#!/bin/bash # # Serval Project testing framework for Bash shell # Copyright 2012 Paul Gardner-Stephen # # 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. # This file is sourced by all testing scripts. A typical test script looks # like this: # # #!/bin/bash # source testframework.sh # setup() { # export BLAH_CONFIG=$TFWTMP/blah.conf # echo "username=$LOGNAME" >$BLAH_CONFIG # } # teardown() { # # $TFWTMP is always removed after every test, so no need to # # remove blah.conf ourselves. # } # doc_feature1='Feature one works' # test_feature1() { # execute programUnderTest --feature1 arg1 arg2 # assertExitStatus '==' 0 # assertRealTime --message='ran in under half a second' '<=' 0.5 # assertStdoutIs "" # assertStderrIs "" # tfw_cat arg1 # } # doc_feature2='Feature two fails with status 1' # setup_feature2() { # # Overrides setup(), so we have to call it ourselves explicitly # # here if we still want it. # setup # echo "option=specialValue" >>$BLAH_CONFIG # } # test_feature2() { # execute programUnderTest --feature2 arg1 arg2 # assertExitStatus '==' 1 # assertStdoutIs -e "Response:\tok\n" # assertStderrGrep "^ERROR: missing arg3$" # } # runTests "$@" usage() { echo -n "\ Usage: ${0##*/} [options] [--] Options: -t, --trace Enable shell "set -x" tracing during tests, output to test log -v, --verbose Send test log to output during execution -j, --jobs Run all tests in parallel (by default runs as --jobs=1) --jobs=N Run tests in parallel, at most N at a time -E, --stop-on-error Do not execute any tests after an ERROR occurs -F, --stop-on-failure Do not execute any tests after a FAIL occurs --filter=PREFIX Only execute tests whose names start with PREFIX --filter=N Only execute test number N --filter=M-N Only execute tests with numbers in range M-N inclusive --filter=-N Only execute tests with numbers <= N --filter=N- Only execute tests with numbers >= N --filter=M,N,... Only execute tests with number M or N or ... " } # Internal utility for setting shopt variables and restoring their original # value: # _tfw_shopt -s extglob -u extdebug # ... # _tfw_shopt_restore _tfw_shopt() { if [ -n "$_tfw_shopt_orig" ]; then _tfw_fatal "unrestored shopt settings: $_tfw_shopt_orig" fi _tfw_shopt_orig= local op=s while [ $# -ne 0 ] do case "$1" in -s) op=s;; -u) op=u;; *) local opt="$1" _tfw_shopt_orig="${restore:+$restore; }shopt -$(shopt -q $opt && echo s || echo u) $opt" shopt -$op $opt ;; esac shift done } _tfw_shopt_restore() { if [ -n "$_tfw_shopt_orig" ]; then eval "$_tfw_shopt_orig" _tfw_shopt_orig= fi } _tfw_shopt_orig= declare -a _tfw_running_pids # The rest of this file is parsed for extended glob patterns. _tfw_shopt -s extglob runTests() { _tfw_stdout=1 _tfw_stderr=2 _tfw_checkBashVersion _tfw_checkTerminfo _tfw_invoking_script=$(abspath "${BASH_SOURCE[1]}") _tfw_suite_name="${_tfw_invoking_script##*/}" _tfw_cwd=$(abspath "$PWD") _tfw_tmpdir="${TFW_TMPDIR:-${TMPDIR:-/tmp}}/_tfw-$$" trap '_tfw_status=$?; rm -rf "$_tfw_tmpdir"; exit $_tfw_status' EXIT SIGHUP SIGINT SIGTERM rm -rf "$_tfw_tmpdir" mkdir -p "$_tfw_tmpdir" || return $? _tfw_logdir="${TFW_LOGDIR:-$_tfw_cwd/testlog}/$_tfw_suite_name" _tfw_trace=false _tfw_verbose=false _tfw_stop_on_error=false _tfw_stop_on_failure=false _tfw_default_timeout=60 local allargs="$*" local -a filters=() local njobs=1 _tfw_shopt -s extglob while [ $# -ne 0 ]; do case "$1" in --help) usage; exit 0;; -t|--trace) _tfw_trace=true;; -v|--verbose) _tfw_verbose=true;; --filter=*) filters+=("${1#*=}");; -j|--jobs) njobs=0;; --jobs=+([0-9])) njobs="${1#*=}";; --jobs=*) _tfw_fatal "invalid option: $1";; -E|--stop-on-error) _tfw_stop_on_error=true;; -F|--stop-on-failure) _tfw_stop_on_failure=true;; --) shift; break;; --*) _tfw_fatal "unsupported option: $1";; *) _tfw_fatal "spurious argument: $1";; esac shift done _tfw_shopt_restore if $_tfw_verbose && [ $njobs -ne 1 ]; then _tfw_fatal "--verbose is incompatible with --jobs=$njobs" fi # Create an empty results directory. _tfw_results_dir="$_tfw_tmpdir/results" mkdir "$_tfw_results_dir" || return $? # Create an empty log directory. mkdir -p "$_tfw_logdir" || return $? rm -f "$_tfw_logdir"/* # Enumerate all the test cases. _tfw_find_tests "${filters[@]}" # Iterate through all test cases, starting a new test whenever the number of # running tests is less than the job limit. _tfw_testcount=0 _tfw_passcount=0 _tfw_failcount=0 _tfw_errorcount=0 _tfw_fatalcount=0 _tfw_running_pids=() _tfw_test_number_watermark=0 local testNumber local testPosition=0 for ((testNumber = 1; testNumber <= ${#_tfw_tests[*]}; ++testNumber)); do testName="${_tfw_tests[$(($testNumber - 1))]}" [ -z "$testName" ] && continue let ++testPosition let ++_tfw_testcount # Wait for any existing child process to finish. while [ $njobs -ne 0 -a ${#_tfw_running_pids[*]} -ge $njobs ]; do _tfw_harvest_processes done [ $_tfw_fatalcount -ne 0 ] && break $_tfw_stop_on_error && [ $_tfw_errorcount -ne 0 ] && break $_tfw_stop_on_failure && [ $_tfw_failcount -ne 0 ] && break # Start the next test in a child process. _tfw_echo_intro $testPosition $testNumber $testName if $_tfw_verbose || [ $njobs -ne 1 ]; then echo fi echo "$testPosition $testNumber $testName" >"$_tfw_results_dir/$testName" ( _tfw_test_name="$testName" # Pick a unique decimal number that must not coincide with other tests # being run concurrently, _including tests being run in other test # scripts by other users on the same host_. We cannot simply use # $testNumber. The subshell process ID is ideal. We don't use # $BASHPID because MacOS only has Bash-3.2, and $BASHPID was introduced # in Bash-4. _tfw_unique=$($BASH -c 'echo $PPID') # All files created by this test belong inside a temporary directory. # The path name must be kept short because it is used to construct # named socket paths, which have a limited length. _tfw_tmp=/tmp/_tfw-$_tfw_unique trap '_tfw_status=$?; rm -rf "$_tfw_tmp"; exit $_tfw_status' EXIT SIGHUP SIGINT SIGTERM local start_time=$(_tfw_timestamp) local finish_time=unknown ( trap '_tfw_status=$?; _tfw_teardown; exit $_tfw_status' EXIT SIGHUP SIGINT SIGTERM _tfw_result=ERROR mkdir $_tfw_tmp || exit 255 _tfw_setup _tfw_result=FAIL _tfw_phase=testcase tfw_log "# CALL test_$_tfw_test_name()" $_tfw_trace && set -x test_$_tfw_test_name _tfw_result=PASS case $_tfw_result in PASS) exit 0;; FAIL) exit 1;; ERROR) exit 254;; esac exit 255 ) local stat=$? finish_time=$(_tfw_timestamp) local result=FATAL case $stat in 254) result=ERROR;; 1) result=FAIL;; 0) result=PASS;; esac echo "$testPosition $testNumber $testName $result" >"$_tfw_results_dir/$testName" { echo "Name: $testName" echo "Result: $result" echo "Started: $start_time" echo "Finished: $finish_time" echo '++++++++++ log.stdout ++++++++++' cat $_tfw_tmp/log.stdout echo '++++++++++' echo '++++++++++ log.stderr ++++++++++' cat $_tfw_tmp/log.stderr echo '++++++++++' if $_tfw_trace; then echo '++++++++++ log.xtrace ++++++++++' cat $_tfw_tmp/log.xtrace echo '++++++++++' fi } >"$_tfw_logdir/$testNumber.$testName.$result" exit 0 ) & _tfw_running_pids+=($!) ln -s "$_tfw_results_dir/$testName" "$_tfw_results_dir/pid-$!" done # Wait for all child processes to finish. while [ ${#_tfw_running_pids[*]} -ne 0 ]; do _tfw_harvest_processes done # Clean up working directory. rm -rf "$_tfw_tmpdir" trap - EXIT SIGHUP SIGINT SIGTERM # Echo result summary and exit with success if no failures or errors. s=$([ $_tfw_testcount -eq 1 ] || echo s) echo "$_tfw_testcount test$s, $_tfw_passcount pass, $_tfw_failcount fail, $_tfw_errorcount error" [ $_tfw_fatalcount -eq 0 -a $_tfw_failcount -eq 0 -a $_tfw_errorcount -eq 0 ] } _tfw_echo_intro() { local docvar="doc_$3" echo -n "$2. ${!docvar:-$3}..." [ $1 -gt $_tfw_test_number_watermark ] && _tfw_test_number_watermark=$1 } _tfw_harvest_processes() { # # This is the only way known to get the effect of a 'wait' builtin that will # return when _any_ child dies or after a one-second timeout. trap 'kill $spid 2>/dev/null' SIGCHLD sleep 1 & spid=$! set -m wait $spid 2>/dev/null trap - SIGCHLD # local -a surviving_pids=() local pid for pid in ${_tfw_running_pids[*]}; do if kill -0 $pid 2>/dev/null; then surviving_pids+=($pid) elif [ -s "$_tfw_results_dir/pid-$pid" ]; then set -- $(<"$_tfw_results_dir/pid-$pid") local testPosition="$1" local testNumber="$2" local testName="$3" local result="$4" case "$result" in ERROR) let _tfw_errorcount=_tfw_errorcount+1 ;; PASS) let _tfw_passcount=_tfw_passcount+1 ;; FAIL) let _tfw_failcount=_tfw_failcount+1 ;; *) result=FATAL let _tfw_fatalcount=_tfw_fatalcount+1 ;; esac local lines if ! $_tfw_verbose && [ $njobs -eq 1 ]; then echo -n " " _tfw_echo_result "$result" echo elif ! $_tfw_verbose && lines=$($_tfw_tput lines); then local travel=$(($_tfw_test_number_watermark - $testPosition + 1)) if [ $travel -gt 0 -a $travel -lt $lines ] && $_tfw_tput cuu $travel ; then _tfw_echo_intro $testPosition $testNumber $testName echo -n " " _tfw_echo_result "$result" echo travel=$(($_tfw_test_number_watermark - $testPosition)) [ $travel -gt 0 ] && $_tfw_tput cud $travel fi else echo -n "$testNumber. ... " _tfw_echo_result "$result" echo fi else _tfw_echoerr "${BASH_SOURCE[1]}: child process $pid terminated without result" fi done _tfw_running_pids=(${surviving_pids[*]}) } _tfw_echo_result() { local result="$1" case "$result" in ERROR | FATAL) $_tfw_tput setf 4 $_tfw_tput rev echo -n "$result" $_tfw_tput sgr0 $_tfw_tput op ;; PASS) $_tfw_tput setf 2 echo -n "$result" $_tfw_tput op ;; FAIL) $_tfw_tput setf 4 echo -n "$result" $_tfw_tput op ;; *) echo -n "$result" ;; esac } # The following functions can be overridden by a test script to provide a # default fixture for all test cases. setup() { : } teardown() { : } # The following functions are provided to facilitate writing test cases and # fixtures. # Echo the absolute path (containing symlinks if given) of the given # file/directory, which does not have to exist or even be accessible. abspath() { _tfw_abspath -L "$1" } # Echo the absolute path (resolving all symlinks) of the given file/directory, # which does not have to exist or even be accessible. realpath() { _tfw_abspath -P "$1" } # Escape all grep(1) basic regular expression metacharacters. escape_grep_basic() { local re="$1" local nil='' re="${re//[\\]/\\\\$nil}" re="${re//./\\.}" re="${re//\*/\\*}" re="${re//^/\\^}" re="${re//\$/\\$}" re="${re//\[/\\[}" re="${re//\]/\\]}" echo "$re" } # Escape all egrep(1) extended regular expression metacharacters. escape_grep_extended() { local re="$1" local nil='' re="${re//[\\]/\\\\$nil}" re="${re//./\\.}" re="${re//\*/\\*}" re="${re//\?/\\?}" re="${re//+/\\+}" re="${re//^/\\^}" re="${re//\$/\\$}" re="${re//(/\\(}" re="${re//)/\\)}" re="${re//|/\\|}" re="${re//\[/\\[}" re="${re//{/\\{}" echo "$re" } # Executes its arguments as a command: # - captures the standard output and error in temporary files for later # examination # - captures the exit status for later assertions # - sets the $executed variable to a description of the command that was # executed execute() { tfw_log "# execute" $(_tfw_shellarg "$@") _tfw_getopts execute "$@" shift $_tfw_getopts_shift _tfw_execute "$@" } executeOk() { tfw_log "# executeOk" $(_tfw_shellarg "$@") _tfw_getopts executeok "$@" _tfw_opt_exit_status=0 _tfw_dump_on_fail --stderr shift $_tfw_getopts_shift _tfw_execute "$@" } # Wait until a given condition is met: # - can specify the timeout with --timeout=SECONDS # - can specify the sleep interval with --sleep=SECONDS # - the condition is a command that is executed repeatedly until returns zero # status # where SECONDS may be fractional, eg, 1.5 wait_until() { tfw_log "# wait_until" $(_tfw_shellarg "$@") local start=$SECONDS _tfw_getopts wait_until "$@" shift $_tfw_getopts_shift sleep ${_tfw_opt_timeout:-$_tfw_default_timeout} & local timeout_pid=$! while true; do "$@" && break kill -0 $timeout_pid 2>/dev/null || fail "timeout" sleep ${_tfw_opt_sleep:-1} done local end=$SECONDS tfw_log "# waited for" $((end - start)) "seconds" return 0 } # Executes its arguments as a command in the current shell process (not in a # child process), so that side effects like functions setting variables will # have effect. # - if the exit status is non-zero, then fails the current test # - otherwise, logs a message indicating the assertion passed assert() { _tfw_getopts assert "$@" shift $_tfw_getopts_shift [ -z "$_tfw_message" ] && _tfw_message=$(_tfw_shellarg "$@") _tfw_assert "$@" || _tfw_failexit tfw_log "# assert $_tfw_message" return 0 } assertExpr() { _tfw_getopts assertexpr "$@" shift $_tfw_getopts_shift local awkexpr=$(_tfw_expr_to_awkexpr "$@") _tfw_message="${_tfw_message+$_tfw_message }($(_tfw_shellarg \"$awkexpr\"))" _tfw_assert _tfw_eval_awkexpr "$awkexpr" || _tfw_failexit tfw_log "# assert $_tfw_message" return 0 } fail() { _tfw_getopts fail "$@" shift $_tfw_getopts_shift [ $# -ne 0 ] && _tfw_failmsg "$1" _tfw_backtrace _tfw_failexit } error() { _tfw_getopts error "$@" shift $_tfw_getopts_shift [ $# -ne 0 ] && _tfw_errormsg "$1" _tfw_backtrace _tfw_errorexit } fatal() { [ $# -eq 0 ] && set -- "no reason given" _tfw_fatalmsg "$@" _tfw_backtrace _tfw_fatalexit } # Append a message to the test case's stdout log. A normal 'echo' to stdout # will also do this, but tfw_log will work even in a context that stdout (fd 1) # is redirected. tfw_log() { local ts=$(_tfw_timestamp) cat >&$_tfw_log_fd <&$_tfw_log_fd } tfw_core_backtrace() { local executable="$1" local corefile="$2" echo backtrace >"$_tfw_tmpdir/backtrace.gdb" tfw_log "#--- gdb backtrace from $executable $corefile ---" gdb -n -batch -x "$_tfw_tmpdir/backtrace.gdb" "$executable" "$corefile" &1 5>&2 6>$_tfw_tmp/log.stdout 1>&6 2>$_tfw_tmp/log.stderr 7>$_tfw_tmp/log.xtrace BASH_XTRACEFD=7 _tfw_log_fd=6 _tfw_stdout=5 _tfw_stderr=5 if $_tfw_verbose; then # Find the PID of the current subshell process. Cannot use $BASHPID # because MacOS only has Bash-3.2, and $BASHPID was introduced in Bash-4. local mypid=$($BASH -c 'echo $PPID') # These tail processes will die when the current subshell exits. tail --pid=$mypid --follow $_tfw_tmp/log.stdout >&$_tfw_stdout 2>/dev/null & tail --pid=$mypid --follow $_tfw_tmp/log.stderr >&$_tfw_stderr 2>/dev/null & fi export TFWUNIQUE=$_tfw_unique export TFWVAR=$_tfw_tmp/var mkdir $TFWVAR export TFWTMP=$_tfw_tmp/tmp mkdir $TFWTMP cd $TFWTMP tfw_log '# SETUP' case `type -t setup_$_tfw_test_name` in function) tfw_log "# call setup_$_tfw_test_name()" $_tfw_trace && set -x setup_$_tfw_test_name $_tfw_test_name set +x ;; *) tfw_log "# call setup($_tfw_test_name)" $_tfw_trace && set -x setup $_tfw_test_name set +x ;; esac tfw_log '# END SETUP' } _tfw_teardown() { _tfw_phase=teardown tfw_log '# TEARDOWN' case `type -t teardown_$_tfw_test_name` in function) tfw_log "# call teardown_$_tfw_test_name()" $_tfw_trace && set -x teardown_$_tfw_test_name set +x ;; *) tfw_log "# call teardown($_tfw_test_name)" $_tfw_trace && set -x teardown $_tfw_test_name set +x ;; esac tfw_log '# END TEARDOWN' } # Executes $_tfw_executable with the given arguments. _tfw_execute() { executed=$(_tfw_shellarg "${_tfw_executable##*/}" "$@") if $_tfw_opt_core_backtrace; then ulimit -S -c unlimited rm -f core fi { time -p "$_tfw_executable" "$@" >$_tfw_tmp/stdout 2>$_tfw_tmp/stderr ; } 2>$_tfw_tmp/times _tfw_exitStatus=$? # Deal with core dump. if $_tfw_opt_core_backtrace && [ -s core ]; then tfw_core_backtrace "$_tfw_executable" core fi # Deal with exit status. if [ -n "$_tfw_opt_exit_status" ]; then _tfw_message="exit status ($_tfw_exitStatus) of ($executed) is $_tfw_opt_exit_status" _tfw_dump_stderr_on_fail=true _tfw_assert [ "$_tfw_exitStatus" -eq "$_tfw_opt_exit_status" ] || _tfw_failexit tfw_log "# assert $_tfw_message" else tfw_log "# exit status of ($executed) = $_tfw_exitStatus" fi # Parse execution time report. if ! _tfw_parse_times_to_milliseconds real realtime_ms || ! _tfw_parse_times_to_milliseconds user usertime_ms || ! _tfw_parse_times_to_milliseconds sys systime_ms then tfw_log '# malformed output from time:' tfw_cat --header=times -v $_tfw_tmp/times fi return 0 } _tfw_parse_times_to_milliseconds() { local label="$1" local var="$2" local milliseconds=$(awk '$1 == "'"$label"'" { value = $2 minutes = 0 if (match(value, "[0-9]+m")) { minutes = substr(value, RSTART, RLENGTH - 1) value = substr(value, 1, RSTART - 1) substr(value, RSTART + RLENGTH) } if (substr(value, length(value)) == "s") { value = substr(value, 1, length(value) - 1) } if (match(value, "^[0-9]+(\.[0-9]+)?$")) { seconds = value + 0 print (minutes * 60 + seconds) * 1000 } }' $_tfw_tmp/times) [ -z "$milliseconds" ] && return 1 [ -n "$var" ] && eval $var=$milliseconds return 0 } _tfw_assert() { if ! "$@"; then _tfw_failmsg "assertion failed: ${_tfw_message:-$*}" _tfw_backtrace return 1 fi return 0 } declare -a _tfw_opt_dump_on_fail _tfw_dump_on_fail() { for arg; do local _found=false local _f for _f in "${_tfw_opt_dump_on_fail[@]}"; do if [ "$_f" = "$arg" ]; then _found=true break fi done $_found || _tfw_opt_dump_on_fail+=("$arg") done } _tfw_getopts() { local context="$1" shift _tfw_executable= _tfw_opt_core_backtrace=false _tfw_message= _tfw_opt_dump_on_fail=() _tfw_opt_error_on_fail=false _tfw_opt_exit_status= _tfw_opt_timeout= _tfw_opt_sleep= _tfw_opt_matches= _tfw_opt_line= _tfw_getopts_shift=0 _tfw_shopt -s extglob while [ $# -ne 0 ]; do case "$context:$1" in *:--stdout) _tfw_dump_on_fail --stdout;; *:--stderr) _tfw_dump_on_fail --stderr;; assert*:--dump-on-fail=*) _tfw_dump_on_fail "${1#*=}";; execute:--exit-status=+([0-9])) _tfw_opt_exit_status="${1#*=}";; execute:--exit-status=*) _tfw_error "invalid value: $1";; execute*:--executable=) _tfw_error "missing value: $1";; execute*:--executable=*) _tfw_executable="${1#*=}";; execute*:--core-backtrace) _tfw_opt_core_backtrace=true;; wait_until:--timeout=@(+([0-9])?(.+([0-9]))|*([0-9]).+([0-9]))) _tfw_opt_timeout="${1#*=}";; wait_until:--timeout=*) _tfw_error "invalid value: $1";; wait_until:--sleep=@(+([0-9])?(.+([0-9]))|*([0-9]).+([0-9]))) _tfw_opt_sleep="${1#*=}";; wait_until:--sleep=*) _tfw_error "invalid value: $1";; assert*:--error-on-fail) _tfw_opt_error_on_fail=true;; assert*:--message=*) _tfw_message="${1#*=}";; assertgrep:--matches=+([0-9])) _tfw_opt_matches="${1#*=}";; assertgrep:--matches=*) _tfw_error "invalid value: $1";; assertfilecontent:--line=+([0-9])) _tfw_opt_line="${1#*=}";; assertfilecontent:--line=*) _tfw_error "invalid value: $1";; *:--) let _tfw_getopts_shift=_tfw_getopts_shift+1; shift; break;; *:--*) _tfw_error "unsupported option: $1";; *) break;; esac let _tfw_getopts_shift=_tfw_getopts_shift+1 shift done case "$context" in execute*) if [ -z "$_tfw_executable" ]; then _tfw_executable="$1" let _tfw_getopts_shift=_tfw_getopts_shift+1 shift fi [ -z "$_tfw_executable" ] && _tfw_error "missing executable argument" ;; esac _tfw_shopt_restore return 0 } _tfw_expr_to_awkexpr() { local awkexpr= for arg; do if [ -z "${arg//[0-9]}" ]; then awkexpr="${awkexpr:+$awkexpr }$arg" else case $arg in '==' | '!=' | '<' | '<=' | '>' | '>=' | \ '~' | '!~' | '&&' | '||' | '!' ) awkexpr="${awkexpr:+$awkexpr }$arg" ;; *) arg=${arg//\\/\\\\} #} restore Vim syntax highlighting arg=${arg//"/\\"} awkexpr="${awkexpr:+$awkexpr }\"$arg\"" ;; esac fi done echo $awkexpr } _tfw_eval_awkexpr() { local awkerrs # on separate line so we don't lose exit status awkerrs=$(awk "BEGIN { exit(($*) ? 0 : 1) }" &1) local stat=$? if [ -n "$awkerrs" ]; then _tfw_error "invalid expression: $*" stat=254 fi return $stat } _tfw_assertExpr() { local awkexpr=$(_tfw_expr_to_awkexpr "$@") _tfw_assert _tfw_eval_awkexpr "$awkexpr" || _tfw_failexit } _tfw_assert_stdxxx_is() { local qual="$1" shift _tfw_getopts assertfilecontent --$qual "$@" shift $((_tfw_getopts_shift - 1)) if [ $# -lt 1 ]; then _tfw_error "incorrect arguments" return 254 fi case "$_tfw_opt_line" in '') ln -f "$_tfw_tmp/$qual" "$_tfw_tmp/content";; *) sed -n -e "${_tfw_opt_line}p" "$_tfw_tmp/$qual" >"$_tfw_tmp/content";; esac local message="${_tfw_message:-${_tfw_opt_line:+line $_tfw_opt_line of }$qual of ($executed) is $(_tfw_shellarg "$@")}" echo -n "$@" >$_tfw_tmp/stdxxx_is.tmp if ! cmp --quiet $_tfw_tmp/stdxxx_is.tmp "$_tfw_tmp/content"; then _tfw_failmsg "assertion failed: $message" _tfw_backtrace return 1 fi tfw_log "# assert $message" return 0 } _tfw_assert_stdxxx_linecount() { local qual="$1" shift _tfw_getopts assertfilecontent --$qual "$@" shift $((_tfw_getopts_shift - 1)) if [ $# -lt 1 ]; then _tfw_error "incorrect arguments" return 254 fi local lineCount=$(( $(cat $_tfw_tmp/$qual | wc -l) + 0 )) [ -z "$_tfw_message" ] && _tfw_message="$qual line count ($lineCount) $*" _tfw_assertExpr "$lineCount" "$@" || _tfw_failexit tfw_log "# assert $_tfw_message" return 0 } _tfw_assert_stdxxx_grep() { local qual="$1" shift _tfw_getopts assertgrep --$qual "$@" shift $((_tfw_getopts_shift - 1)) if [ $# -ne 1 ]; then _tfw_error "incorrect arguments" return 254 fi _tfw_assert_grep "$qual of ($executed)" $_tfw_tmp/$qual "$@" } _tfw_assert_grep() { local label="$1" local file="$2" local pattern="$3" local message= if ! [ -e "$file" ]; then _tfw_error "$file does not exist" ret=254 elif ! [ -f "$file" ]; then _tfw_error "$file is not a regular file" ret=254 elif ! [ -r "$file" ]; then _tfw_error "$file is not readable" ret=254 else local matches=$(( $(grep --regexp="$pattern" "$file" | wc -l) + 0 )) local done=false local ret=0 _tfw_shopt -s extglob case "$_tfw_opt_matches" in '') done=true message="${_tfw_message:-$label contains a line matching \"$pattern\"}" if [ $matches -ne 0 ]; then tfw_log "# assert $message" else _tfw_failmsg "assertion failed: $message" ret=1 fi ;; esac case "$_tfw_opt_matches" in +([0-9])) done=true local s=$([ $_tfw_opt_matches -ne 1 ] && echo s) message="${_tfw_message:-$label contains exactly $_tfw_opt_matches line$s matching \"$pattern\"}" if [ $matches -eq $_tfw_opt_matches ]; then tfw_log "# assert $message" else _tfw_failmsg "assertion failed: $message" ret=1 fi ;; esac case "$_tfw_opt_matches" in +([0-9])-*([0-9])) done=true local bound=${_tfw_opt_matches%-*} local s=$([ $bound -ne 1 ] && echo s) message="${_tfw_message:-$label contains at least $bound line$s matching \"$pattern\"}" if [ $matches -ge $bound ]; then tfw_log "# assert $message" else _tfw_failmsg "assertion failed: $message" ret=1 fi ;; esac case "$_tfw_opt_matches" in *([0-9])-+([0-9])) done=true local bound=${_tfw_opt_matches#*-} local s=$([ $bound -ne 1 ] && echo s) message="${_tfw_message:-$label contains at most $bound line$s matching \"$pattern\"}" if [ $matches -le $bound ]; then tfw_log "# assert $message" else _tfw_failmsg "assertion failed: $message" ret=1 fi ;; esac if ! $done; then _tfw_error "unsupported value for --matches=$_tfw_opt_matches" ret=254 fi _tfw_shopt_restore fi if [ $ret -ne 0 ]; then _tfw_backtrace fi return $ret } # Write a message to the real stderr of the test script, so the user sees it # immediately. Also write the message to the test log, so it can be recovered # later. _tfw_echoerr() { echo "$@" >&$_tfw_stderr if [ $_tfw_stderr -ne 2 ]; then echo "$@" >&2 fi } _tfw_checkBashVersion() { [ -z "$BASH_VERSION" ] && _tfw_fatal "not running in Bash (/bin/bash) shell" if [ -n "${BASH_VERSINFO[*]}" ]; then [ ${BASH_VERSINFO[0]} -gt 3 ] && return 0 if [ ${BASH_VERSINFO[0]} -eq 3 ]; then [ ${BASH_VERSINFO[1]} -gt 2 ] && return 0 if [ ${BASH_VERSINFO[1]} -eq 2 ]; then [ ${BASH_VERSINFO[2]} -ge 48 ] && return 0 fi fi fi _tfw_fatal "unsupported Bash version: $BASH_VERSION" } _tfw_checkTerminfo() { _tfw_tput=false case $(type -p tput) in */tput) _tfw_tput=tput;; esac } # Return a list of test names in the _tfw_tests array variable, in the order # that the test_TestName functions were defined. Test names must start with # an alphabetic character (not numeric or '_'). _tfw_find_tests() { _tfw_tests=() _tfw_shopt -s extdebug local name for name in $(builtin declare -F | sed -n -e '/^declare -f test_[A-Za-z]/s/^declare -f test_//p' | while read name; do builtin declare -F "test_$name"; done | sort --key 2,2n --key 3,3 | sed -e 's/^test_//' -e 's/[ ].*//') do local number=$((${#_tfw_tests[*]} + 1)) local testName= if [ $# -eq 0 ]; then testName="$name" else local filter for filter; do case "$filter" in +([0-9])) if [ $number -eq $filter ]; then testName="$name" break fi ;; +([0-9])*(,+([0-9]))) local oIFS="$IFS" IFS=, local -a numbers=($filter) IFS="$oIFS" local n for n in ${numbers[*]}; do if [ $number -eq $n ]; then testName="$name" break 2 fi done ;; +([0-9])-) local start=${filter%-} if [ $number -ge $start ]; then testName="$name" break fi ;; -+([0-9])) local end=${filter#-} if [ $number -le $end ]; then testName="$name" break fi ;; +([0-9])-+([0-9])) local start=${filter%-*} local end=${filter#*-} if [ $number -ge $start -a $number -le $end ]; then testName="$name" break fi ;; *) case "$name" in "$filter"*) testName="$name"; break;; esac ;; esac done fi _tfw_tests+=("$testName") done _tfw_shopt_restore } # A "fail" event occurs when any assertion fails, and indicates that the test # has not passed. Other tests may still proceed. A "fail" event during setup # or teardown is treated as an error, not a failure. _tfw_failmsg() { # A failure during setup or teardown is treated as an error. case $_tfw_phase in testcase) if ! $_tfw_opt_error_on_fail; then tfw_log "FAIL: $*" return 0; fi ;; esac tfw_log "ERROR: $*" } _tfw_backtrace() { tfw_log '#--- backtrace ---' local -i up=1 while [ "${BASH_SOURCE[$up]}" == "${BASH_SOURCE[0]}" ]; do let up=up+1 done local -i i=0 while [ $up -lt ${#FUNCNAME[*]} -a "${BASH_SOURCE[$up]}" != "${BASH_SOURCE[0]}" ]; do tfw_log "[$i] ${FUNCNAME[$(($up-1))]}() called from ${FUNCNAME[$up]}() at line ${BASH_LINENO[$(($up-1))]} of ${BASH_SOURCE[$up]}" let up=up+1 let i=i+1 done tfw_log '#---' } _tfw_failexit() { # When exiting a test case due to a failure, log any diagnostic output that # has been requested. tfw_cat "${_tfw_opt_dump_on_fail[@]}" # A failure during setup or teardown is treated as an error. case $_tfw_phase in testcase) if ! $_tfw_opt_error_on_fail; then exit 1 fi ;; esac _tfw_errorexit } # An "error" event prevents a test from running, so it neither passes nor fails. # Other tests may still proceed. _tfw_errormsg() { [ $# -eq 0 ] && set -- "(no message)" local -i up=1 local -i top=${#FUNCNAME[*]} let top=top-1 while [ $up -lt $top -a "${BASH_SOURCE[$up]}" == "${BASH_SOURCE[0]}" ]; do let up=up+1 done tfw_log "ERROR in ${FUNCNAME[$up]}: $*" } _tfw_error() { _tfw_errormsg "ERROR: $*" _tfw_backtrace _tfw_errorexit } _tfw_errorexit() { # Do not exit process during teardown _tfw_result=ERROR case $_tfw_phase in teardown) [ $_tfw_status -lt 254 ] && _tfw_status=254;; *) exit 254;; esac return 254 } # A "fatal" event stops the entire test run, and generally indicates an # insurmountable problem in the test script or in the test framework itself. _tfw_fatalmsg() { _tfw_echoerr "${BASH_SOURCE[1]}: FATAL: $*" } _tfw_fatal() { [ $# -eq 0 ] && set -- exiting _tfw_echoerr "${BASH_SOURCE[1]}: FATAL: $*" _tfw_fatalexit } _tfw_fatalexit() { exit 255 } # Restore the caller's shopt preferences before returning. _tfw_shopt_restore