esp-idf/tools/tiny-test-fw/Utility/CaseConfig.py

226 lines
8.1 KiB
Python
Raw Normal View History

2017-10-09 22:44:55 -04:00
# Copyright 2015-2017 Espressif Systems (Shanghai) PTE LTD
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http:#www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Processing case config files.
This is mainly designed for CI, we need to auto create and assign test jobs.
Template Config File::
TestConfig:
app:
path: Users/Test/TinyTestFW/IDF/IDFApp.py
class: Example
dut:
path:
class:
config_file: /somewhere/config_file_for_runner
test_name: CI_test_job_1
Filter:
chip: ESP32
env_tag: default
CaseConfig:
- name: test_examples_protocol_https_request
# optional
extra_data: some extra data passed to case with kwarg extra_data
overwrite: # overwrite test configs
app:
path: Users/Test/TinyTestFW/IDF/IDFApp.py
class: Example
- name: xxx
"""
# TODO: add a function to use suitable import lib for python2 and python3
import imp
import yaml
import TestCase
def _convert_to_lower_case_bytes(item):
"""
bot filter is always lower case string.
this function will convert to all string to lower case.
Note: Unicode strings are converted to bytes.
"""
if isinstance(item, (tuple, list)):
output = [_convert_to_lower_case_bytes(v) for v in item]
elif type(item) == type(b''):
output = item.lower()
elif type(item) == type(u''):
output = item.encode().lower()
else:
output = item
return output
2017-10-09 22:44:55 -04:00
def _filter_one_case(test_method, case_filter):
""" Apply filter for one case (the filter logic is the same as described in ``filter_test_cases``) """
filter_result = True
# filter keys are lower case. Do map lower case keys with original keys.
key_mapping = {x.lower(): x for x in test_method.case_info.keys()}
for orig_key in case_filter:
key = key_mapping[orig_key]
2017-10-09 22:44:55 -04:00
if key in test_method.case_info:
# the filter key is both in case and filter
# we need to check if they match
filter_item = _convert_to_lower_case_bytes(case_filter[orig_key])
accepted_item = _convert_to_lower_case_bytes(test_method.case_info[key])
2017-10-09 22:44:55 -04:00
if isinstance(filter_item, (tuple, list)) \
and isinstance(accepted_item, (tuple, list)):
# both list/tuple, check if they have common item
filter_result = True if set(filter_item) & set(accepted_item) else False
elif isinstance(filter_item, (tuple, list)):
# filter item list/tuple, check if case accepted value in filter item list/tuple
filter_result = True if accepted_item in filter_item else False
elif isinstance(accepted_item, (tuple, list)):
# accepted item list/tuple, check if case filter value is in accept item list/tuple
filter_result = True if filter_item in accepted_item else False
else:
if type(filter_item) != type(accepted_item):
# This will catch silent ignores of test cases when Unicode and bytes are compared
raise AssertionError(filter_item, '!=', accepted_item)
2017-10-09 22:44:55 -04:00
# both string/int, just do string compare
filter_result = (filter_item == accepted_item)
else:
# key in filter only, which means the case supports all values for this filter key, match succeed
pass
if not filter_result:
# match failed
break
return filter_result
def filter_test_cases(test_methods, case_filter):
"""
filter test case. filter logic:
1. if filter key both in case attribute and filter:
* if both value is string/int, then directly compare
* if one is list/tuple, the other one is string/int, then check if string/int is in list/tuple
* if both are list/tuple, then check if they have common item
2. if only case attribute or filter have the key, filter succeed
3. will do case insensitive compare for string
2017-10-09 22:44:55 -04:00
for example, the following are match succeed scenarios
(the rule is symmetric, result is same if exchange values for user filter and case attribute):
* user case filter is ``chip: ["esp32", "esp32c"]``, case doesn't have ``chip`` attribute
* user case filter is ``chip: ["esp32", "esp32c"]``, case attribute is ``chip: "esp32"``
* user case filter is ``chip: "esp32"``, case attribute is ``chip: "esp32"``
:param test_methods: a list of test methods functions
:param case_filter: case filter
:return: filtered test methods
"""
filtered_test_methods = []
for test_method in test_methods:
if _filter_one_case(test_method, case_filter):
filtered_test_methods.append(test_method)
return filtered_test_methods
class Parser(object):
DEFAULT_CONFIG = {
"TestConfig": dict(),
"Filter": dict(),
"CaseConfig": [{"extra_data": None}],
}
@classmethod
def parse_config_file(cls, config_file):
"""
parse from config file and then update to default config.
:param config_file: config file path
:return: configs
"""
configs = cls.DEFAULT_CONFIG.copy()
if config_file:
with open(config_file, "r") as f:
configs.update(yaml.load(f))
return configs
@classmethod
def handle_overwrite_args(cls, overwrite):
"""
handle overwrite configs. import module from path and then get the required class.
:param overwrite: overwrite args
:return: dict of (original key: class)
"""
output = dict()
for key in overwrite:
_path = overwrite[key]["path"]
# TODO: add a function to use suitable import lib for python2 and python3
_module = imp.load_source(str(hash(_path)), overwrite[key]["path"])
output[key] = _module.__getattribute__(overwrite[key]["class"])
return output
@classmethod
def apply_config(cls, test_methods, config_file):
"""
apply config for test methods
:param test_methods: a list of test methods functions
:param config_file: case filter file
:return: filtered cases
"""
configs = cls.parse_config_file(config_file)
test_case_list = []
for _config in configs["CaseConfig"]:
_filter = configs["Filter"].copy()
_filter.update(_config)
_overwrite = cls.handle_overwrite_args(_filter.pop("overwrite", dict()))
_extra_data = _filter.pop("extra_data", None)
for test_method in test_methods:
if _filter_one_case(test_method, _filter):
test_case_list.append(TestCase.TestCase(test_method, _extra_data, **_overwrite))
return test_case_list
class Generator(object):
""" Case config file generator """
def __init__(self):
self.default_config = {
"TestConfig": dict(),
"Filter": dict(),
}
def set_default_configs(self, test_config, case_filter):
"""
:param test_config: "TestConfig" value
:param case_filter: "Filter" value
:return: None
"""
self.default_config = {"TestConfig": test_config, "Filter": case_filter}
def generate_config(self, case_configs, output_file):
"""
:param case_configs: "CaseConfig" value
:param output_file: output file path
:return: None
"""
config = self.default_config.copy()
config.update({"CaseConfig": case_configs})
with open(output_file, "w") as f:
yaml.dump(config, f)