Source code for CIME.bless_test_results

import CIME.compare_namelists, CIME.simple_compare
from CIME.test_scheduler import NAMELIST_PHASE
from CIME.utils import run_cmd, expect, get_scripts_root, get_model, EnvironmentContext
from CIME.test_status import *
from CIME.hist_utils import generate_baseline, compare_baseline
from CIME.case import Case

import os, glob, time, six
logger = logging.getLogger(__name__)

###############################################################################
[docs]def bless_namelists(test_name, test_dir, report_only, force, baseline_name, baseline_root): ############################################################################### # Be aware that restart test will overwrite the original namelist files # with versions of the files that should not be blessed. This forces us to # re-run create_test. # Update namelist files logger.info("Test '{}' had namelist diff".format(test_name)) if (not report_only and (force or six.moves.input("Update namelists (y/n)? ").upper() in ["Y", "YES"])): if baseline_name is None: stat, baseline_name, _ = run_cmd("./xmlquery --value BASELINE_NAME_CMP", from_dir=test_dir) if stat != 0 or not baseline_name: baseline_name = CIME.utils.get_current_branch(repo=CIME.utils.get_cime_root()) if baseline_root is None: stat, baseline_root, _ = run_cmd("./xmlquery --value BASELINE_ROOT", from_dir=test_dir) if stat != 0 or not baseline_root: return False, "Could not determine baseline root" create_test_gen_args = " -g {} ".format(baseline_name if get_model() == "cesm" else " -g -b {} ".format(baseline_name)) stat, _, err = run_cmd("{}/create_test {} -n {} --baseline-root {} -o".format(get_scripts_root(), test_name, create_test_gen_args, baseline_root)) if stat != 0: return False, "Namelist regen failed: '{}'".format(err) else: return True, None else: return True, None
###############################################################################
[docs]def bless_history(test_name, test_dir, baseline_name, baseline_root, report_only, force): ############################################################################### with Case(test_dir) as case: real_user = case.get_value("REALUSER") with EnvironmentContext(USER=real_user): if baseline_name is None: baseline_name = case.get_value("BASELINE_NAME_CMP") if not baseline_name: baseline_name = CIME.utils.get_current_branch(repo=CIME.utils.get_cime_root()) if baseline_root is None: baseline_root = case.get_value("BASELINE_ROOT") baseline_full_dir = os.path.join(baseline_root, baseline_name, case.get_value("CASEBASEID")) cmp_result, cmp_comments = compare_baseline(case, baseline_dir=baseline_full_dir, outfile_suffix=None) if cmp_result: logger.info("Diff appears to have been already resolved.") return True, None else: logger.info(cmp_comments) if (not report_only and (force or six.moves.input("Update this diff (y/n)? ").upper() in ["Y", "YES"])): gen_result, gen_comments = generate_baseline(case, baseline_dir=baseline_full_dir) if not gen_result: logger.warning("Hist file bless FAILED for test {}".format(test_name)) return False, "Generate baseline failed: {}".format(gen_comments) else: logger.info(gen_comments) return True, None else: return True, None
###############################################################################
[docs]def bless_test_results(baseline_name, baseline_root, test_root, compiler, test_id=None, namelists_only=False, hist_only=False, report_only=False, force=False, bless_tests=None, no_skip_pass=False): ############################################################################### test_id_glob = "*{}*".format(compiler) if test_id is None else "*{}*".format(test_id) test_status_files = glob.glob("{}/{}/{}".format(test_root, test_id_glob, TEST_STATUS_FILENAME)) expect(test_status_files, "No matching test cases found in for {}/{}/{}".format(test_root, test_id_glob, TEST_STATUS_FILENAME)) broken_blesses = [] for test_status_file in test_status_files: test_dir = os.path.dirname(test_status_file) ts = TestStatus(test_dir=test_dir) test_name = ts.get_name() if test_name is None: case_dir = os.path.basename(test_dir) test_name = CIME.utils.normalize_case_id(case_dir) if (bless_tests in [[], None] or CIME.utils.match_any(test_name, bless_tests)): broken_blesses.append((test_name, "test had invalid TestStatus file: '{}'".format(test_status_file))) else: continue if (bless_tests in [[], None] or CIME.utils.match_any(test_name, bless_tests)): overall_result = ts.get_overall_test_status() # See if we need to bless namelist if (not hist_only): if no_skip_pass: nl_bless = True else: nl_bless = ts.get_status(NAMELIST_PHASE) != TEST_PASS_STATUS else: nl_bless = False # See if we need to bless baselines if (not namelists_only): run_result = ts.get_status(RUN_PHASE) if (run_result is None): broken_blesses.append((test_name, "no run phase")) logger.warning("Test '{}' did not make it to run phase".format(test_name)) hist_bless = False elif (run_result != TEST_PASS_STATUS): broken_blesses.append((test_name, "test did not pass")) logger.warning("Test '{}' did not pass, not safe to bless".format(test_name)) hist_bless = False elif no_skip_pass: hist_bless = True else: hist_bless = ts.get_status(BASELINE_PHASE) != TEST_PASS_STATUS else: hist_bless = False # Now, do the bless if not nl_bless and not hist_bless: logger.info("Nothing to bless for test: {}, overall status: {}".format(test_name, overall_result)) else: logger.info("###############################################################################") logger.info("Blessing results for test: {}, most recent result: {}".format(test_name, overall_result)) logger.info("Case dir: {}".format(test_dir)) logger.info("###############################################################################") if not force: time.sleep(2) # Bless namelists if nl_bless: success, reason = bless_namelists(test_name, test_dir, report_only, force, baseline_name, baseline_root) if not success: broken_blesses.append((test_name, reason)) # Bless hist files if hist_bless: if "HOMME" in test_name: success = False reason = "HOMME tests cannot be blessed with bless_for_tests" else: success, reason = bless_history(test_name, test_dir, baseline_name, baseline_root, report_only, force) if (not success): broken_blesses.append((test_name, reason)) # Make sure user knows that some tests were not blessed success = True for broken_bless, reason in broken_blesses: logger.warning("FAILED TO BLESS TEST: {}, reason {}".format(broken_bless, reason)) success = False return success