#!/bin/bash set -eu function usage { echo "Usage: $0 [OPTION]..." echo "Run Cinder's test suite(s)" echo "" echo " -V, --virtual-env Always use virtualenv. Install automatically if not present" echo " -N, --no-virtual-env Don't use virtualenv. Run tests in local environment" echo " -s, --no-site-packages Isolate the virtualenv from the global Python environment" echo " -r, --recreate-db Recreate the test database (deprecated, as this is now the default)." echo " -n, --no-recreate-db Don't recreate the test database." echo " -f, --force Force a clean re-build of the virtual environment. Useful when dependencies have been added." echo " -u, --update Update the virtual environment with any newer package versions" echo " -p, --pep8 Just run PEP8 and HACKING compliance check" echo " -8, --pep8-only-changed Just run PEP8 and HACKING compliance check on files changed since HEAD~1" echo " -P, --no-pep8 Don't run static code checks" echo " -c, --coverage Generate coverage report" echo " -d, --debug Run tests with testtools instead of testr. This allows you to use the debugger." echo " -h, --help Print this usage message" echo " --hide-elapsed Don't print the elapsed time for each test along with slow test list" echo " --virtual-env-path Location of the virtualenv directory" echo " Default: \$(pwd)" echo " --virtual-env-name Name of the virtualenv directory" echo " Default: .venv" echo " --tools-path Location of the tools directory" echo " Default: \$(pwd)" echo " --concurrency How many processes to use when running the tests. A value of 0 autodetects concurrency from your CPU count" echo " Default: 1" echo "" echo "Note: with no options specified, the script will try to run the tests in a virtual environment," echo " If no virtualenv is found, the script will ask if you would like to create one. If you " echo " prefer to run tests NOT in a virtual environment, simply pass the -N option." exit } function process_options { i=1 while [ $i -le $# ]; do case "${!i}" in -h|--help) usage;; -V|--virtual-env) always_venv=1; never_venv=0;; -N|--no-virtual-env) always_venv=0; never_venv=1;; -s|--no-site-packages) no_site_packages=1;; -r|--recreate-db) recreate_db=1;; -n|--no-recreate-db) recreate_db=0;; -f|--force) force=1;; -u|--update) update=1;; -p|--pep8) just_pep8=1;; -8|--pep8-only-changed) just_pep8_changed=1;; -P|--no-pep8) no_pep8=1;; -c|--coverage) coverage=1;; -d|--debug) debug=1;; --virtual-env-path) (( i++ )) venv_path=${!i} ;; --virtual-env-name) (( i++ )) venv_dir=${!i} ;; --tools-path) (( i++ )) tools_path=${!i} ;; --concurrency) (( i++ )) concurrency=${!i} ;; -*) testropts="$testropts ${!i}";; *) testrargs="$testrargs ${!i}" esac (( i++ )) done } tool_path=${tools_path:-$(pwd)} venv_path=${venv_path:-$(pwd)} venv_dir=${venv_name:-.venv} with_venv=tools/with_venv.sh always_venv=0 never_venv=0 force=0 no_site_packages=0 installvenvopts= testrargs= testropts= wrapper="" just_pep8=0 just_pep8_changed=0 no_pep8=0 coverage=0 debug=0 recreate_db=1 update=0 concurrency=1 LANG=en_US.UTF-8 LANGUAGE=en_US:en LC_ALL=C process_options $@ # Make our paths available to other scripts we call export venv_path export venv_dir export venv_name export tools_dir export venv=${venv_path}/${venv_dir} if [ $no_site_packages -eq 1 ]; then installvenvopts="--no-site-packages" fi function run_tests { # Cleanup *pyc ${wrapper} find . -type f -name "*.pyc" -delete if [ $debug -eq 1 ]; then if [ "$testropts" = "" ] && [ "$testrargs" = "" ]; then # Default to running all tests if specific test is not # provided. testrargs="discover ./cinder/tests" fi ${wrapper} python -m testtools.run $testropts $testrargs # Short circuit because all of the testr and coverage stuff # below does not make sense when running testtools.run for # debugging purposes. return $? fi if [ $coverage -eq 1 ]; then TESTRTESTS="$TESTRTESTS --coverage" else TESTRTESTS="$TESTRTESTS" fi # Just run the test suites in current environment set +e testrargs=`echo "$testrargs" | sed -e's/^\s*\(.*\)\s*$/\1/'` TESTRTESTS="$TESTRTESTS --testr-args='--subunit --concurrency $concurrency $testropts $testrargs'" if [ setup.cfg -nt cinder.egg-info/entry_points.txt ] then ${wrapper} python setup.py egg_info fi echo "Running \`${wrapper} $TESTRTESTS\`" if ${wrapper} which subunit-2to1 2>&1 > /dev/null then # subunit-2to1 is present, testr subunit stream should be in version 2 # format. Convert to version one before colorizing. bash -c "${wrapper} $TESTRTESTS | ${wrapper} subunit-2to1 | ${wrapper} tools/colorizer.py" else bash -c "${wrapper} $TESTRTESTS | ${wrapper} tools/colorizer.py" fi RESULT=$? set -e copy_subunit_log if [ $coverage -eq 1 ]; then echo "Generating coverage report in covhtml/" # Don't compute coverage for common code, which is tested elsewhere ${wrapper} coverage combine ${wrapper} coverage html --include='cinder/*' --omit='cinder/openstack/common/*' -d covhtml -i fi return $RESULT } function copy_subunit_log { LOGNAME=`cat .testrepository/next-stream` LOGNAME=$(($LOGNAME - 1)) LOGNAME=".testrepository/${LOGNAME}" cp $LOGNAME subunit.log } function warn_on_flake8_without_venv { if [ $never_venv -eq 1 ]; then echo "**WARNING**:" echo "Running flake8 without virtual env may miss OpenStack HACKING detection" fi } function run_pep8 { echo "Running flake8 ..." warn_on_flake8_without_venv bash -c "${wrapper} flake8 cinder* bin/*" } TESTRTESTS="python setup.py testr" if [ $never_venv -eq 0 ] then # Remove the virtual environment if --force used if [ $force -eq 1 ]; then echo "Cleaning virtualenv..." rm -rf ${venv} fi if [ $update -eq 1 ]; then echo "Updating virtualenv..." python tools/install_venv.py $installvenvopts fi if [ -e ${venv} ]; then wrapper="${with_venv}" else if [ $always_venv -eq 1 ]; then # Automatically install the virtualenv python tools/install_venv.py $installvenvopts wrapper="${with_venv}" else echo -e "No virtual environment found...create one? (Y/n) \c" read use_ve if [ "x$use_ve" = "xY" -o "x$use_ve" = "x" -o "x$use_ve" = "xy" ]; then # Install the virtualenv and run the test suite in it python tools/install_venv.py $installvenvopts wrapper=${with_venv} fi fi fi fi # Delete old coverage data from previous runs if [ $coverage -eq 1 ]; then ${wrapper} coverage erase fi if [ $just_pep8 -eq 1 ]; then run_pep8 ${wrapper} bash ./tools/config/check_uptodate.sh exit fi if [ $recreate_db -eq 1 ]; then rm -f tests.sqlite fi if [ $just_pep8_changed -eq 1 ]; then # NOTE(gilliard) We want to use flake8 to check the # entirety of every file that has a change in it. # Unfortunately the --filenames argument to flake8 only accepts # file *names* and there are no files named (eg) "nova/compute/manager.py". The # --diff argument behaves surprisingly as well, because although you feed it a # diff, it actually checks the file on disk anyway. files=$(git diff --name-only HEAD~1 | tr '\n' ' ') echo "Running flake8 on ${files}" warn_on_flake8_without_venv bash -c "diff -u --from-file /dev/null ${files} | ${wrapper} flake8 --diff" exit fi run_tests # NOTE(sirp): we only want to run pep8 when we're running the full-test suite, # not when we're running tests individually. To handle this, we need to # distinguish between options (testropts), which begin with a '-', and # arguments (testrargs). if [ -z "$testrargs" ]; then if [ $no_pep8 -eq 0 ]; then run_pep8 ${wrapper} bash ./tools/config/check_uptodate.sh fi fi