrap;
|
|
padding-left: 12px;
|
|
padding-right: 4px;
|
|
background-color: #A7FC9D;
|
|
}
|
|
|
|
/* All views: header item coverage table entry for medium coverage rate */
|
|
td.headerCovTableEntryMed
|
|
{
|
|
text-align: right;
|
|
color: #000000;
|
|
font-family: sans-serif;
|
|
font-weight: bold;
|
|
white-space: nowrap;
|
|
padding-left: 12px;
|
|
padding-right: 4px;
|
|
background-color: #FFEA20;
|
|
}
|
|
|
|
/* All views: header item coverage table entry for ow coverage rate */
|
|
td.headerCovTableEntryLo
|
|
{
|
|
text-align: right;
|
|
color: #000000;
|
|
font-family: sans-serif;
|
|
font-weight: bold;
|
|
white-space: nowrap;
|
|
padding-left: 12px;
|
|
padding-right: 4px;
|
|
background-color: #FF0000;
|
|
}
|
|
|
|
/* All views: header legend value for legend entry */
|
|
td.headerValueLeg
|
|
{
|
|
text-align: left;
|
|
color: #000000;
|
|
font-family: sans-serif;
|
|
font-size: 80%;
|
|
white-space: nowrap;
|
|
padding-top: 4px;
|
|
}
|
|
|
|
/* All views: color of horizontal ruler */
|
|
td.ruler
|
|
{
|
|
background-color: #6688D4;
|
|
}
|
|
|
|
/* All views: version string format */
|
|
td.versionInfo
|
|
{
|
|
text-align: center;
|
|
padding-top: 2px;
|
|
font-family: sans-serif;
|
|
font-style: italic;
|
|
}
|
|
|
|
/* Directory view/File view (all)/Test case descriptions:
|
|
table headline format */
|
|
td.tableHead
|
|
{
|
|
text-align: center;
|
|
color: #FFFFFF;
|
|
background-color: #6688D4;
|
|
font-family: sans-serif;
|
|
font-size: 120%;
|
|
font-weight: bold;
|
|
white-space: nowrap;
|
|
padding-left: 4px;
|
|
padding-right: 4px;
|
|
}
|
|
|
|
span.tableHeadSort
|
|
{
|
|
padding-right: 4px;
|
|
}
|
|
|
|
/* Directory view/File view (all): filename entry format */
|
|
td.coverFile
|
|
{
|
|
text-align: left;
|
|
padding-left: 10px;
|
|
padding-right: 20px;
|
|
color: #284FA8;
|
|
background-color: #DAE7FE;
|
|
font-family: monospace;
|
|
}
|
|
|
|
/* Directory view/File view (all): bar-graph entry format*/
|
|
td.coverBar
|
|
{
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #DAE7FE;
|
|
}
|
|
|
|
/* Directory view/File view (all): bar-graph outline color */
|
|
td.coverBarOutline
|
|
{
|
|
background-color: #000000;
|
|
}
|
|
|
|
/* Directory view/File view (all): percentage entry for files with
|
|
high coverage rate */
|
|
td.coverPerHi
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #A7FC9D;
|
|
font-weight: bold;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* Directory view/File view (all): line count entry for files with
|
|
high coverage rate */
|
|
td.coverNumHi
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #A7FC9D;
|
|
white-space: nowrap;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* Directory view/File view (all): percentage entry for files with
|
|
medium coverage rate */
|
|
td.coverPerMed
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #FFEA20;
|
|
font-weight: bold;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* Directory view/File view (all): line count entry for files with
|
|
medium coverage rate */
|
|
td.coverNumMed
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #FFEA20;
|
|
white-space: nowrap;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* Directory view/File view (all): percentage entry for files with
|
|
low coverage rate */
|
|
td.coverPerLo
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #FF0000;
|
|
font-weight: bold;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* Directory view/File view (all): line count entry for files with
|
|
low coverage rate */
|
|
td.coverNumLo
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #FF0000;
|
|
white-space: nowrap;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* File view (all): "show/hide details" link format */
|
|
a.detail:link
|
|
{
|
|
color: #B8D0FF;
|
|
font-size:80%;
|
|
}
|
|
|
|
/* File view (all): "show/hide details" link - visited format */
|
|
a.detail:visited
|
|
{
|
|
color: #B8D0FF;
|
|
font-size:80%;
|
|
}
|
|
|
|
/* File view (all): "show/hide details" link - activated format */
|
|
a.detail:active
|
|
{
|
|
color: #FFFFFF;
|
|
font-size:80%;
|
|
}
|
|
|
|
/* File view (detail): test name entry */
|
|
td.testName
|
|
{
|
|
text-align: right;
|
|
padding-right: 10px;
|
|
background-color: #DAE7FE;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* File view (detail): test percentage entry */
|
|
td.testPer
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #DAE7FE;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* File view (detail): test lines count entry */
|
|
td.testNum
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #DAE7FE;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* Test case descriptions: test name format*/
|
|
dt
|
|
{
|
|
font-family: sans-serif;
|
|
font-weight: bold;
|
|
}
|
|
|
|
/* Test case descriptions: description table body */
|
|
td.testDescription
|
|
{
|
|
padding-top: 10px;
|
|
padding-left: 30px;
|
|
padding-bottom: 10px;
|
|
padding-right: 30px;
|
|
background-color: #DAE7FE;
|
|
}
|
|
|
|
/* Source code view: function entry */
|
|
td.coverFn
|
|
{
|
|
text-align: left;
|
|
padding-left: 10px;
|
|
padding-right: 20px;
|
|
color: #284FA8;
|
|
background-color: #DAE7FE;
|
|
font-family: monospace;
|
|
}
|
|
|
|
/* Source code view: function entry zero count*/
|
|
td.coverFnLo
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #FF0000;
|
|
font-weight: bold;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* Source code view: function entry nonzero count*/
|
|
td.coverFnHi
|
|
{
|
|
text-align: right;
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
background-color: #DAE7FE;
|
|
font-weight: bold;
|
|
font-family: sans-serif;
|
|
}
|
|
|
|
/* Source code view: source code format */
|
|
pre.source
|
|
{
|
|
font-family: monospace;
|
|
white-space: pre;
|
|
margin-top: 2px;
|
|
}
|
|
|
|
/* Source code view: line number format */
|
|
span.lineNum
|
|
{
|
|
background-color: #EFE383;
|
|
}
|
|
|
|
/* Source code view: format for lines which were executed */
|
|
td.lineCov,
|
|
span.lineCov
|
|
{
|
|
background-color: #CAD7FE;
|
|
}
|
|
|
|
/* Source code view: format for Cov legend */
|
|
span.coverLegendCov
|
|
{
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
padding-bottom: 2px;
|
|
background-color: #CAD7FE;
|
|
}
|
|
|
|
/* Source code view: format for lines which were not executed */
|
|
td.lineNoCov,
|
|
span.lineNoCov
|
|
{
|
|
background-color: #FF6230;
|
|
}
|
|
|
|
/* Source code view: format for NoCov legend */
|
|
span.coverLegendNoCov
|
|
{
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
padding-bottom: 2px;
|
|
background-color: #FF6230;
|
|
}
|
|
|
|
/* Source code view (function table): standard link - visited format */
|
|
td.lineNoCov > a:visited,
|
|
td.lineCov > a:visited
|
|
{
|
|
color: black;
|
|
text-decoration: underline;
|
|
}
|
|
|
|
/* Source code view: format for lines which were executed only in a
|
|
previous version */
|
|
span.lineDiffCov
|
|
{
|
|
background-color: #B5F7AF;
|
|
}
|
|
|
|
/* Source code view: format for branches which were executed
|
|
* and taken */
|
|
span.branchCov
|
|
{
|
|
background-color: #CAD7FE;
|
|
}
|
|
|
|
/* Source code view: format for branches which were executed
|
|
* but not taken */
|
|
span.branchNoCov
|
|
{
|
|
background-color: #FF6230;
|
|
}
|
|
|
|
/* Source code view: format for branches which were not executed */
|
|
span.branchNoExec
|
|
{
|
|
background-color: #FF6230;
|
|
}
|
|
|
|
/* Source code view: format for the source code heading line */
|
|
pre.sourceHeading
|
|
{
|
|
white-space: pre;
|
|
font-family: monospace;
|
|
font-weight: bold;
|
|
margin: 0px;
|
|
}
|
|
|
|
/* All views: header legend value for low rate */
|
|
td.headerValueLegL
|
|
{
|
|
font-family: sans-serif;
|
|
text-align: center;
|
|
white-space: nowrap;
|
|
padding-left: 4px;
|
|
padding-right: 2px;
|
|
background-color: #FF0000;
|
|
font-size: 80%;
|
|
}
|
|
|
|
/* All views: header legend value for med rate */
|
|
td.headerValueLegM
|
|
{
|
|
font-family: sans-serif;
|
|
text-align: center;
|
|
white-space: nowrap;
|
|
padding-left: 2px;
|
|
padding-right: 2px;
|
|
background-color: #FFEA20;
|
|
font-size: 80%;
|
|
}
|
|
|
|
/* All views: header legend value for hi rate */
|
|
td.headerValueLegH
|
|
{
|
|
font-family: sans-serif;
|
|
text-align: center;
|
|
white-space: nowrap;
|
|
padding-left: 2px;
|
|
padding-right: 4px;
|
|
background-color: #A7FC9D;
|
|
font-size: 80%;
|
|
}
|
|
|
|
/* All views except source code view: legend format for low coverage */
|
|
span.coverLegendCovLo
|
|
{
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
padding-top: 2px;
|
|
background-color: #FF0000;
|
|
}
|
|
|
|
/* All views except source code view: legend format for med coverage */
|
|
span.coverLegendCovMed
|
|
{
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
padding-top: 2px;
|
|
background-color: #FFEA20;
|
|
}
|
|
|
|
/* All views except source code view: legend format for hi coverage */
|
|
span.coverLegendCovHi
|
|
{
|
|
padding-left: 10px;
|
|
padding-right: 10px;
|
|
padding-top: 2px;
|
|
background-color: #A7FC9D;
|
|
}
|
|
END_OF_CSS
|
|
;
|
|
|
|
# *************************************************************
|
|
|
|
|
|
# Remove leading tab from all lines
|
|
$css_data =~ s/^\t//gm;
|
|
|
|
print(CSS_HANDLE $css_data);
|
|
|
|
close(CSS_HANDLE);
|
|
}
|
|
|
|
|
|
#
|
|
# get_bar_graph_code(base_dir, cover_found, cover_hit)
|
|
#
|
|
# Return a string containing HTML code which implements a bar graph display
|
|
# for a coverage rate of cover_hit * 100 / cover_found.
|
|
#
|
|
|
|
sub get_bar_graph_code($$$)
|
|
{
|
|
my ($base_dir, $found, $hit) = @_;
|
|
my $rate;
|
|
my $alt;
|
|
my $width;
|
|
my $remainder;
|
|
my $png_name;
|
|
my $graph_code;
|
|
|
|
# Check number of instrumented lines
|
|
if ($_[1] == 0) { return ""; }
|
|
|
|
$alt = rate($hit, $found, "%");
|
|
$width = rate($hit, $found, undef, 0);
|
|
$remainder = 100 - $width;
|
|
|
|
# Decide which .png file to use
|
|
$png_name = $rate_png[classify_rate($found, $hit, $med_limit,
|
|
$hi_limit)];
|
|
|
|
if ($width == 0)
|
|
{
|
|
# Zero coverage
|
|
$graph_code = (<<END_OF_HTML)
|
|
<table border=0 cellspacing=0 cellpadding=1><tr><td class="coverBarOutline"><img src="$_[0]snow.png" width=100 height=10 alt="$alt"></td></tr></table>
|
|
END_OF_HTML
|
|
;
|
|
}
|
|
elsif ($width == 100)
|
|
{
|
|
# Full coverage
|
|
$graph_code = (<<END_OF_HTML)
|
|
<table border=0 cellspacing=0 cellpadding=1><tr><td class="coverBarOutline"><img src="$_[0]$png_name" width=100 height=10 alt="$alt"></td></tr></table>
|
|
END_OF_HTML
|
|
;
|
|
}
|
|
else
|
|
{
|
|
# Positive coverage
|
|
$graph_code = (<<END_OF_HTML)
|
|
<table border=0 cellspacing=0 cellpadding=1><tr><td class="coverBarOutline"><img src="$_[0]$png_name" width=$width height=10 alt="$alt"><img src="$_[0]snow.png" width=$remainder height=10 alt="$alt"></td></tr></table>
|
|
END_OF_HTML
|
|
;
|
|
}
|
|
|
|
# Remove leading tabs from all lines
|
|
$graph_code =~ s/^\t+//gm;
|
|
chomp($graph_code);
|
|
|
|
return($graph_code);
|
|
}
|
|
|
|
#
|
|
# sub classify_rate(found, hit, med_limit, high_limit)
|
|
#
|
|
# Return 0 for low rate, 1 for medium rate and 2 for hi rate.
|
|
#
|
|
|
|
sub classify_rate($$$$)
|
|
{
|
|
my ($found, $hit, $med, $hi) = @_;
|
|
my $rate;
|
|
|
|
if ($found == 0) {
|
|
return 2;
|
|
}
|
|
$rate = rate($hit, $found);
|
|
if ($rate < $med) {
|
|
return 0;
|
|
} elsif ($rate < $hi) {
|
|
return 1;
|
|
}
|
|
return 2;
|
|
}
|
|
|
|
|
|
#
|
|
# write_html(filehandle, html_code)
|
|
#
|
|
# Write out HTML_CODE to FILEHANDLE while removing a leading tabulator mark
|
|
# in each line of HTML_CODE.
|
|
#
|
|
|
|
sub write_html(*$)
|
|
{
|
|
local *HTML_HANDLE = $_[0];
|
|
my $html_code = $_[1];
|
|
|
|
# Remove leading tab from all lines
|
|
$html_code =~ s/^\t//gm;
|
|
|
|
print(HTML_HANDLE $html_code)
|
|
or die("ERROR: cannot write HTML data ($!)\n");
|
|
}
|
|
|
|
|
|
#
|
|
# write_html_prolog(filehandle, base_dir, pagetitle)
|
|
#
|
|
# Write an HTML prolog common to all HTML files to FILEHANDLE. PAGETITLE will
|
|
# be used as HTML page title. BASE_DIR contains a relative path which points
|
|
# to the base directory.
|
|
#
|
|
|
|
sub write_html_prolog(*$$)
|
|
{
|
|
my $basedir = $_[1];
|
|
my $pagetitle = $_[2];
|
|
my $prolog;
|
|
|
|
$prolog = $html_prolog;
|
|
$prolog =~ s/\@pagetitle\@/$pagetitle/g;
|
|
$prolog =~ s/\@basedir\@/$basedir/g;
|
|
|
|
write_html($_[0], $prolog);
|
|
}
|
|
|
|
|
|
#
|
|
# write_header_prolog(filehandle, base_dir)
|
|
#
|
|
# Write beginning of page header HTML code.
|
|
#
|
|
|
|
sub write_header_prolog(*$)
|
|
{
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
<table width="100%" border=0 cellspacing=0 cellpadding=0>
|
|
<tr><td class="title">$title</td></tr>
|
|
<tr><td class="ruler"><img src="$_[1]glass.png" width=3 height=3 alt=""></td></tr>
|
|
|
|
<tr>
|
|
<td width="100%">
|
|
<table cellpadding=1 border=0 width="100%">
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# write_header_line(handle, content)
|
|
#
|
|
# Write a header line with the specified table contents.
|
|
#
|
|
|
|
sub write_header_line(*@)
|
|
{
|
|
my ($handle, @content) = @_;
|
|
my $entry;
|
|
|
|
write_html($handle, " <tr>\n");
|
|
foreach $entry (@content) {
|
|
my ($width, $class, $text, $colspan) = @{$entry};
|
|
|
|
if (defined($width)) {
|
|
$width = " width=\"$width\"";
|
|
} else {
|
|
$width = "";
|
|
}
|
|
if (defined($class)) {
|
|
$class = " class=\"$class\"";
|
|
} else {
|
|
$class = "";
|
|
}
|
|
if (defined($colspan)) {
|
|
$colspan = " colspan=\"$colspan\"";
|
|
} else {
|
|
$colspan = "";
|
|
}
|
|
$text = "" if (!defined($text));
|
|
write_html($handle,
|
|
" <td$width$class$colspan>$text</td>\n");
|
|
}
|
|
write_html($handle, " </tr>\n");
|
|
}
|
|
|
|
|
|
#
|
|
# write_header_epilog(filehandle, base_dir)
|
|
#
|
|
# Write end of page header HTML code.
|
|
#
|
|
|
|
sub write_header_epilog(*$)
|
|
{
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
<tr><td><img src="$_[1]glass.png" width=3 height=3 alt=""></td></tr>
|
|
</table>
|
|
</td>
|
|
</tr>
|
|
|
|
<tr><td class="ruler"><img src="$_[1]glass.png" width=3 height=3 alt=""></td></tr>
|
|
</table>
|
|
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# write_file_table_prolog(handle, file_heading, ([heading, num_cols], ...))
|
|
#
|
|
# Write heading for file table.
|
|
#
|
|
|
|
sub write_file_table_prolog(*$@)
|
|
{
|
|
my ($handle, $file_heading, @columns) = @_;
|
|
my $num_columns = 0;
|
|
my $file_width;
|
|
my $col;
|
|
my $width;
|
|
|
|
$width = 20 if (scalar(@columns) == 1);
|
|
$width = 10 if (scalar(@columns) == 2);
|
|
$width = 8 if (scalar(@columns) > 2);
|
|
|
|
foreach $col (@columns) {
|
|
my ($heading, $cols) = @{$col};
|
|
|
|
$num_columns += $cols;
|
|
}
|
|
$file_width = 100 - $num_columns * $width;
|
|
|
|
# Table definition
|
|
write_html($handle, <<END_OF_HTML);
|
|
<center>
|
|
<table width="80%" cellpadding=1 cellspacing=1 border=0>
|
|
|
|
<tr>
|
|
<td width="$file_width%"><br></td>
|
|
END_OF_HTML
|
|
# Empty first row
|
|
foreach $col (@columns) {
|
|
my ($heading, $cols) = @{$col};
|
|
|
|
while ($cols-- > 0) {
|
|
write_html($handle, <<END_OF_HTML);
|
|
<td width="$width%"></td>
|
|
END_OF_HTML
|
|
}
|
|
}
|
|
# Next row
|
|
write_html($handle, <<END_OF_HTML);
|
|
</tr>
|
|
|
|
<tr>
|
|
<td class="tableHead">$file_heading</td>
|
|
END_OF_HTML
|
|
# Heading row
|
|
foreach $col (@columns) {
|
|
my ($heading, $cols) = @{$col};
|
|
my $colspan = "";
|
|
|
|
$colspan = " colspan=$cols" if ($cols > 1);
|
|
write_html($handle, <<END_OF_HTML);
|
|
<td class="tableHead"$colspan>$heading</td>
|
|
END_OF_HTML
|
|
}
|
|
write_html($handle, <<END_OF_HTML);
|
|
</tr>
|
|
END_OF_HTML
|
|
}
|
|
|
|
|
|
# write_file_table_entry(handle, base_dir, filename, page_link,
|
|
# ([ found, hit, med_limit, hi_limit, graph ], ..)
|
|
#
|
|
# Write an entry of the file table.
|
|
#
|
|
|
|
sub write_file_table_entry(*$$$@)
|
|
{
|
|
my ($handle, $base_dir, $filename, $page_link, @entries) = @_;
|
|
my $file_code;
|
|
my $entry;
|
|
my $esc_filename = escape_html($filename);
|
|
|
|
# Add link to source if provided
|
|
if (defined($page_link) && $page_link ne "") {
|
|
$file_code = "<a href=\"$page_link\">$esc_filename</a>";
|
|
} else {
|
|
$file_code = $esc_filename;
|
|
}
|
|
|
|
# First column: filename
|
|
write_html($handle, <<END_OF_HTML);
|
|
<tr>
|
|
<td class="coverFile">$file_code</td>
|
|
END_OF_HTML
|
|
# Columns as defined
|
|
foreach $entry (@entries) {
|
|
my ($found, $hit, $med, $hi, $graph) = @{$entry};
|
|
my $bar_graph;
|
|
my $class;
|
|
my $rate;
|
|
|
|
# Generate bar graph if requested
|
|
if ($graph) {
|
|
$bar_graph = get_bar_graph_code($base_dir, $found,
|
|
$hit);
|
|
write_html($handle, <<END_OF_HTML);
|
|
<td class="coverBar" align="center">
|
|
$bar_graph
|
|
</td>
|
|
END_OF_HTML
|
|
}
|
|
# Get rate color and text
|
|
if ($found == 0) {
|
|
$rate = "-";
|
|
$class = "Hi";
|
|
} else {
|
|
$rate = rate($hit, $found, " %");
|
|
$class = $rate_name[classify_rate($found, $hit,
|
|
$med, $hi)];
|
|
}
|
|
write_html($handle, <<END_OF_HTML);
|
|
<td class="coverPer$class">$rate</td>
|
|
<td class="coverNum$class">$hit / $found</td>
|
|
END_OF_HTML
|
|
}
|
|
# End of row
|
|
write_html($handle, <<END_OF_HTML);
|
|
</tr>
|
|
END_OF_HTML
|
|
}
|
|
|
|
|
|
#
|
|
# write_file_table_detail_entry(filehandle, test_name, ([found, hit], ...))
|
|
#
|
|
# Write entry for detail section in file table.
|
|
#
|
|
|
|
sub write_file_table_detail_entry(*$@)
|
|
{
|
|
my ($handle, $test, @entries) = @_;
|
|
my $entry;
|
|
|
|
if ($test eq "") {
|
|
$test = "<span style=\"font-style:italic\"><unnamed></span>";
|
|
} elsif ($test =~ /^(.*),diff$/) {
|
|
$test = $1." (converted)";
|
|
}
|
|
# Testname
|
|
write_html($handle, <<END_OF_HTML);
|
|
<tr>
|
|
<td class="testName" colspan=2>$test</td>
|
|
END_OF_HTML
|
|
# Test data
|
|
foreach $entry (@entries) {
|
|
my ($found, $hit) = @{$entry};
|
|
my $rate = rate($hit, $found, " %");
|
|
|
|
write_html($handle, <<END_OF_HTML);
|
|
<td class="testPer">$rate</td>
|
|
<td class="testNum">$hit / $found</td>
|
|
END_OF_HTML
|
|
}
|
|
|
|
write_html($handle, <<END_OF_HTML);
|
|
</tr>
|
|
|
|
END_OF_HTML
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# write_file_table_epilog(filehandle)
|
|
#
|
|
# Write end of file table HTML code.
|
|
#
|
|
|
|
sub write_file_table_epilog(*)
|
|
{
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
</table>
|
|
</center>
|
|
<br>
|
|
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# write_test_table_prolog(filehandle, table_heading)
|
|
#
|
|
# Write heading for test case description table.
|
|
#
|
|
|
|
sub write_test_table_prolog(*$)
|
|
{
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
<center>
|
|
<table width="80%" cellpadding=2 cellspacing=1 border=0>
|
|
|
|
<tr>
|
|
<td><br></td>
|
|
</tr>
|
|
|
|
<tr>
|
|
<td class="tableHead">$_[1]</td>
|
|
</tr>
|
|
|
|
<tr>
|
|
<td class="testDescription">
|
|
<dl>
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# write_test_table_entry(filehandle, test_name, test_description)
|
|
#
|
|
# Write entry for the test table.
|
|
#
|
|
|
|
sub write_test_table_entry(*$$)
|
|
{
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
<dt>$_[1]<a name="$_[1]"> </a></dt>
|
|
<dd>$_[2]<br><br></dd>
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# write_test_table_epilog(filehandle)
|
|
#
|
|
# Write end of test description table HTML code.
|
|
#
|
|
|
|
sub write_test_table_epilog(*)
|
|
{
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
</dl>
|
|
</td>
|
|
</tr>
|
|
</table>
|
|
</center>
|
|
<br>
|
|
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
sub fmt_centered($$)
|
|
{
|
|
my ($width, $text) = @_;
|
|
my $w0 = length($text);
|
|
my $w1 = int(($width - $w0) / 2);
|
|
my $w2 = $width - $w0 - $w1;
|
|
|
|
return (" "x$w1).$text.(" "x$w2);
|
|
}
|
|
|
|
|
|
#
|
|
# write_source_prolog(filehandle)
|
|
#
|
|
# Write start of source code table.
|
|
#
|
|
|
|
sub write_source_prolog(*)
|
|
{
|
|
my $lineno_heading = " ";
|
|
my $branch_heading = "";
|
|
my $line_heading = fmt_centered($line_field_width, "Line data");
|
|
my $source_heading = " Source code";
|
|
|
|
if ($br_coverage) {
|
|
$branch_heading = fmt_centered($br_field_width, "Branch data").
|
|
" ";
|
|
}
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
<table cellpadding=0 cellspacing=0 border=0>
|
|
<tr>
|
|
<td><br></td>
|
|
</tr>
|
|
<tr>
|
|
<td>
|
|
<pre class="sourceHeading">${lineno_heading}${branch_heading}${line_heading} ${source_heading}</pre>
|
|
<pre class="source">
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# get_branch_blocks(brdata)
|
|
#
|
|
# Group branches that belong to the same basic block.
|
|
#
|
|
# Returns: [block1, block2, ...]
|
|
# block: [branch1, branch2, ...]
|
|
# branch: [block_num, branch_num, taken_count, text_length, open, close]
|
|
#
|
|
|
|
sub get_branch_blocks($)
|
|
{
|
|
my ($brdata) = @_;
|
|
my $last_block_num;
|
|
my $block = [];
|
|
my @blocks;
|
|
my $i;
|
|
my $num = br_ivec_len($brdata);
|
|
|
|
# Group branches
|
|
for ($i = 0; $i < $num; $i++) {
|
|
my ($block_num, $branch, $taken) = br_ivec_get($brdata, $i);
|
|
my $br;
|
|
|
|
if (defined($last_block_num) && $block_num != $last_block_num) {
|
|
push(@blocks, $block);
|
|
$block = [];
|
|
}
|
|
$br = [$block_num, $branch, $taken, 3, 0, 0];
|
|
push(@{$block}, $br);
|
|
$last_block_num = $block_num;
|
|
}
|
|
push(@blocks, $block) if (scalar(@{$block}) > 0);
|
|
|
|
# Add braces to first and last branch in group
|
|
foreach $block (@blocks) {
|
|
$block->[0]->[$BR_OPEN] = 1;
|
|
$block->[0]->[$BR_LEN]++;
|
|
$block->[scalar(@{$block}) - 1]->[$BR_CLOSE] = 1;
|
|
$block->[scalar(@{$block}) - 1]->[$BR_LEN]++;
|
|
}
|
|
|
|
return @blocks;
|
|
}
|
|
|
|
#
|
|
# get_block_len(block)
|
|
#
|
|
# Calculate total text length of all branches in a block of branches.
|
|
#
|
|
|
|
sub get_block_len($)
|
|
{
|
|
my ($block) = @_;
|
|
my $len = 0;
|
|
my $branch;
|
|
|
|
foreach $branch (@{$block}) {
|
|
$len += $branch->[$BR_LEN];
|
|
}
|
|
|
|
return $len;
|
|
}
|
|
|
|
|
|
#
|
|
# get_branch_html(brdata)
|
|
#
|
|
# Return a list of HTML lines which represent the specified branch coverage
|
|
# data in source code view.
|
|
#
|
|
|
|
sub get_branch_html($)
|
|
{
|
|
my ($brdata) = @_;
|
|
my @blocks = get_branch_blocks($brdata);
|
|
my $block;
|
|
my $branch;
|
|
my $line_len = 0;
|
|
my $line = []; # [branch2|" ", branch|" ", ...]
|
|
my @lines; # [line1, line2, ...]
|
|
my @result;
|
|
|
|
# Distribute blocks to lines
|
|
foreach $block (@blocks) {
|
|
my $block_len = get_block_len($block);
|
|
|
|
# Does this block fit into the current line?
|
|
if ($line_len + $block_len <= $br_field_width) {
|
|
# Add it
|
|
$line_len += $block_len;
|
|
push(@{$line}, @{$block});
|
|
next;
|
|
} elsif ($block_len <= $br_field_width) {
|
|
# It would fit if the line was empty - add it to new
|
|
# line
|
|
push(@lines, $line);
|
|
$line_len = $block_len;
|
|
$line = [ @{$block} ];
|
|
next;
|
|
}
|
|
# Split the block into several lines
|
|
foreach $branch (@{$block}) {
|
|
if ($line_len + $branch->[$BR_LEN] >= $br_field_width) {
|
|
# Start a new line
|
|
if (($line_len + 1 <= $br_field_width) &&
|
|
scalar(@{$line}) > 0 &&
|
|
!$line->[scalar(@$line) - 1]->[$BR_CLOSE]) {
|
|
# Try to align branch symbols to be in
|
|
# one # row
|
|
push(@{$line}, " ");
|
|
}
|
|
push(@lines, $line);
|
|
$line_len = 0;
|
|
$line = [];
|
|
}
|
|
push(@{$line}, $branch);
|
|
$line_len += $branch->[$BR_LEN];
|
|
}
|
|
}
|
|
push(@lines, $line);
|
|
|
|
# Convert to HTML
|
|
foreach $line (@lines) {
|
|
my $current = "";
|
|
my $current_len = 0;
|
|
|
|
foreach $branch (@$line) {
|
|
# Skip alignment space
|
|
if ($branch eq " ") {
|
|
$current .= " ";
|
|
$current_len++;
|
|
next;
|
|
}
|
|
|
|
my ($block_num, $br_num, $taken, $len, $open, $close) =
|
|
@{$branch};
|
|
my $class;
|
|
my $title;
|
|
my $text;
|
|
|
|
if ($taken eq '-') {
|
|
$class = "branchNoExec";
|
|
$text = " # ";
|
|
$title = "Branch $br_num was not executed";
|
|
} elsif ($taken == 0) {
|
|
$class = "branchNoCov";
|
|
$text = " - ";
|
|
$title = "Branch $br_num was not taken";
|
|
} else {
|
|
$class = "branchCov";
|
|
$text = " + ";
|
|
$title = "Branch $br_num was taken $taken ".
|
|
"time";
|
|
$title .= "s" if ($taken > 1);
|
|
}
|
|
$current .= "[" if ($open);
|
|
$current .= "<span class=\"$class\" title=\"$title\">";
|
|
$current .= $text."</span>";
|
|
$current .= "]" if ($close);
|
|
$current_len += $len;
|
|
}
|
|
|
|
# Right-align result text
|
|
if ($current_len < $br_field_width) {
|
|
$current = (" "x($br_field_width - $current_len)).
|
|
$current;
|
|
}
|
|
push(@result, $current);
|
|
}
|
|
|
|
return @result;
|
|
}
|
|
|
|
|
|
#
|
|
# format_count(count, width)
|
|
#
|
|
# Return a right-aligned representation of count that fits in width characters.
|
|
#
|
|
|
|
sub format_count($$)
|
|
{
|
|
my ($count, $width) = @_;
|
|
my $result;
|
|
my $exp;
|
|
|
|
$result = sprintf("%*.0f", $width, $count);
|
|
while (length($result) > $width) {
|
|
last if ($count < 10);
|
|
$exp++;
|
|
$count = int($count/10);
|
|
$result = sprintf("%*s", $width, ">$count*10^$exp");
|
|
}
|
|
return $result;
|
|
}
|
|
|
|
#
|
|
# write_source_line(filehandle, line_num, source, hit_count, converted,
|
|
# brdata, add_anchor)
|
|
#
|
|
# Write formatted source code line. Return a line in a format as needed
|
|
# by gen_png()
|
|
#
|
|
|
|
sub write_source_line(*$$$$$$)
|
|
{
|
|
my ($handle, $line, $source, $count, $converted, $brdata,
|
|
$add_anchor) = @_;
|
|
my $source_format;
|
|
my $count_format;
|
|
my $result;
|
|
my $anchor_start = "";
|
|
my $anchor_end = "";
|
|
my $count_field_width = $line_field_width - 1;
|
|
my @br_html;
|
|
my $html;
|
|
|
|
# Get branch HTML data for this line
|
|
@br_html = get_branch_html($brdata) if ($br_coverage);
|
|
|
|
if (!defined($count)) {
|
|
$result = "";
|
|
$source_format = "";
|
|
$count_format = " "x$count_field_width;
|
|
}
|
|
elsif ($count == 0) {
|
|
$result = $count;
|
|
$source_format = '<span class="lineNoCov">';
|
|
$count_format = format_count($count, $count_field_width);
|
|
}
|
|
elsif ($converted && defined($highlight)) {
|
|
$result = "*".$count;
|
|
$source_format = '<span class="lineDiffCov">';
|
|
$count_format = format_count($count, $count_field_width);
|
|
}
|
|
else {
|
|
$result = $count;
|
|
$source_format = '<span class="lineCov">';
|
|
$count_format = format_count($count, $count_field_width);
|
|
}
|
|
$result .= ":".$source;
|
|
|
|
# Write out a line number navigation anchor every $nav_resolution
|
|
# lines if necessary
|
|
if ($add_anchor)
|
|
{
|
|
$anchor_start = "<a name=\"$_[1]\">";
|
|
$anchor_end = "</a>";
|
|
}
|
|
|
|
|
|
# *************************************************************
|
|
|
|
$html = $anchor_start;
|
|
$html .= "<span class=\"lineNum\">".sprintf("%8d", $line)." </span>";
|
|
$html .= shift(@br_html).":" if ($br_coverage);
|
|
$html .= "$source_format$count_format : ";
|
|
$html .= escape_html($source);
|
|
$html .= "</span>" if ($source_format);
|
|
$html .= $anchor_end."\n";
|
|
|
|
write_html($handle, $html);
|
|
|
|
if ($br_coverage) {
|
|
# Add lines for overlong branch information
|
|
foreach (@br_html) {
|
|
write_html($handle, "<span class=\"lineNum\">".
|
|
" </span>$_\n");
|
|
}
|
|
}
|
|
# *************************************************************
|
|
|
|
return($result);
|
|
}
|
|
|
|
|
|
#
|
|
# write_source_epilog(filehandle)
|
|
#
|
|
# Write end of source code table.
|
|
#
|
|
|
|
sub write_source_epilog(*)
|
|
{
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
</pre>
|
|
</td>
|
|
</tr>
|
|
</table>
|
|
<br>
|
|
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# write_html_epilog(filehandle, base_dir[, break_frames])
|
|
#
|
|
# Write HTML page footer to FILEHANDLE. BREAK_FRAMES should be set when
|
|
# this page is embedded in a frameset, clicking the URL link will then
|
|
# break this frameset.
|
|
#
|
|
|
|
sub write_html_epilog(*$;$)
|
|
{
|
|
my $basedir = $_[1];
|
|
my $break_code = "";
|
|
my $epilog;
|
|
|
|
if (defined($_[2]))
|
|
{
|
|
$break_code = " target=\"_parent\"";
|
|
}
|
|
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
<table width="100%" border=0 cellspacing=0 cellpadding=0>
|
|
<tr><td class="ruler"><img src="$_[1]glass.png" width=3 height=3 alt=""></td></tr>
|
|
<tr><td class="versionInfo">Generated by: <a href="$lcov_url"$break_code>$lcov_version</a></td></tr>
|
|
</table>
|
|
<br>
|
|
END_OF_HTML
|
|
;
|
|
|
|
$epilog = $html_epilog;
|
|
$epilog =~ s/\@basedir\@/$basedir/g;
|
|
|
|
write_html($_[0], $epilog);
|
|
}
|
|
|
|
|
|
#
|
|
# write_frameset(filehandle, basedir, basename, pagetitle)
|
|
#
|
|
#
|
|
|
|
sub write_frameset(*$$$)
|
|
{
|
|
my $frame_width = $overview_width + 40;
|
|
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Frameset//EN">
|
|
|
|
<html lang="en">
|
|
|
|
<head>
|
|
<meta http-equiv="Content-Type" content="text/html; charset=$charset">
|
|
<title>$_[3]</title>
|
|
<link rel="stylesheet" type="text/css" href="$_[1]gcov.css">
|
|
</head>
|
|
|
|
<frameset cols="$frame_width,*">
|
|
<frame src="$_[2].gcov.overview.$html_ext" name="overview">
|
|
<frame src="$_[2].gcov.$html_ext" name="source">
|
|
<noframes>
|
|
<center>Frames not supported by your browser!<br></center>
|
|
</noframes>
|
|
</frameset>
|
|
|
|
</html>
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# sub write_overview_line(filehandle, basename, line, link)
|
|
#
|
|
#
|
|
|
|
sub write_overview_line(*$$$)
|
|
{
|
|
my $y1 = $_[2] - 1;
|
|
my $y2 = $y1 + $nav_resolution - 1;
|
|
my $x2 = $overview_width - 1;
|
|
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
<area shape="rect" coords="0,$y1,$x2,$y2" href="$_[1].gcov.$html_ext#$_[3]" target="source" alt="overview">
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
#
|
|
# write_overview(filehandle, basedir, basename, pagetitle, lines)
|
|
#
|
|
#
|
|
|
|
sub write_overview(*$$$$)
|
|
{
|
|
my $index;
|
|
my $max_line = $_[4] - 1;
|
|
my $offset;
|
|
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
|
|
|
|
<html lang="en">
|
|
|
|
<head>
|
|
<title>$_[3]</title>
|
|
<meta http-equiv="Content-Type" content="text/html; charset=$charset">
|
|
<link rel="stylesheet" type="text/css" href="$_[1]gcov.css">
|
|
</head>
|
|
|
|
<body>
|
|
<map name="overview">
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
|
|
# Make $offset the next higher multiple of $nav_resolution
|
|
$offset = ($nav_offset + $nav_resolution - 1) / $nav_resolution;
|
|
$offset = sprintf("%d", $offset ) * $nav_resolution;
|
|
|
|
# Create image map for overview image
|
|
for ($index = 1; $index <= $_[4]; $index += $nav_resolution)
|
|
{
|
|
# Enforce nav_offset
|
|
if ($index < $offset + 1)
|
|
{
|
|
write_overview_line($_[0], $_[2], $index, 1);
|
|
}
|
|
else
|
|
{
|
|
write_overview_line($_[0], $_[2], $index, $index - $offset);
|
|
}
|
|
}
|
|
|
|
# *************************************************************
|
|
|
|
write_html($_[0], <<END_OF_HTML)
|
|
</map>
|
|
|
|
<center>
|
|
<a href="$_[2].gcov.$html_ext#top" target="source">Top</a><br><br>
|
|
<img src="$_[2].gcov.png" width=$overview_width height=$max_line alt="Overview" border=0 usemap="#overview">
|
|
</center>
|
|
</body>
|
|
</html>
|
|
END_OF_HTML
|
|
;
|
|
|
|
# *************************************************************
|
|
}
|
|
|
|
|
|
sub max($$)
|
|
{
|
|
my ($a, $b) = @_;
|
|
|
|
return $a if ($a > $b);
|
|
return $b;
|
|
}
|
|
|
|
|
|
#
|
|
# write_header(filehandle, type, trunc_file_name, rel_file_name, lines_found,
|
|
# lines_hit, funcs_found, funcs_hit, sort_type)
|
|
#
|
|
# Write a complete standard page header. TYPE may be (0, 1, 2, 3, 4)
|
|
# corresponding to (directory view header, file view header, source view
|
|
# header, test case description header, function view header)
|
|
#
|
|
|
|
sub write_header(*$$$$$$$$$$)
|
|
{
|
|
local *HTML_HANDLE = $_[0];
|
|
my $type = $_[1];
|
|
my $trunc_name = $_[2];
|
|
my $rel_filename = $_[3];
|
|
my $lines_found = $_[4];
|
|
my $lines_hit = $_[5];
|
|
my $fn_found = $_[6];
|
|
my $fn_hit = $_[7];
|
|
my $br_found = $_[8];
|
|
my $br_hit = $_[9];
|
|
my $sort_type = $_[10];
|
|
my $base_dir;
|
|
my $view;
|
|
my $test;
|
|
my $base_name;
|
|
my $style;
|
|
my $rate;
|
|
my @row_left;
|
|
my @row_right;
|
|
my $num_rows;
|
|
my $i;
|
|
my $esc_trunc_name = escape_html($trunc_name);
|
|
|
|
$base_name = basename($rel_filename);
|
|
|
|
# Prepare text for "current view" field
|
|
if ($type == $HDR_DIR)
|
|
{
|
|
# Main overview
|
|
$base_dir = "";
|
|
$view = $overview_title;
|
|
}
|
|
elsif ($type == $HDR_FILE)
|
|
{
|
|
# Directory overview
|
|
$base_dir = get_relative_base_path($rel_filename);
|
|
$view = "<a href=\"$base_dir"."index.$html_ext\">".
|
|
"$overview_title</a> - $esc_trunc_name";
|
|
}
|
|
elsif ($type == $HDR_SOURCE || $type == $HDR_FUNC)
|
|
{
|
|
# File view
|
|
my $dir_name = dirname($rel_filename);
|
|
my $esc_base_name = escape_html($base_name);
|
|
my $esc_dir_name = escape_html($dir_name);
|
|
|
|
$base_dir = get_relative_base_path($dir_name);
|
|
if ($frames)
|
|
{
|
|
# Need to break frameset when clicking any of these
|
|
# links
|
|
$view = "<a href=\"$base_dir"."index.$html_ext\" ".
|
|
"target=\"_parent\">$overview_title</a> - ".
|
|
"<a href=\"index.$html_ext\" target=\"_parent\">".
|
|
"$esc_dir_name</a> - $esc_base_name";
|
|
}
|
|
else
|
|
{
|
|
$view = "<a href=\"$base_dir"."index.$html_ext\">".
|
|
"$overview_title</a> - ".
|
|
"<a href=\"index.$html_ext\">".
|
|
"$esc_dir_name</a> - $esc_base_name";
|
|
}
|
|
|
|
# Add function suffix
|
|
if ($func_coverage) {
|
|
$view .= "<span style=\"font-size: 80%;\">";
|
|
if ($type == $HDR_SOURCE) {
|
|
$view .= " (source / <a href=\"$base_name.func.$html_ext\">functions</a>)";
|
|
} elsif ($type == $HDR_FUNC) {
|
|
$view .= " (<a href=\"$base_name.gcov.$html_ext\">source</a> / functions)";
|
|
}
|
|
$view .= "</span>";
|
|
}
|
|
}
|
|
elsif ($type == $HDR_TESTDESC)
|
|
{
|
|
# Test description header
|
|
$base_dir = "";
|
|
$view = "<a href=\"$base_dir"."index.$html_ext\">".
|
|
"$overview_title</a> - test case descriptions";
|
|
}
|
|
|
|
# Prepare text for "test" field
|
|
$test = escape_html($test_title);
|
|
|
|
# Append link to test description page if available
|
|
if (%test_description && ($type != $HDR_TESTDESC))
|
|
{
|
|
if ($frames && ($type == $HDR_SOURCE || $type == $HDR_FUNC))
|
|
{
|
|
# Need to break frameset when clicking this link
|
|
$test .= " ( <span style=\"font-size:80%;\">".
|
|
"<a href=\"$base_dir".
|
|
"descriptions.$html_ext\" target=\"_parent\">".
|
|
"view descriptions</a></span> )";
|
|
}
|
|
else
|
|
{
|
|
$test .= " ( <span style=\"font-size:80%;\">".
|
|
"<a href=\"$base_dir".
|
|
"descriptions.$html_ext\">".
|
|
"view descriptions</a></span> )";
|
|
}
|
|
}
|
|
|
|
# Write header
|
|
write_header_prolog(*HTML_HANDLE, $base_dir);
|
|
|
|
# Left row
|
|
push(@row_left, [[ "10%", "headerItem", "Current view:" ],
|
|
[ "35%", "headerValue", $view ]]);
|
|
push(@row_left, [[undef, "headerItem", "Test:"],
|
|
[undef, "headerValue", $test]]);
|
|
push(@row_left, [[undef, "headerItem", "Date:"],
|
|
[undef, "headerValue", $date]]);
|
|
|
|
# Right row
|
|
if ($legend && ($type == $HDR_SOURCE || $type == $HDR_FUNC)) {
|
|
my $text = <<END_OF_HTML;
|
|
Lines:
|
|
<span class="coverLegendCov">hit</span>
|
|
<span class="coverLegendNoCov">not hit</span>
|
|
END_OF_HTML
|
|
if ($br_coverage) {
|
|
$text .= <<END_OF_HTML;
|
|
| Branches:
|
|
<span class="coverLegendCov">+</span> taken
|
|
<span class="coverLegendNoCov">-</span> not taken
|
|
<span class="coverLegendNoCov">#</span> not executed
|
|
END_OF_HTML
|
|
}
|
|
push(@row_left, [[undef, "headerItem", "Legend:"],
|
|
[undef, "headerValueLeg", $text]]);
|
|
} elsif ($legend && ($type != $HDR_TESTDESC)) {
|
|
my $text = <<END_OF_HTML;
|
|
Rating:
|
|
<span class="coverLegendCovLo" title="Coverage rates below $med_limit % are classified as low">low: < $med_limit %</span>
|
|
<span class="coverLegendCovMed" title="Coverage rates between $med_limit % and $hi_limit % are classified as medium">medium: >= $med_limit %</span>
|
|
<span class="coverLegendCovHi" title="Coverage rates of $hi_limit % and more are classified as high">high: >= $hi_limit %</span>
|
|
END_OF_HTML
|
|
push(@row_left, [[undef, "headerItem", "Legend:"],
|
|
[undef, "headerValueLeg", $text]]);
|
|
}
|
|
if ($type == $HDR_TESTDESC) {
|
|
push(@row_right, [[ "55%" ]]);
|
|
} else {
|
|
push(@row_right, [["15%", undef, undef ],
|
|
["10%", "headerCovTableHead", "Hit" ],
|
|
["10%", "headerCovTableHead", "Total" ],
|
|
["15%", "headerCovTableHead", "Coverage"]]);
|
|
}
|
|
# Line coverage
|
|
$style = $rate_name[classify_rate($lines_found, $lines_hit,
|
|
$med_limit, $hi_limit)];
|
|
$rate = rate($lines_hit, $lines_found, " %");
|
|
push(@row_right, [[undef, "headerItem", "Lines:"],
|
|
[undef, "headerCovTableEntry", $lines_hit],
|
|
[undef, "headerCovTableEntry", $lines_found],
|
|
[undef, "headerCovTableEntry$style", $rate]])
|
|
if ($type != $HDR_TESTDESC);
|
|
# Function coverage
|
|
if ($func_coverage) {
|
|
$style = $rate_name[classify_rate($fn_found, $fn_hit,
|
|
$fn_med_limit, $fn_hi_limit)];
|
|
$rate = rate($fn_hit, $fn_found, " %");
|
|
push(@row_right, [[undef, "headerItem", "Functions:"],
|
|
[undef, "headerCovTableEntry", $fn_hit],
|
|
[undef, "headerCovTableEntry", $fn_found],
|
|
[undef, "headerCovTableEntry$style", $rate]])
|
|
if ($type != $HDR_TESTDESC);
|
|
}
|
|
# Branch coverage
|
|
if ($br_coverage) {
|
|
$style = $rate_name[classify_rate($br_found, $br_hit,
|
|
$br_med_limit, $br_hi_limit)];
|
|
$rate = rate($br_hit, $br_found, " %");
|
|
push(@row_right, [[undef, "headerItem", "Branches:"],
|
|
[undef, "headerCovTableEntry", $br_hit],
|
|
[undef, "headerCovTableEntry", $br_found],
|
|
[undef, "headerCovTableEntry$style", $rate]])
|
|
if ($type != $HDR_TESTDESC);
|
|
}
|
|
|
|
# Print rows
|
|
$num_rows = max(scalar(@row_left), scalar(@row_right));
|
|
for ($i = 0; $i < $num_rows; $i++) {
|
|
my $left = $row_left[$i];
|
|
my $right = $row_right[$i];
|
|
|
|
if (!defined($left)) {
|
|
$left = [[undef, undef, undef], [undef, undef, undef]];
|
|
}
|
|
if (!defined($right)) {
|
|
$right = [];
|
|
}
|
|
write_header_line(*HTML_HANDLE, @{$left},
|
|
[ $i == 0 ? "5%" : undef, undef, undef],
|
|
@{$right});
|
|
}
|
|
|
|
# Fourth line
|
|
write_header_epilog(*HTML_HANDLE, $base_dir);
|
|
}
|
|
|
|
|
|
#
|
|
# get_sorted_keys(hash_ref, sort_type)
|
|
#
|
|
|
|
sub get_sorted_keys($$)
|
|
{
|
|
my ($hash, $type) = @_;
|
|
|
|
if ($type == $SORT_FILE) {
|
|
# Sort by name
|
|
return sort(keys(%{$hash}));
|
|
} elsif ($type == $SORT_LINE) {
|
|
# Sort by line coverage
|
|
return sort({$hash->{$a}[7] <=> $hash->{$b}[7]} keys(%{$hash}));
|
|
} elsif ($type == $SORT_FUNC) {
|
|
# Sort by function coverage;
|
|
return sort({$hash->{$a}[8] <=> $hash->{$b}[8]} keys(%{$hash}));
|
|
} elsif ($type == $SORT_BRANCH) {
|
|
# Sort by br coverage;
|
|
return sort({$hash->{$a}[9] <=> $hash->{$b}[9]} keys(%{$hash}));
|
|
}
|
|
}
|
|
|
|
sub get_sort_code($$$)
|
|
{
|
|
my ($link, $alt, $base) = @_;
|
|
my $png;
|
|
my $link_start;
|
|
my $link_end;
|
|
|
|
if (!defined($link)) {
|
|
$png = "glass.png";
|
|
$link_start = "";
|
|
$link_end = "";
|
|
} else {
|
|
$png = "updown.png";
|
|
$link_start = '<a href="'.$link.'">';
|
|
$link_end = "</a>";
|
|
}
|
|
|
|
return ' <span class="tableHeadSort">'.$link_start.
|
|
'<img src="'.$base.$png.'" width=10 height=14 '.
|
|
'alt="'.$alt.'" title="'.$alt.'" border=0>'.$link_end.'</span>';
|
|
}
|
|
|
|
sub get_file_code($$$$)
|
|
{
|
|
my ($type, $text, $sort_button, $base) = @_;
|
|
my $result = $text;
|
|
my $link;
|
|
|
|
if ($sort_button) {
|
|
if ($type == $HEAD_NO_DETAIL) {
|
|
$link = "index.$html_ext";
|
|
} else {
|
|
$link = "index-detail.$html_ext";
|
|
}
|
|
}
|
|
$result .= get_sort_code($link, "Sort by name", $base);
|
|
|
|
return $result;
|
|
}
|
|
|
|
sub get_line_code($$$$$)
|
|
{
|
|
my ($type, $sort_type, $text, $sort_button, $base) = @_;
|
|
my $result = $text;
|
|
my $sort_link;
|
|
|
|
if ($type == $HEAD_NO_DETAIL) {
|
|
# Just text
|
|
if ($sort_button) {
|
|
$sort_link = "index-sort-l.$html_ext";
|
|
}
|
|
} elsif ($type == $HEAD_DETAIL_HIDDEN) {
|
|
# Text + link to detail view
|
|
$result .= ' ( <a class="detail" href="index-detail'.
|
|
$fileview_sortname[$sort_type].'.'.$html_ext.
|
|
'">show details</a> )';
|
|
if ($sort_button) {
|
|
$sort_link = "index-sort-l.$html_ext";
|
|
}
|
|
} else {
|
|
# Text + link to standard view
|
|
$result .= ' ( <a class="detail" href="index'.
|
|
$fileview_sortname[$sort_type].'.'.$html_ext.
|
|
'">hide details</a> )';
|
|
if ($sort_button) {
|
|
$sort_link = "index-detail-sort-l.$html_ext";
|
|
}
|
|
}
|
|
# Add sort button
|
|
$result .= get_sort_code($sort_link, "Sort by line coverage", $base);
|
|
|
|
return $result;
|
|
}
|
|
|
|
sub get_func_code($$$$)
|
|
{
|
|
my ($type, $text, $sort_button, $base) = @_;
|
|
my $result = $text;
|
|
my $link;
|
|
|
|
if ($sort_button) {
|
|
if ($type == $HEAD_NO_DETAIL) {
|
|
$link = "index-sort-f.$html_ext";
|
|
} else {
|
|
$link = "index-detail-sort-f.$html_ext";
|
|
}
|
|
}
|
|
$result .= get_sort_code($link, "Sort by function coverage", $base);
|
|
return $result;
|
|
}
|
|
|
|
sub get_br_code($$$$)
|
|
{
|
|
my ($type, $text, $sort_button, $base) = @_;
|
|
my $result = $text;
|
|
my $link;
|
|
|
|
if ($sort_button) {
|
|
if ($type == $HEAD_NO_DETAIL) {
|
|
$link = "index-sort-b.$html_ext";
|
|
} else {
|
|
$link = "index-detail-sort-b.$html_ext";
|
|
}
|
|
}
|
|
$result .= get_sort_code($link, "Sort by branch coverage", $base);
|
|
return $result;
|
|
}
|
|
|
|
#
|
|
# write_file_table(filehandle, base_dir, overview, testhash, testfnchash,
|
|
# testbrhash, fileview, sort_type)
|
|
#
|
|
# Write a complete file table. OVERVIEW is a reference to a hash containing
|
|
# the following mapping:
|
|
#
|
|
# filename -> "lines_found,lines_hit,funcs_found,funcs_hit,page_link,
|
|
# func_link"
|
|
#
|
|
# TESTHASH is a reference to the following hash:
|
|
#
|
|
# filename -> \%testdata
|
|
# %testdata: name of test affecting this file -> \%testcount
|
|
# %testcount: line number -> execution count for a single test
|
|
#
|
|
# Heading of first column is "Filename" if FILEVIEW is true, "Directory name"
|
|
# otherwise.
|
|
#
|
|
|
|
sub write_file_table(*$$$$$$$)
|
|
{
|
|
local *HTML_HANDLE = $_[0];
|
|
my $base_dir = $_[1];
|
|
my $overview = $_[2];
|
|
my $testhash = $_[3];
|
|
my $testfnchash = $_[4];
|
|
my $testbrhash = $_[5];
|
|
my $fileview = $_[6];
|
|
my $sort_type = $_[7];
|
|
my $filename;
|
|
my $bar_graph;
|
|
my $hit;
|
|
my $found;
|
|
my $fn_found;
|
|
my $fn_hit;
|
|
my $br_found;
|
|
my $br_hit;
|
|
my $page_link;
|
|
my $testname;
|
|
my $testdata;
|
|
my $testfncdata;
|
|
my $testbrdata;
|
|
my %affecting_tests;
|
|
my $line_code = "";
|
|
my $func_code;
|
|
my $br_code;
|
|
my $file_code;
|
|
my @head_columns;
|
|
|
|
# Determine HTML code for column headings
|
|
if (($base_dir ne "") && $show_details)
|
|
{
|
|
my $detailed = keys(%{$testhash});
|
|
|
|
$file_code = get_file_code($detailed ? $HEAD_DETAIL_HIDDEN :
|
|
$HEAD_NO_DETAIL,
|
|
$fileview ? "Filename" : "Directory",
|
|
$sort && $sort_type != $SORT_FILE,
|
|
$base_dir);
|
|
$line_code = get_line_code($detailed ? $HEAD_DETAIL_SHOWN :
|
|
$HEAD_DETAIL_HIDDEN,
|
|
$sort_type,
|
|
"Line Coverage",
|
|
$sort && $sort_type != $SORT_LINE,
|
|
$base_dir);
|
|
$func_code = get_func_code($detailed ? $HEAD_DETAIL_HIDDEN :
|
|
$HEAD_NO_DETAIL,
|
|
"Functions",
|
|
$sort && $sort_type != $SORT_FUNC,
|
|
$base_dir);
|
|
$br_code = get_br_code($detailed ? $HEAD_DETAIL_HIDDEN :
|
|
$HEAD_NO_DETAIL,
|
|
"Branches",
|
|
$sort && $sort_type != $SORT_BRANCH,
|
|
$base_dir);
|
|
} else {
|
|
$file_code = get_file_code($HEAD_NO_DETAIL,
|
|
$fileview ? "Filename" : "Directory",
|
|
$sort && $sort_type != $SORT_FILE,
|
|
$base_dir);
|
|
$line_code = get_line_code($HEAD_NO_DETAIL, $sort_type, "Line Coverage",
|
|
$sort && $sort_type != $SORT_LINE,
|
|
$base_dir);
|
|
$func_code = get_func_code($HEAD_NO_DETAIL, "Functions",
|
|
$sort && $sort_type != $SORT_FUNC,
|
|
$base_dir);
|
|
$br_code = get_br_code($HEAD_NO_DETAIL, "Branches",
|
|
$sort && $sort_type != $SORT_BRANCH,
|
|
$base_dir);
|
|
}
|
|
push(@head_columns, [ $line_code, 3 ]);
|
|
push(@head_columns, [ $func_code, 2]) if ($func_coverage);
|
|
push(@head_columns, [ $br_code, 2]) if ($br_coverage);
|
|
|
|
write_file_table_prolog(*HTML_HANDLE, $file_code, @head_columns);
|
|
|
|
foreach $filename (get_sorted_keys($overview, $sort_type))
|
|
{
|
|
my @columns;
|
|
($found, $hit, $fn_found, $fn_hit, $br_found, $br_hit,
|
|
$page_link) = @{$overview->{$filename}};
|
|
|
|
# Line coverage
|
|
push(@columns, [$found, $hit, $med_limit, $hi_limit, 1]);
|
|
# Function coverage
|
|
if ($func_coverage) {
|
|
push(@columns, [$fn_found, $fn_hit, $fn_med_limit,
|
|
$fn_hi_limit, 0]);
|
|
}
|
|
# Branch coverage
|
|
if ($br_coverage) {
|
|
push(@columns, [$br_found, $br_hit, $br_med_limit,
|
|
$br_hi_limit, 0]);
|
|
}
|
|
write_file_table_entry(*HTML_HANDLE, $base_dir, $filename,
|
|
$page_link, @columns);
|
|
|
|
$testdata = $testhash->{$filename};
|
|
$testfncdata = $testfnchash->{$filename};
|
|
$testbrdata = $testbrhash->{$filename};
|
|
|
|
# Check whether we should write test specific coverage
|
|
# as well
|
|
if (!($show_details && $testdata)) { next; }
|
|
|
|
# Filter out those tests that actually affect this file
|
|
%affecting_tests = %{ get_affecting_tests($testdata,
|
|
$testfncdata, $testbrdata) };
|
|
|
|
# Does any of the tests affect this file at all?
|
|
if (!%affecting_tests) { next; }
|
|
|
|
foreach $testname (keys(%affecting_tests))
|
|
{
|
|
my @results;
|
|
($found, $hit, $fn_found, $fn_hit, $br_found, $br_hit) =
|
|
split(",", $affecting_tests{$testname});
|
|
|
|
# Insert link to description of available
|
|
if ($test_description{$testname})
|
|
{
|
|
$testname = "<a href=\"$base_dir".
|
|
"descriptions.$html_ext#$testname\">".
|
|
"$testname</a>";
|
|
}
|
|
|
|
push(@results, [$found, $hit]);
|
|
push(@results, [$fn_found, $fn_hit]) if ($func_coverage);
|
|
push(@results, [$br_found, $br_hit]) if ($br_coverage);
|
|
write_file_table_detail_entry(*HTML_HANDLE, $testname,
|
|
@results);
|
|
}
|
|
}
|
|
|
|
write_file_table_epilog(*HTML_HANDLE);
|
|
}
|
|
|
|
|
|
#
|
|
# get_found_and_hit(hash)
|
|
#
|
|
# Return the count for entries (found) and entries with an execution count
|
|
# greater than zero (hit) in a hash (linenumber -> execution count) as
|
|
# a list (found, hit)
|
|
#
|
|
|
|
sub get_found_and_hit($)
|
|
{
|
|
my %hash = %{$_[0]};
|
|
my $found = 0;
|
|
my $hit = 0;
|
|
|
|
# Calculate sum
|
|
$found = 0;
|
|
$hit = 0;
|
|
|
|
foreach (keys(%hash))
|
|
{
|
|
$found++;
|
|
if ($hash{$_}>0) { $hit++; }
|
|
}
|
|
|
|
return ($found, $hit);
|
|
}
|
|
|
|
|
|
#
|
|
# get_func_found_and_hit(sumfnccount)
|
|
#
|
|
# Return (f_found, f_hit) for sumfnccount
|
|
#
|
|
|
|
sub get_func_found_and_hit($)
|
|
{
|
|
my ($sumfnccount) = @_;
|
|
my $function;
|
|
my $fn_found;
|
|
my $fn_hit;
|
|
|
|
$fn_found = scalar(keys(%{$sumfnccount}));
|
|
$fn_hit = 0;
|
|
foreach $function (keys(%{$sumfnccount})) {
|
|
if ($sumfnccount->{$function} > 0) {
|
|
$fn_hit++;
|
|
}
|
|
}
|
|
return ($fn_found, $fn_hit);
|
|
}
|
|
|
|
|
|
#
|
|
# br_taken_to_num(taken)
|
|
#
|
|
# Convert a branch taken value .info format to number format.
|
|
#
|
|
|
|
sub br_taken_to_num($)
|
|
{
|
|
my ($taken) = @_;
|
|
|
|
return 0 if ($taken eq '-');
|
|
return $taken + 1;
|
|
}
|
|
|
|
|
|
#
|
|
# br_num_to_taken(taken)
|
|
#
|
|
# Convert a branch taken value in number format to .info format.
|
|
#
|
|
|
|
sub br_num_to_taken($)
|
|
{
|
|
my ($taken) = @_;
|
|
|
|
return '-' if ($taken == 0);
|
|
return $taken - 1;
|
|
}
|
|
|
|
|
|
#
|
|
# br_taken_add(taken1, taken2)
|
|
#
|
|
# Return the result of taken1 + taken2 for 'branch taken' values.
|
|
#
|
|
|
|
sub br_taken_add($$)
|
|
{
|
|
my ($t1, $t2) = @_;
|
|
|
|
return $t1 if (!defined($t2));
|
|
return $t2 if (!defined($t1));
|
|
return $t1 if ($t2 eq '-');
|
|
return $t2 if ($t1 eq '-');
|
|
return $t1 + $t2;
|
|
}
|
|
|
|
|
|
#
|
|
# br_taken_sub(taken1, taken2)
|
|
#
|
|
# Return the result of taken1 - taken2 for 'branch taken' values. Return 0
|
|
# if the result would become negative.
|
|
#
|
|
|
|
sub br_taken_sub($$)
|
|
{
|
|
my ($t1, $t2) = @_;
|
|
|
|
return $t1 if (!defined($t2));
|
|
return undef if (!defined($t1));
|
|
return $t1 if ($t1 eq '-');
|
|
return $t1 if ($t2 eq '-');
|
|
return 0 if $t2 > $t1;
|
|
return $t1 - $t2;
|
|
}
|
|
|
|
|
|
#
|
|
# br_ivec_len(vector)
|
|
#
|
|
# Return the number of entries in the branch coverage vector.
|
|
#
|
|
|
|
sub br_ivec_len($)
|
|
{
|
|
my ($vec) = @_;
|
|
|
|
return 0 if (!defined($vec));
|
|
return (length($vec) * 8 / $BR_VEC_WIDTH) / $BR_VEC_ENTRIES;
|
|
}
|
|
|
|
|
|
#
|
|
# br_ivec_get(vector, number)
|
|
#
|
|
# Return an entry from the branch coverage vector.
|
|
#
|
|
|
|
sub br_ivec_get($$)
|
|
{
|
|
my ($vec, $num) = @_;
|
|
my $block;
|
|
my $branch;
|
|
my $taken;
|
|
my $offset = $num * $BR_VEC_ENTRIES;
|
|
|
|
# Retrieve data from vector
|
|
$block = vec($vec, $offset + $BR_BLOCK, $BR_VEC_WIDTH);
|
|
$branch = vec($vec, $offset + $BR_BRANCH, $BR_VEC_WIDTH);
|
|
$taken = vec($vec, $offset + $BR_TAKEN, $BR_VEC_WIDTH);
|
|
|
|
# Decode taken value from an integer
|
|
$taken = br_num_to_taken($taken);
|
|
|
|
return ($block, $branch, $taken);
|
|
}
|
|
|
|
|
|
#
|
|
# br_ivec_push(vector, block, branch, taken)
|
|
#
|
|
# Add an entry to the branch coverage vector. If an entry with the same
|
|
# branch ID already exists, add the corresponding taken values.
|
|
#
|
|
|
|
sub br_ivec_push($$$$)
|
|
{
|
|
my ($vec, $block, $branch, $taken) = @_;
|
|
my $offset;
|
|
my $num = br_ivec_len($vec);
|
|
my $i;
|
|
|
|
$vec = "" if (!defined($vec));
|
|
|
|
# Check if branch already exists in vector
|
|
for ($i = 0; $i < $num; $i++) {
|
|
my ($v_block, $v_branch, $v_taken) = br_ivec_get($vec, $i);
|
|
|
|
next if ($v_block != $block || $v_branch != $branch);
|
|
|
|
# Add taken counts
|
|
$taken = br_taken_add($taken, $v_taken);
|
|
last;
|
|
}
|
|
|
|
$offset = $i * $BR_VEC_ENTRIES;
|
|
$taken = br_taken_to_num($taken);
|
|
|
|
# Add to vector
|
|
vec($vec, $offset + $BR_BLOCK, $BR_VEC_WIDTH) = $block;
|
|
vec($vec, $offset + $BR_BRANCH, $BR_VEC_WIDTH) = $branch;
|
|
vec($vec, $offset + $BR_TAKEN, $BR_VEC_WIDTH) = $taken;
|
|
|
|
return $vec;
|
|
}
|
|
|
|
|
|
#
|
|
# get_br_found_and_hit(sumbrcount)
|
|
#
|
|
# Return (br_found, br_hit) for sumbrcount
|
|
#
|
|
|
|
sub get_br_found_and_hit($)
|
|
{
|
|
my ($sumbrcount) = @_;
|
|
my $line;
|
|
my $br_found = 0;
|
|
my $br_hit = 0;
|
|
|
|
foreach $line (keys(%{$sumbrcount})) {
|
|
my $brdata = $sumbrcount->{$line};
|
|
my $i;
|
|
my $num = br_ivec_len($brdata);
|
|
|
|
for ($i = 0; $i < $num; $i++) {
|
|
my $taken;
|
|
|
|
(undef, undef, $taken) = br_ivec_get($brdata, $i);
|
|
|
|
$br_found++;
|
|
$br_hit++ if ($taken ne "-" && $taken > 0);
|
|
}
|
|
}
|
|
|
|
return ($br_found, $br_hit);
|
|
}
|
|
|
|
|
|
#
|
|
# get_affecting_tests(testdata, testfncdata, testbrdata)
|
|
#
|
|
# HASHREF contains a mapping filename -> (linenumber -> exec count). Return
|
|
# a hash containing mapping filename -> "lines found, lines hit" for each
|
|
# filename which has a nonzero hit count.
|
|
#
|
|
|
|
sub get_affecting_tests($$$)
|
|
{
|
|
my ($testdata, $testfncdata, $testbrdata) = @_;
|
|
my $testname;
|
|
my $testcount;
|
|
my $testfnccount;
|
|
my $testbrcount;
|
|
my %result;
|
|
my $found;
|
|
my $hit;
|
|
my $fn_found;
|
|
my $fn_hit;
|
|
my $br_found;
|
|
my $br_hit;
|
|
|
|
foreach $testname (keys(%{$testdata}))
|
|
{
|
|
# Get (line number -> count) hash for this test case
|
|
$testcount = $testdata->{$testname};
|
|
$testfnccount = $testfncdata->{$testname};
|
|
$testbrcount = $testbrdata->{$testname};
|
|
|
|
# Calculate sum
|
|
($found, $hit) = get_found_and_hit($testcount);
|
|
($fn_found, $fn_hit) = get_func_found_and_hit($testfnccount);
|
|
($br_found, $br_hit) = get_br_found_and_hit($testbrcount);
|
|
|
|
if ($hit>0)
|
|
{
|
|
$result{$testname} = "$found,$hit,$fn_found,$fn_hit,".
|
|
"$br_found,$br_hit";
|
|
}
|
|
}
|
|
|
|
return(\%result);
|
|
}
|
|
|
|
|
|
sub get_hash_reverse($)
|
|
{
|
|
my ($hash) = @_;
|
|
my %result;
|
|
|
|
foreach (keys(%{$hash})) {
|
|
$result{$hash->{$_}} = $_;
|
|
}
|
|
|
|
return \%result;
|
|
}
|
|
|
|
#
|
|
# write_source(filehandle, source_filename, count_data, checksum_data,
|
|
# converted_data, func_data, sumbrcount)
|
|
#
|
|
# Write an HTML view of a source code file. Returns a list containing
|
|
# data as needed by gen_png().
|
|
#
|
|
# Die on error.
|
|
#
|
|
|
|
sub write_source($$$$$$$)
|
|
{
|
|
local *HTML_HANDLE = $_[0];
|
|
local *SOURCE_HANDLE;
|
|
my $source_filename = $_[1];
|
|
my %count_data;
|
|
my $line_number;
|
|
my @result;
|
|
my $checkdata = $_[3];
|
|
my $converted = $_[4];
|
|
my $funcdata = $_[5];
|
|
my $sumbrcount = $_[6];
|
|
my $datafunc = get_hash_reverse($funcdata);
|
|
my $add_anchor;
|
|
my @file;
|
|
|
|
if ($_[2])
|
|
{
|
|
%count_data = %{$_[2]};
|
|
}
|
|
|
|
if (!open(SOURCE_HANDLE, "<", $source_filename)) {
|
|
my @lines;
|
|
my $last_line = 0;
|
|
|
|
if (!$ignore[$ERROR_SOURCE]) {
|
|
die("ERROR: cannot read $source_filename\n");
|
|
}
|
|
|
|
# Continue without source file
|
|
warn("WARNING: cannot read $source_filename!\n");
|
|
|
|
@lines = sort( { $a <=> $b } keys(%count_data));
|
|
if (@lines) {
|
|
$last_line = $lines[scalar(@lines) - 1];
|
|
}
|
|
return ( ":" ) if ($last_line < 1);
|
|
|
|
# Simulate gcov behavior
|
|
for ($line_number = 1; $line_number <= $last_line;
|
|
$line_number++) {
|
|
push(@file, "/* EOF */");
|
|
}
|
|
} else {
|
|
@file = <SOURCE_HANDLE>;
|
|
}
|
|
|
|
write_source_prolog(*HTML_HANDLE);
|
|
$line_number = 0;
|
|
foreach (@file) {
|
|
$line_number++;
|
|
chomp($_);
|
|
|
|
# Also remove CR from line-end
|
|
s/\015$//;
|
|
|
|
# Source code matches coverage data?
|
|
if (defined($checkdata->{$line_number}) &&
|
|
($checkdata->{$line_number} ne md5_base64($_)))
|
|
{
|
|
die("ERROR: checksum mismatch at $source_filename:".
|
|
"$line_number\n");
|
|
}
|
|
|
|
$add_anchor = 0;
|
|
if ($frames) {
|
|
if (($line_number - 1) % $nav_resolution == 0) {
|
|
$add_anchor = 1;
|
|
}
|
|
}
|
|
if ($func_coverage) {
|
|
if ($line_number == 1) {
|
|
$add_anchor = 1;
|
|
} elsif (defined($datafunc->{$line_number +
|
|
$func_offset})) {
|
|
$add_anchor = 1;
|
|
}
|
|
}
|
|
push (@result,
|
|
write_source_line(HTML_HANDLE, $line_number,
|
|
$_, $count_data{$line_number},
|
|
$converted->{$line_number},
|
|
$sumbrcount->{$line_number}, $add_anchor));
|
|
}
|
|
|
|
close(SOURCE_HANDLE);
|
|
write_source_epilog(*HTML_HANDLE);
|
|
return(@result);
|
|
}
|
|
|
|
|
|
sub funcview_get_func_code($$$)
|
|
{
|
|
my ($name, $base, $type) = @_;
|
|
my $result;
|
|
my $link;
|
|
|
|
if ($sort && $type == 1) {
|
|
$link = "$name.func.$html_ext";
|
|
}
|
|
$result = "Function Name";
|
|
$result .= get_sort_code($link, "Sort by function name", $base);
|
|
|
|
return $result;
|
|
}
|
|
|
|
sub funcview_get_count_code($$$)
|
|
{
|
|
my ($name, $base, $type) = @_;
|
|
my $result;
|
|
my $link;
|
|
|
|
if ($sort && $type == 0) {
|
|
$link = "$name.func-sort-c.$html_ext";
|
|
}
|
|
$result = "Hit count";
|
|
$result .= get_sort_code($link, "Sort by hit count", $base);
|
|
|
|
return $result;
|
|
}
|
|
|
|
#
|
|
# funcview_get_sorted(funcdata, sumfncdata, sort_type)
|
|
#
|
|
# Depending on the value of sort_type, return a list of functions sorted
|
|
# by name (type 0) or by the associated call count (type 1).
|
|
#
|
|
|
|
sub funcview_get_sorted($$$)
|
|
{
|
|
my ($funcdata, $sumfncdata, $type) = @_;
|
|
|
|
if ($type == 0) {
|
|
return sort(keys(%{$funcdata}));
|
|
}
|
|
return sort({$sumfncdata->{$b} <=> $sumfncdata->{$a}}
|
|
keys(%{$sumfncdata}));
|
|
}
|
|
|
|
#
|
|
# write_function_table(filehandle, source_file, sumcount, funcdata,
|
|
# sumfnccount, testfncdata, sumbrcount, testbrdata,
|
|
# base_name, base_dir, sort_type)
|
|
#
|
|
# Write an HTML table listing all functions in a source file, including
|
|
# also function call counts and line coverages inside of each function.
|
|
#
|
|
# Die on error.
|
|
#
|
|
|
|
sub write_function_table(*$$$$$$$$$$)
|
|
{
|
|
local *HTML_HANDLE = $_[0];
|
|
my $source = $_[1];
|
|
my $sumcount = $_[2];
|
|
my $funcdata = $_[3];
|
|
my $sumfncdata = $_[4];
|
|
my $testfncdata = $_[5];
|
|
my $sumbrcount = $_[6];
|
|
my $testbrdata = $_[7];
|
|
my $name = $_[8];
|
|
my $base = $_[9];
|
|
my $type = $_[10];
|
|
my $func;
|
|
my $func_code;
|
|
my $count_code;
|
|
|
|
# Get HTML code for headings
|
|
$func_code = funcview_get_func_code($name, $base, $type);
|
|
$count_code = funcview_get_count_code($name, $base, $type);
|
|
write_html(*HTML_HANDLE, <<END_OF_HTML)
|
|
<center>
|
|
<table width="60%" cellpadding=1 cellspacing=1 border=0>
|
|
<tr><td><br></td></tr>
|
|
<tr>
|
|
<td width="80%" class="tableHead">$func_code</td>
|
|
<td width="20%" class="tableHead">$count_code</td>
|
|
</tr>
|
|
END_OF_HTML
|
|
;
|
|
|
|
# Get a sorted table
|
|
foreach $func (funcview_get_sorted($funcdata, $sumfncdata, $type)) {
|
|
if (!defined($funcdata->{$func}))
|
|
{
|
|
next;
|
|
}
|
|
|
|
my $startline = $funcdata->{$func} - $func_offset;
|
|
my $name = $func;
|
|
my $count = $sumfncdata->{$name};
|
|
my $countstyle;
|
|
|
|
# Demangle C++ function names if requested
|
|
if ($demangle_cpp) {
|
|
$name = `c++filt "$name"`;
|
|
chomp($name);
|
|
}
|
|
# Escape any remaining special characters
|
|
$name = escape_html($name);
|
|
if ($startline < 1) {
|
|
$startline = 1;
|
|
}
|
|
if ($count == 0) {
|
|
$countstyle = "coverFnLo";
|
|
} else {
|
|
$countstyle = "coverFnHi";
|
|
}
|
|
|
|
write_html(*HTML_HANDLE, <<END_OF_HTML)
|
|
<tr>
|
|
<td class="coverFn"><a href="$source#$startline">$name</a></td>
|
|
<td class="$countstyle">$count</td>
|
|
</tr>
|
|
END_OF_HTML
|
|
;
|
|
}
|
|
write_html(*HTML_HANDLE, <<END_OF_HTML)
|
|
</table>
|
|
<br>
|
|
</center>
|
|
END_OF_HTML
|
|
;
|
|
}
|
|
|
|
|
|
#
|
|
# info(printf_parameter)
|
|
#
|
|
# Use printf to write PRINTF_PARAMETER to stdout only when the $quiet flag
|
|
# is not set.
|
|
#
|
|
|
|
sub info(@)
|
|
{
|
|
if (!$quiet)
|
|
{
|
|
# Print info string
|
|
printf(@_);
|
|
}
|
|
}
|
|
|
|
|
|
#
|
|
# subtract_counts(data_ref, base_ref)
|
|
#
|
|
|
|
sub subtract_counts($$)
|
|
{
|
|
my %data = %{$_[0]};
|
|
my %base = %{$_[1]};
|
|
my $line;
|
|
my $data_count;
|
|
my $base_count;
|
|
my $hit = 0;
|
|
my $found = 0;
|
|
|
|
foreach $line (keys(%data))
|
|
{
|
|
$found++;
|
|
$data_count = $data{$line};
|
|
$base_count = $base{$line};
|
|
|
|
if (defined($base_count))
|
|
{
|
|
$data_count -= $base_count;
|
|
|
|
# Make sure we don't get negative numbers
|
|
if ($data_count<0) { $data_count = 0; }
|
|
}
|
|
|
|
$data{$line} = $data_count;
|
|
if ($data_count > 0) { $hit++; }
|
|
}
|
|
|
|
return (\%data, $found, $hit);
|
|
}
|
|
|
|
|
|
#
|
|
# subtract_fnccounts(data, base)
|
|
#
|
|
# Subtract function call counts found in base from those in data.
|
|
# Return (data, f_found, f_hit).
|
|
#
|
|
|
|
sub subtract_fnccounts($$)
|
|
{
|
|
my %data;
|
|
my %base;
|
|
my $func;
|
|
my $data_count;
|
|
my $base_count;
|
|
my $fn_hit = 0;
|
|
my $fn_found = 0;
|
|
|
|
%data = %{$_[0]} if (defined($_[0]));
|
|
%base = %{$_[1]} if (defined($_[1]));
|
|
foreach $func (keys(%data)) {
|
|
$fn_found++;
|
|
$data_count = $data{$func};
|
|
$base_count = $base{$func};
|
|
|
|
if (defined($base_count)) {
|
|
$data_count -= $base_count;
|
|
|
|
# Make sure we don't get negative numbers
|
|
if ($data_count < 0) {
|
|
$data_count = 0;
|
|
}
|
|
}
|
|
|
|
$data{$func} = $data_count;
|
|
if ($data_count > 0) {
|
|
$fn_hit++;
|
|
}
|
|
}
|
|
|
|
return (\%data, $fn_found, $fn_hit);
|
|
}
|
|
|
|
|
|
#
|
|
# apply_baseline(data_ref, baseline_ref)
|
|
#
|
|
# Subtract the execution counts found in the baseline hash referenced by
|
|
# BASELINE_REF from actual data in DATA_REF.
|
|
#
|
|
|
|
sub apply_baseline($$)
|
|
{
|
|
my %data_hash = %{$_[0]};
|
|
my %base_hash = %{$_[1]};
|
|
my $filename;
|
|
my $testname;
|
|
my $data;
|
|
my $data_testdata;
|
|
my $data_funcdata;
|
|
my $data_checkdata;
|
|
my $data_testfncdata;
|
|
my $data_testbrdata;
|
|
my $data_count;
|
|
my $data_testfnccount;
|
|
my $data_testbrcount;
|
|
my $base;
|
|
my $base_checkdata;
|
|
my $base_sumfnccount;
|
|
my $base_sumbrcount;
|
|
my $base_count;
|
|
my $sumcount;
|
|
my $sumfnccount;
|
|
my $sumbrcount;
|
|
my $found;
|
|
my $hit;
|
|
my $fn_found;
|
|
my $fn_hit;
|
|
my $br_found;
|
|
my $br_hit;
|
|
|
|
foreach $filename (keys(%data_hash))
|
|
{
|
|
# Get data set for data and baseline
|
|
$data = $data_hash{$filename};
|
|
$base = $base_hash{$filename};
|
|
|
|
# Skip data entries for which no base entry exists
|
|
if (!defined($base))
|
|
{
|
|
next;
|
|
}
|
|
|
|
# Get set entries for data and baseline
|
|
($data_testdata, undef, $data_funcdata, $data_checkdata,
|
|
$data_testfncdata, undef, $data_testbrdata) =
|
|
get_info_entry($data);
|
|
(undef, $base_count, undef, $base_checkdata, undef,
|
|
$base_sumfnccount, undef, $base_sumbrcount) =
|
|
get_info_entry($base);
|
|
|
|
# Check for compatible checksums
|
|
merge_checksums($data_checkdata, $base_checkdata, $filename);
|
|
|
|
# sumcount has to be calculated anew
|
|
$sumcount = {};
|
|
$sumfnccount = {};
|
|
$sumbrcount = {};
|
|
|
|
# For each test case, subtract test specific counts
|
|
foreach $testname (keys(%{$data_testdata}))
|
|
{
|
|
# Get counts of both data and baseline
|
|
$data_count = $data_testdata->{$testname};
|
|
$data_testfnccount = $data_testfncdata->{$testname};
|
|
$data_testbrcount = $data_testbrdata->{$testname};
|
|
|
|
($data_count, undef, $hit) =
|
|
subtract_counts($data_count, $base_count);
|
|
($data_testfnccount) =
|
|
subtract_fnccounts($data_testfnccount,
|
|
$base_sumfnccount);
|
|
($data_testbrcount) =
|
|
combine_brcount($data_testbrcount,
|
|
$base_sumbrcount, $BR_SUB);
|
|
|
|
|
|
# Check whether this test case did hit any line at all
|
|
if ($hit > 0)
|
|
{
|
|
# Write back resulting hash
|
|
$data_testdata->{$testname} = $data_count;
|
|
$data_testfncdata->{$testname} =
|
|
$data_testfnccount;
|
|
$data_testbrdata->{$testname} =
|
|
$data_testbrcount;
|
|
}
|
|
else
|
|
{
|
|
# Delete test case which did not impact this
|
|
# file
|
|
delete($data_testdata->{$testname});
|
|
delete($data_testfncdata->{$testname});
|
|
delete($data_testbrdata->{$testname});
|
|
}
|
|
|
|
# Add counts to sum of counts
|
|
($sumcount, $found, $hit) =
|
|
add_counts($sumcount, $data_count);
|
|
($sumfnccount, $fn_found, $fn_hit) =
|
|
add_fnccount($sumfnccount, $data_testfnccount);
|
|
($sumbrcount, $br_found, $br_hit) =
|
|
combine_brcount($sumbrcount, $data_testbrcount,
|
|
$BR_ADD);
|
|
}
|
|
|
|
# Write back resulting entry
|
|
set_info_entry($data, $data_testdata, $sumcount, $data_funcdata,
|
|
$data_checkdata, $data_testfncdata, $sumfnccount,
|
|
$data_testbrdata, $sumbrcount, $found, $hit,
|
|
$fn_found, $fn_hit, $br_found, $br_hit);
|
|
|
|
$data_hash{$filename} = $data;
|
|
}
|
|
|
|
return (\%data_hash);
|
|
}
|
|
|
|
|
|
#
|
|
# remove_unused_descriptions()
|
|
#
|
|
# Removes all test descriptions from the global hash %test_description which
|
|
# are not present in %info_data.
|
|
#
|
|
|
|
sub remove_unused_descriptions()
|
|
{
|
|
my $filename; # The current filename
|
|
my %test_list; # Hash containing found test names
|
|
my $test_data; # Reference to hash test_name -> count_data
|
|
my $before; # Initial number of descriptions
|
|
my $after; # Remaining number of descriptions
|
|
|
|
$before = scalar(keys(%test_description));
|
|
|
|
foreach $filename (keys(%info_data))
|
|
{
|
|
($test_data) = get_info_entry($info_data{$filename});
|
|
foreach (keys(%{$test_data}))
|
|
{
|
|
$test_list{$_} = "";
|
|
}
|
|
}
|
|
|
|
# Remove descriptions for tests which are not in our list
|
|
foreach (keys(%test_description))
|
|
{
|
|
if (!defined($test_list{$_}))
|
|
{
|
|
delete($test_description{$_});
|
|
}
|
|
}
|
|
|
|
$after = scalar(keys(%test_description));
|
|
if ($after < $before)
|
|
{
|
|
info("Removed ".($before - $after).
|
|
" unused descriptions, $after remaining.\n");
|
|
}
|
|
}
|
|
|
|
|
|
#
|
|
# apply_prefix(filename, prefix)
|
|
#
|
|
# If FILENAME begins with PREFIX, remove PREFIX from FILENAME and return
|
|
# resulting string, otherwise return FILENAME.
|
|
#
|
|
|
|
sub apply_prefix($$)
|
|
{
|
|
my $filename = $_[0];
|
|
my $prefix = $_[1];
|
|
|
|
if (defined($prefix) && ($prefix ne ""))
|
|
{
|
|
if ($filename =~ /^\Q$prefix\E\/(.*)$/)
|
|
{
|
|
return substr($filename, length($prefix) + 1);
|
|
}
|
|
}
|
|
|
|
return $filename;
|
|
}
|
|
|
|
|
|
#
|
|
# system_no_output(mode, parameters)
|
|
#
|
|
# Call an external program using PARAMETERS while suppressing depending on
|
|
# the value of MODE:
|
|
#
|
|
# MODE & 1: suppress STDOUT
|
|
# MODE & 2: suppress STDERR
|
|
#
|
|
# Return 0 on success, non-zero otherwise.
|
|
#
|
|
|
|
sub system_no_output($@)
|
|
{
|
|
my $mode = shift;
|
|
my $result;
|
|
local *OLD_STDERR;
|
|
local *OLD_STDOUT;
|
|
|
|
# Save old stdout and stderr handles
|
|
($mode & 1) && open(OLD_STDOUT, ">>&", "STDOUT");
|
|
($mode & 2) && open(OLD_STDERR, ">>&", "STDERR");
|
|
|
|
# Redirect to /dev/null
|
|
($mode & 1) && open(STDOUT, ">", "/dev/null");
|
|
($mode & 2) && open(STDERR, ">", "/dev/null");
|
|
|
|
system(@_);
|
|
$result = $?;
|
|
|
|
# Close redirected handles
|
|
($mode & 1) && close(STDOUT);
|
|
($mode & 2) && close(STDERR);
|
|
|
|
# Restore old handles
|
|
($mode & 1) && open(STDOUT, ">>&", "OLD_STDOUT");
|
|
($mode & 2) && open(STDERR, ">>&", "OLD_STDERR");
|
|
|
|
return $result;
|
|
}
|
|
|
|
|
|
#
|
|
# read_config(filename)
|
|
#
|
|
# Read configuration file FILENAME and return a reference to a hash containing
|
|
# all valid key=value pairs found.
|
|
#
|
|
|
|
sub read_config($)
|
|
{
|
|
my $filename = $_[0];
|
|
my %result;
|
|
my $key;
|
|
my $value;
|
|
local *HANDLE;
|
|
|
|
if (!open(HANDLE, "<", $filename))
|
|
{
|
|
warn("WARNING: cannot read configuration file $filename\n");
|
|
return undef;
|
|
}
|
|
while (<HANDLE>)
|
|
{
|
|
chomp;
|
|
# Skip comments
|
|
s/#.*//;
|
|
# Remove leading blanks
|
|
s/^\s+//;
|
|
# Remove trailing blanks
|
|
s/\s+$//;
|
|
next unless length;
|
|
($key, $value) = split(/\s*=\s*/, $_, 2);
|
|
if (defined($key) && defined($value))
|
|
{
|
|
$result{$key} = $value;
|
|
}
|
|
else
|
|
{
|
|
warn("WARNING: malformed statement in line $. ".
|
|
"of configuration file $filename\n");
|
|
}
|
|
}
|
|
close(HANDLE);
|
|
return \%result;
|
|
}
|
|
|
|
|
|
#
|
|
# apply_config(REF)
|
|
#
|
|
# REF is a reference to a hash containing the following mapping:
|
|
#
|
|
# key_string => var_ref
|
|
#
|
|
# where KEY_STRING is a keyword and VAR_REF is a reference to an associated
|
|
# variable. If the global configuration hashes CONFIG or OPT_RC contain a value
|
|
# for keyword KEY_STRING, VAR_REF will be assigned the value for that keyword.
|
|
#
|
|
|
|
sub apply_config($)
|
|
{
|
|
my $ref = $_[0];
|
|
|
|
foreach (keys(%{$ref}))
|
|
{
|
|
if (defined($opt_rc{$_})) {
|
|
${$ref->{$_}} = $opt_rc{$_};
|
|
} elsif (defined($config->{$_})) {
|
|
${$ref->{$_}} = $config->{$_};
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
#
|
|
# get_html_prolog(FILENAME)
|
|
#
|
|
# If FILENAME is defined, return contents of file. Otherwise return default
|
|
# HTML prolog. Die on error.
|
|
#
|
|
|
|
sub get_html_prolog($)
|
|
{
|
|
my $filename = $_[0];
|
|
my $result = "";
|
|
|
|
if (defined($filename))
|
|
{
|
|
local *HANDLE;
|
|
|
|
open(HANDLE, "<", $filename)
|
|
or die("ERROR: cannot open html prolog $filename!\n");
|
|
while (<HANDLE>)
|
|
{
|
|
$result .= $_;
|
|
}
|
|
close(HANDLE);
|
|
}
|
|
else
|
|
{
|
|
$result = <<END_OF_HTML
|
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
|
|
|
|
<html lang="en">
|
|
|
|
<head>
|
|
<meta http-equiv="Content-Type" content="text/html; charset=$charset">
|
|
<title>\@pagetitle\@</title>
|
|
<link rel="stylesheet" type="text/css" href="\@basedir\@gcov.css">
|
|
</head>
|
|
|
|
<body>
|
|
|
|
END_OF_HTML
|
|
;
|
|
}
|
|
|
|
return $result;
|
|
}
|
|
|
|
|
|
#
|
|
# get_html_epilog(FILENAME)
|
|
#
|
|
# If FILENAME is defined, return contents of file. Otherwise return default
|
|
# HTML epilog. Die on error.
|
|
#
|
|
sub get_html_epilog($)
|
|
{
|
|
my $filename = $_[0];
|
|
my $result = "";
|
|
|
|
if (defined($filename))
|
|
{
|
|
local *HANDLE;
|
|
|
|
open(HANDLE, "<", $filename)
|
|
or die("ERROR: cannot open html epilog $filename!\n");
|
|
while (<HANDLE>)
|
|
{
|
|
$result .= $_;
|
|
}
|
|
close(HANDLE);
|
|
}
|
|
else
|
|
{
|
|
$result = <<END_OF_HTML
|
|
|
|
</body>
|
|
</html>
|
|
END_OF_HTML
|
|
;
|
|
}
|
|
|
|
return $result;
|
|
|
|
}
|
|
|
|
sub warn_handler($)
|
|
{
|
|
my ($msg) = @_;
|
|
|
|
warn("$tool_name: $msg");
|
|
}
|
|
|
|
sub die_handler($)
|
|
{
|
|
my ($msg) = @_;
|
|
|
|
die("$tool_name: $msg");
|
|
}
|
|
|
|
#
|
|
# parse_ignore_errors(@ignore_errors)
|
|
#
|
|
# Parse user input about which errors to ignore.
|
|
#
|
|
|
|
sub parse_ignore_errors(@)
|
|
{
|
|
my (@ignore_errors) = @_;
|
|
my @items;
|
|
my $item;
|
|
|
|
return if (!@ignore_errors);
|
|
|
|
foreach $item (@ignore_errors) {
|
|
$item =~ s/\s//g;
|
|
if ($item =~ /,/) {
|
|
# Split and add comma-separated parameters
|
|
push(@items, split(/,/, $item));
|
|
} else {
|
|
# Add single parameter
|
|
push(@items, $item);
|
|
}
|
|
}
|
|
foreach $item (@items) {
|
|
my $item_id = $ERROR_ID{lc($item)};
|
|
|
|
if (!defined($item_id)) {
|
|
die("ERROR: unknown argument for --ignore-errors: ".
|
|
"$item\n");
|
|
}
|
|
$ignore[$item_id] = 1;
|
|
}
|
|
}
|
|
|
|
#
|
|
# rate(hit, found[, suffix, precision, width])
|
|
#
|
|
# Return the coverage rate [0..100] for HIT and FOUND values. 0 is only
|
|
# returned when HIT is 0. 100 is only returned when HIT equals FOUND.
|
|
# PRECISION specifies the precision of the result. SUFFIX defines a
|
|
# string that is appended to the result if FOUND is non-zero. Spaces
|
|
# are added to the start of the resulting string until it is at least WIDTH
|
|
# characters wide.
|
|
#
|
|
|
|
sub rate($$;$$$)
|
|
{
|
|
my ($hit, $found, $suffix, $precision, $width) = @_;
|
|
my $rate;
|
|
|
|
# Assign defaults if necessary
|
|
$precision = 1 if (!defined($precision));
|
|
$suffix = "" if (!defined($suffix));
|
|
$width = 0 if (!defined($width));
|
|
|
|
return sprintf("%*s", $width, "-") if (!defined($found) || $found == 0);
|
|
$rate = sprintf("%.*f", $precision, $hit * 100 / $found);
|
|
|
|
# Adjust rates if necessary
|
|
if ($rate == 0 && $hit > 0) {
|
|
$rate = sprintf("%.*f", $precision, 1 / 10 ** $precision);
|
|
} elsif ($rate == 100 && $hit != $found) {
|
|
$rate = sprintf("%.*f", $precision, 100 - 1 / 10 ** $precision);
|
|
}
|
|
|
|
return sprintf("%*s", $width, $rate.$suffix);
|
|
}
|