DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Juraj Linkeš" <juraj.linkes@pantheon.tech>
To: thomas@monjalon.net, david.marchand@redhat.com,
	Honnappa.Nagarahalli@arm.com, ohilyard@iol.unh.edu,
	lijuan.tu@intel.com
Cc: dev@dpdk.org, "Juraj Linkeš" <juraj.linkes@pantheon.tech>
Subject: [RFC PATCH v1 05/15] dts: merge DTS framework/asan_test.py to DPDK
Date: Wed,  6 Apr 2022 14:55:56 +0000	[thread overview]
Message-ID: <20220406145606.2913834-6-juraj.linkes@pantheon.tech> (raw)
In-Reply-To: <20220406145606.2913834-1-juraj.linkes@pantheon.tech>

---
 dts/framework/asan_test.py | 439 +++++++++++++++++++++++++++++++++++++
 1 file changed, 439 insertions(+)
 create mode 100644 dts/framework/asan_test.py

diff --git a/dts/framework/asan_test.py b/dts/framework/asan_test.py
new file mode 100644
index 0000000000..8ca0536bcc
--- /dev/null
+++ b/dts/framework/asan_test.py
@@ -0,0 +1,439 @@
+import configparser
+import os
+import re
+from contextlib import contextmanager
+
+import xlrd
+
+import framework.settings as settings
+from framework.excel_reporter import ExcelReporter
+from framework.test_result import Result
+
+from .json_reporter import JSONReporter
+from .stats_reporter import StatsReporter
+
+ASan_CONFIG_SECT = "ASan"
+ASan_FILTER_BOUNDS = "filter_bounds"
+ASan_PARAM_KEY = "build_param"
+ASan_CONFIG_FILE_PATH = "%s/asan.cfg" % settings.CONFIG_ROOT_PATH
+COMMAND_PATTERN_OF_ADDRESS_RANDOM_SWITCH = (
+    "echo %s > /proc/sys/kernel/randomize_va_space"
+)
+COMMAND_OF_CLOSE_ADDRESS_RANDOM = COMMAND_PATTERN_OF_ADDRESS_RANDOM_SWITCH % 0
+COMMAND_OF_OPEN_ADDRESS_RANDOM = COMMAND_PATTERN_OF_ADDRESS_RANDOM_SWITCH % 2
+NEW_TEST_REPORT_FILE = "asan_test_results.xls"
+NEW_JSON_REPORT_FILE = "asan_test_results.json"
+NEW_STATS_REPORT_FILE = "asan_statistics.txt"
+ORIGIN_TEST_REPORT_FILE = "test_results.xls"
+MIN_LENGTH_OF_FILTERED_OUTPUT = 50
+REPORT_OUTPUT_PATH = ""
+
+
+class ASanTestProcess(object):
+    IS_SUPPORT_ASan_TEST = False
+
+    @staticmethod
+    def test_prepare(is_support_ASan_test, output_dir):
+        if is_support_ASan_test:
+            ASanTestProcess.IS_SUPPORT_ASan_TEST = True
+            # use framework default or customer defined output dir
+            global REPORT_OUTPUT_PATH
+            REPORT_OUTPUT_PATH = output_dir
+            _FrameworkADAPTER.decorator_dts_run()
+            _FrameworkADAPTER.decorator_send_expect()
+            _FrameworkADAPTER.decorator_build_install_dpdk()
+
+    @staticmethod
+    def test_process():
+        if ASanTestProcess.IS_SUPPORT_ASan_TEST:
+            report_process_obj = _NewReport()
+            report_process_obj.process_report_header()
+            report_process_obj.process_report_detail()
+            report_process_obj.save_report()
+
+
+class _FrameworkADAPTER(object):
+    @staticmethod
+    def decorator_build_install_dpdk():
+        added_param = _ASanConfig().build_param
+        if added_param is not None:
+            from framework.project_dpdk import DPDKdut
+
+            origin_func = DPDKdut.build_install_dpdk
+
+            def new_func(*args, **kwargw):
+                kwargw["extra_options"] = " ".join(
+                    [kwargw.get("extra_options", ""), added_param]
+                )
+                origin_func(*args, **kwargw)
+
+            DPDKdut.build_install_dpdk = new_func
+
+    @staticmethod
+    def decorator_dts_run():
+        import framework.dts as dts
+
+        origin_func = dts.dts_run_suite
+
+        def new_func(*args, **kwargs):
+            duts = args[0]
+            for dut in duts:
+                dut.send_expect(COMMAND_OF_CLOSE_ADDRESS_RANDOM, "#")
+
+            origin_func(*args, **kwargs)
+
+            for dut in duts:
+                dut.send_expect(COMMAND_OF_OPEN_ADDRESS_RANDOM, "#")
+
+        dts.dts_run_suite = new_func
+
+    @staticmethod
+    def decorator_send_expect():
+        import framework.ssh_pexpect as ssh_pexpect
+
+        origin_func = ssh_pexpect.SSHPexpect._SSHPexpect__flush
+
+        def new_func(self):
+            DELETE_CONTENT_PATTERN = r"^\s*\[?PEXPECT\]?#?\s*$"
+            befored_info = re.sub(
+                DELETE_CONTENT_PATTERN, "", self.session.before
+            ).strip()
+            if len(befored_info) > MIN_LENGTH_OF_FILTERED_OUTPUT and self.logger:
+                self.logger.info(f"Buffered info: {befored_info}")
+            origin_func(self)
+
+        ssh_pexpect.SSHPexpect._SSHPexpect__flush = new_func
+
+
+class _ASanConfig(object):
+    def __init__(
+        self,
+    ):
+        self.config = configparser.ConfigParser()
+        self.config.read(ASan_CONFIG_FILE_PATH)
+        self._filter_list = None
+        self._build_params = None
+
+    def _read_ASan_sect_conf(self, key):
+        return self.config.get(ASan_CONFIG_SECT, key)
+
+    def _set_ASan_filter(self):
+        try:
+            origin_filter_string = self._read_ASan_sect_conf(ASan_FILTER_BOUNDS)
+            self._filter_list = [
+                tuple(re.split(r":\s*", _filter))
+                for _filter in re.split(r",\s*", origin_filter_string)
+            ]
+        except KeyError:
+            self._filter_list = []
+
+    def _set_ASan_param(self):
+        try:
+            param_string = self._read_ASan_sect_conf(ASan_PARAM_KEY)
+        except KeyError:
+            param_string = ""
+        self._build_params = param_string
+
+    @property
+    def filter_list(self):
+        self._set_ASan_filter()
+        return self._filter_list
+
+    @property
+    def build_param(self):
+        self._set_ASan_param()
+        return self._build_params
+
+
+class _OldExcelReport(object):
+    def __init__(self):
+        self._report_file = os.path.join(REPORT_OUTPUT_PATH, ORIGIN_TEST_REPORT_FILE)
+        self._workbook: xlrd.Book = xlrd.open_workbook(self._report_file)
+        self._sheet_obj: xlrd.sheet.Sheet = self._workbook.sheets()[0]
+        self._rows = self._sheet_obj.nrows
+        self.current_row_num = 0
+
+    def generator_rows(self):
+        while True:
+            if self.current_row_num >= self._rows:
+                raise IndexError
+            row_number_of_jump_to = yield self._sheet_obj.row(self.current_row_num)
+            row = (
+                row_number_of_jump_to
+                if row_number_of_jump_to is not None
+                else self.current_row_num + 1
+            )
+            self.current_row_num = row
+
+
+class _OldExcelReportReader(object):
+    def __init__(self):
+        self._old_report = _OldExcelReport()
+        self._header_line_num = 1
+        self._test_env_content_column = None
+        self._test_suite_content_column = None
+        self._gen_report_rows = self._old_report.generator_rows()
+        next(self._gen_report_rows)
+        self._report_content_dict = dict()
+        self._current_suite = None
+
+    def get_report_info(self):
+        try:
+            self._get_first_line()
+            self._get_test_env()
+            self._get_cases_result()
+        except IndexError:
+            pass
+        return self._report_content_dict
+
+    def _get_first_line(self):
+        header_row_title = self._gen_report_rows.send(self._header_line_num - 1)
+        header_row_content = self._gen_report_rows.send(self._header_line_num)
+        cell_num = 0
+        while header_row_title[cell_num].value != "Test suite":
+            header_cell_title: str = header_row_title[cell_num].value
+            header_cell_content = header_row_content[cell_num].value
+            self._report_content_dict[
+                header_cell_title.lower().replace(" ", "_")
+            ] = header_cell_content
+            cell_num = cell_num + 1
+        self._test_env_content_column = cell_num - 1
+        self._test_suite_content_column = cell_num
+
+    @staticmethod
+    def _get_value_from_cell(cells_list_of_row):
+        return [cell.value for cell in cells_list_of_row]
+
+    def _get_test_env(self):
+        env_key_list = ["driver", "kdriver", "firmware", "package"]
+        for env_key in env_key_list:
+            env_info_row = next(self._gen_report_rows)
+            env_cell_value = env_info_row[self._test_env_content_column].value
+            if env_cell_value:
+                env_value = env_cell_value.split(": ")[1]
+                self._report_content_dict[env_key] = env_value
+            else:
+                self._report_content_dict[env_key] = None
+                # back to previous line
+                self._gen_report_rows.send(self._old_report.current_row_num - 1)
+
+    def _get_cases_result(self):
+        for row_cells in self._gen_report_rows:
+            suite_content_column_begin = self._test_suite_content_column
+            suite_content_column_end = self._test_suite_content_column + 3
+            suite_name, case_name, original_result_msg = self._get_value_from_cell(
+                row_cells[suite_content_column_begin:suite_content_column_end]
+            )
+            EMPTY_LINE_CONDITION = not suite_name and not case_name
+            NO_CASE_LINE_CONDITION = not case_name
+            SUITE_BEGIN_LINE_CONDITON = suite_name
+            if EMPTY_LINE_CONDITION or NO_CASE_LINE_CONDITION:
+                continue
+
+            if SUITE_BEGIN_LINE_CONDITON:
+                self._add_suite_info(suite_name)
+
+            self._add_case_info(case_name, original_result_msg)
+
+    def _add_suite_info(self, _suite):
+        self._report_content_dict.setdefault(_suite, dict())
+        self._current_suite = _suite
+
+    def _add_case_info(self, _case, _result_msg):
+        self._report_content_dict.get(self._current_suite)[_case] = _result_msg
+
+
+class _SuiteLogReader(object):
+    def __init__(self, suite_name):
+        self._suite_name = suite_name
+
+    @contextmanager
+    def suite_log_file(self):
+        from framework.test_case import TestCase
+        from framework.utils import get_subclasses
+
+        suite_full_name = "TestSuite_" + self._suite_name
+        suite_module = __import__(
+            "tests." + suite_full_name, fromlist=[suite_full_name]
+        )
+        suite_class_name = [
+            test_case_name
+            for test_case_name, _ in get_subclasses(suite_module, TestCase)
+        ][0]
+        log_file_path = os.path.join(REPORT_OUTPUT_PATH, suite_class_name)
+        log_file_obj = open(log_file_path + ".log", "r")
+        yield log_file_obj
+        log_file_obj.close()
+
+
+class _NewReport(object):
+    def __init__(self):
+        self._excel_report_file = os.path.join(REPORT_OUTPUT_PATH, NEW_TEST_REPORT_FILE)
+        self._json_report_file = os.path.join(REPORT_OUTPUT_PATH, NEW_JSON_REPORT_FILE)
+        self._stats_report_file = os.path.join(
+            REPORT_OUTPUT_PATH, NEW_STATS_REPORT_FILE
+        )
+        self._remove_history_asan_report()
+        self._excel_report = ExcelReporter(self._excel_report_file)
+        self._json_report = JSONReporter(self._json_report_file)
+        self._stats_report = StatsReporter(self._stats_report_file)
+        self._result_obj = Result()
+        self._old_report_reader = _OldExcelReportReader()
+        self._old_report_content: dict = self._old_report_reader.get_report_info()
+        self._new_suites_result = dict()
+        self._ASan_filter = _ASanConfig().filter_list
+        self._current_case = None
+        self._current_suite = None
+        self._filtered_line_cache = []
+        self._filter_begin = None
+        self._filter_end = None
+
+    def process_report_header(self):
+        head_key_list = [
+            "dut",
+            "kdriver",
+            "firmware",
+            "package",
+            "driver",
+            "dpdk_version",
+            "target",
+            "nic",
+        ]
+        for head_key in head_key_list:
+            head_value = self._old_report_content.setdefault(head_key, None)
+            self._old_report_content.pop(head_key)
+            setattr(self._result_obj, head_key, head_value)
+
+    def process_report_detail(self):
+        for suite in self._old_report_content.keys():
+            self._get_suite_new_result(suite)
+            self._parse_suite_result_to_result_obj()
+
+    def _get_suite_new_result(self, suite):
+        suite_log_reader = _SuiteLogReader(suite)
+        self._current_suite = suite
+        gen_suite_lines = suite_log_reader.suite_log_file()
+        self._get_case_result(gen_suite_lines)
+
+    def _parse_suite_result_to_result_obj(self):
+        self._result_obj.test_suite = self._current_suite
+        for case in self._old_report_content[self._current_suite]:
+            self._result_obj.test_case = case
+            if case in self._new_suites_result:
+                self._result_obj._Result__set_test_case_result(
+                    *self._new_suites_result[case]
+                )
+            else:
+                origin_result = self._get_origin_case_result(case)
+                self._result_obj._Result__set_test_case_result(*origin_result)
+
+    def save_report(self):
+        for report in (self._excel_report, self._json_report, self._stats_report):
+            report.save(self._result_obj)
+
+    def _remove_history_asan_report(self):
+        for file in (
+            self._excel_report_file,
+            self._json_report_file,
+            self._stats_report_file,
+        ):
+            if os.path.exists(file):
+                os.remove(file)
+
+    def _get_origin_case_result(self, case_name):
+        origin_cases_result: dict = self._old_report_content.get(self._current_suite)
+        origin_case_result: str = origin_cases_result.get(case_name)
+        CASE_RESULT_AND_MSG_PATTERN = r"(\S+)\s?(.*)"
+        result, msg = re.search(
+            CASE_RESULT_AND_MSG_PATTERN, origin_case_result
+        ).groups()
+        if msg:
+            msg = msg.replace("'", "").replace('"', "")
+
+        return result, msg
+
+    def _get_case_result(self, suite_log_reader):
+        with suite_log_reader as log_file:
+            for line in log_file:
+                self._filter_asan_except(line)
+
+            self._log_file_end_handler()
+
+    def _filter_asan_except(self, line):
+        CASE_LOG_BEGIN_PATTERN = r"Test Case test_(\w+) Begin"
+        case_begin_match = re.search(CASE_LOG_BEGIN_PATTERN, line)
+
+        if case_begin_match:
+            case_name = case_begin_match.groups()[0]
+            self._case_begin_handler(case_name)
+            return
+
+        for filter_tuple in self._ASan_filter:
+            begin_filter, end_filter = filter_tuple
+            if begin_filter in line:
+                self._filter_matched_begin_handler(begin_filter, line)
+                return
+
+            if self._filter_begin is not None:
+                self._filter_matched_line_handler(line)
+                return
+
+            if end_filter in line:
+                self._filter_matched_end_handler(end_filter, line)
+                return
+
+    def _case_begin_handler(self, case_name):
+        self._save_previous_case_result_and_clean_env()
+        self._current_case = case_name
+
+    def _filter_matched_begin_handler(self, begin_key, line):
+        self._filter_begin = begin_key
+        self._filtered_line_cache.append(line)
+
+    def _filter_matched_line_handler(self, line):
+        self._filtered_line_cache.append(line)
+
+    def _filter_matched_end_handler(self, end_key, line):
+        self._filtered_line_cache.append(line)
+        self._filter_begin = end_key
+
+    def _log_file_end_handler(self):
+        self._save_previous_case_result_and_clean_env()
+
+    def _save_previous_case_result_and_clean_env(self):
+        exist_previous_case_condition = self._current_case is not None
+        origin_report_contain_previous_case_result = (
+            self._current_case in self._old_report_content.get(self._current_suite)
+        )
+
+        if exist_previous_case_condition and origin_report_contain_previous_case_result:
+            self._save_case_result()
+
+        self._filtered_line_cache.clear()
+        self._filter_begin = None
+        self._filter_end = None
+
+    def _save_case_result(self):
+        cached_content = self._get_filtered_cached_result()
+        if self._current_case in self._new_suites_result:
+            # Run multiple times and keep the last result
+            self._new_suites_result.pop(self._current_case)
+
+        if cached_content:
+            # filter hit scene
+            self._new_suites_result[self._current_case] = ("FAILED", cached_content)
+        else:
+            # filter not hit scene
+            self._new_suites_result[self._current_case] = self._get_origin_case_result(
+                self._current_case
+            )
+
+    def _get_filtered_cached_result(self):
+        ASan_FILTER_CONTENT_PATTERN = (
+            rf"{self._filter_begin}[\s\S]+(?!{self._filter_end})?"
+        )
+        key_search_result = re.findall(
+            ASan_FILTER_CONTENT_PATTERN, "".join(self._filtered_line_cache)
+        )
+
+        return key_search_result[0] if key_search_result else ""
-- 
2.20.1


  parent reply	other threads:[~2022-04-06 14:57 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-04-06 14:55 [RFC PATCH v1 00/15] merge DTS core files " Juraj Linkeš
2022-04-06 14:55 ` [RFC PATCH v1 01/15] dts: merge DTS dep/tclclient.tgz " Juraj Linkeš
2022-04-06 14:55 ` [RFC PATCH v1 02/15] dts: merge DTS dep/tgen.tgz " Juraj Linkeš
2022-04-06 14:55 ` [RFC PATCH v1 03/15] dts: merge DTS dts " Juraj Linkeš
2022-04-06 14:55 ` [RFC PATCH v1 04/15] dts: merge DTS framework/__init__.py " Juraj Linkeš
2022-04-06 14:55 ` Juraj Linkeš [this message]
2022-04-06 14:55 ` [RFC PATCH v1 06/15] dts: merge DTS framework/checkCase.py " Juraj Linkeš
2022-04-06 14:55 ` [RFC PATCH v1 07/15] dts: merge DTS framework/dts.py " Juraj Linkeš
2022-04-06 14:55 ` [RFC PATCH v1 08/15] dts: merge DTS framework/exception.py " Juraj Linkeš
2022-04-06 14:56 ` [RFC PATCH v1 09/15] dts: merge DTS framework/logger.py " Juraj Linkeš
2022-04-06 14:56 ` [RFC PATCH v1 10/15] dts: merge DTS framework/packet.py " Juraj Linkeš
2022-04-06 14:56 ` [RFC PATCH v1 11/15] dts: merge DTS framework/project_dpdk.py " Juraj Linkeš
2022-04-06 14:56 ` [RFC PATCH v1 12/15] dts: merge DTS framework/serializer.py " Juraj Linkeš
2022-04-06 14:56 ` [RFC PATCH v1 13/15] dts: merge DTS framework/utils.py " Juraj Linkeš
2022-04-06 14:56 ` [RFC PATCH v1 14/15] dts: merge DTS main.py " Juraj Linkeš
2022-04-06 14:56 ` [RFC PATCH v1 15/15] dts: merge DTS version.py " Juraj Linkeš
2022-04-07  5:04 ` [RFC PATCH v1 00/15] merge DTS core files " Jerin Jacob
2022-04-07  7:33   ` Thomas Monjalon
2022-04-11  7:41     ` Juraj Linkeš
2022-04-11 17:55       ` Honnappa Nagarahalli
2022-04-11 18:20         ` Owen Hilyard
2022-04-11 19:06   ` Honnappa Nagarahalli

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20220406145606.2913834-6-juraj.linkes@pantheon.tech \
    --to=juraj.linkes@pantheon.tech \
    --cc=Honnappa.Nagarahalli@arm.com \
    --cc=david.marchand@redhat.com \
    --cc=dev@dpdk.org \
    --cc=lijuan.tu@intel.com \
    --cc=ohilyard@iol.unh.edu \
    --cc=thomas@monjalon.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).