blob: f87091d618fb1b6dd890bd7f462f0442555a6dc3 [file] [log] [blame]
#!/usr/bin/env python3
# Copyright (c) 2015-2016 The Khronos Group Inc.
# Copyright (c) 2015-2016 Valve Corporation
# Copyright (c) 2015-2016 LunarG, Inc.
# Copyright (c) 2015-2016 Google Inc.
#
# 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.
#
# Author: Tobin Ehlis <tobin@lunarg.com>
import argparse
import os
import sys
import vulkan
import platform
# vk_layer_documentation_generate.py overview
# This script is intended to generate documentation based on vulkan layers
# It parses known validation layer headers for details of the validation checks
# It parses validation layer source files for specific code where checks are implemented
# structs in a human-readable txt format, as well as utility functions
# to print enum values as strings
# NOTE : Initially the script is performing validation of a hand-written document
# Right now it does 3 checks:
# 1. Verify ENUM codes declared in source are documented
# 2. Verify ENUM codes in document are declared in source
# 3. Verify API function names in document are in the actual API header (vulkan.py)
# Currently script will flag errors in all of these cases
# TODO : Need a formal specification of the syntax for doc generation
# Initially, these are the basics:
# 1. Validation checks have unique ENUM values defined in validation layer header
# 2. ENUM includes comments for 1-line overview of check and more detailed description
# 3. Actual code implementing checks includes ENUM value in callback
# 4. Code to test checks should include reference to ENUM
# TODO : Need list of known validation layers to use as default input
# Just a couple of flat lists right now, but may need to make this input file
# or at least a more dynamic data structure
layer_inputs = { 'draw_state' : {'header' : 'layers/core_validation_error_enums.h',
'source' : 'layers/core_validation.cpp',
'generated' : False,
'error_enum' : 'DRAW_STATE_ERROR'},
'shader_checker' : {'header' : 'layers/core_validation_error_enums.h',
'source' : 'layers/core_validation.cpp',
'generated' : False,
'error_enum' : 'SHADER_CHECKER_ERROR'},
'mem_tracker' : {'header' : 'layers/core_validation_error_enums.h',
'source' : 'layers/core_validation.cpp',
'generated' : False,
'error_enum' : 'MEM_TRACK_ERROR'},
'device_limits' : {'header' : 'layers/core_validation_error_enums.h',
'source' : 'layers/core_validation.cpp',
'generated' : False,
'error_enum' : 'DEV_LIMITS_ERROR',},
'object_tracker' : {'header' : 'layers/object_tracker.h',
'source' : 'layers/object_tracker.cpp',
'generated' : False,
'error_enum' : 'OBJECT_TRACK_ERROR',},
'threading' : {'header' : 'layers/threading.h',
'source' : 'dbuild/layers/threading.cpp',
'generated' : True,
'error_enum' : 'THREADING_CHECKER_ERROR'},
'image' : {'header' : 'layers/image.h',
'source' : 'layers/image.cpp',
'generated' : False,
'error_enum' : 'IMAGE_ERROR',},
'swapchain' : {'header' : 'layers/swapchain.h',
'source' : 'layers/swapchain.cpp',
'generated' : False,
'error_enum' : 'SWAPCHAIN_ERROR',},
'parameter_validation' : {'header' : 'layers/parameter_validation_utils.h',
'source' : 'layers/parameter_validation.cpp',
'generated' : False,
'error_enum' : 'ErrorCode',},
}
builtin_headers = [layer_inputs[ln]['header'] for ln in layer_inputs]
builtin_source = [layer_inputs[ln]['source'] for ln in layer_inputs]
builtin_tests = ['tests/layer_validation_tests.cpp', ]
# List of extensions in layers that are included in documentation, but not in vulkan.py API set
layer_extension_functions = ['objTrackGetObjects', 'objTrackGetObjectsOfType']
def handle_args():
parser = argparse.ArgumentParser(description='Generate layer documenation from source.')
parser.add_argument('--in_headers', required=False, default=builtin_headers, help='The input layer header files from which code will be generated.')
parser.add_argument('--in_source', required=False, default=builtin_source, help='The input layer source files from which code will be generated.')
parser.add_argument('--test_source', required=False, default=builtin_tests, help='The input test source files from which code will be generated.')
parser.add_argument('--layer_doc', required=False, default='layers/vk_validation_layer_details.md', help='Existing layer document to be validated against actual layers.')
parser.add_argument('--validate', action='store_true', default=False, help='Validate that there are no mismatches between layer documentation and source. This includes cross-checking the validation checks, and making sure documented Vulkan API calls exist.')
parser.add_argument('--print_structs', action='store_true', default=False, help='Primarily a debug option that prints out internal data structs used to generate layer docs.')
parser.add_argument('--print_doc_checks', action='store_true', default=False, help='Primarily a debug option that prints out all of the checks that are documented.')
return parser.parse_args()
# Little helper class for coloring cmd line output
class bcolors:
def __init__(self):
self.GREEN = '\033[0;32m'
self.RED = '\033[0;31m'
self.YELLOW = '\033[1;33m'
self.ENDC = '\033[0m'
if 'Linux' != platform.system():
self.GREEN = ''
self.RED = ''
self.YELLOW = ''
self.ENDC = ''
def green(self):
return self.GREEN
def red(self):
return self.RED
def yellow(self):
return self.YELLOW
def endc(self):
return self.ENDC
# Class to parse the validation layer test source and store testnames
class TestParser:
def __init__(self, test_file_list, test_group_name=['VkLayerTest', 'VkWsiEnabledLayerTest']):
self.test_files = test_file_list
self.tests_set = set()
self.test_trigger_txt_list = []
for tg in test_group_name:
self.test_trigger_txt_list.append('TEST_F(%s' % tg)
#print('Test trigger test list: %s' % (self.test_trigger_txt_list))
# Parse test files into internal data struct
def parse(self):
# For each test file, parse test names into set
grab_next_line = False # handle testname on separate line than wildcard
for test_file in self.test_files:
with open(test_file) as tf:
for line in tf:
if True in [line.strip().startswith(comment) for comment in ['//', '/*']]:
continue
if True in [ttt in line for ttt in self.test_trigger_txt_list]:
#print('Test wildcard in line: %s' % (line))
testname = line.split(',')[-1]
testname = testname.strip().strip(' {)')
#print('Inserting test: "%s"' % (testname))
if ('' == testname):
grab_next_line = True
continue
self.tests_set.add(testname)
if grab_next_line: # test name on its own line
grab_next_line = False
testname = testname.strip().strip(' {)')
self.tests_set.add(testname)
# Class to parse the layer source code and store details in internal data structs
class LayerParser:
def __init__(self, header_file_list, source_file_list):
self.header_files = header_file_list
self.source_files = source_file_list
self.layer_dict = {}
self.api_dict = {}
# Parse layer header files into internal dict data structs
def parse(self):
# For each header file, parse details into dicts
# TODO : Should have a global dict element to track overall list of checks
store_enum = False
for layer_name in layer_inputs:
hf = layer_inputs[layer_name]['header']
self.layer_dict[layer_name] = {} # initialize a new dict for this layer
self.layer_dict[layer_name]['CHECKS'] = [] # enum of checks is stored in a list
#print('Parsing header file %s as layer name %s' % (hf, layer_name))
with open(hf) as f:
for line in f:
if True in [line.strip().startswith(comment) for comment in ['//', '/*']]:
#print("Skipping comment line: %s" % line)
# For now skipping lines starting w/ comment, may use these to capture
# documentation in the future
continue
# Find enums
if store_enum:
if '}' in line: # we're done with enum definition
store_enum = False
continue
# grab the enum name as a unique check
if ',' in line:
# TODO : When documentation for a check is contained in the source,
# this is where we should also capture that documentation so that
# it can then be transformed into desired doc format
enum_name = line.split(',')[0].strip()
# Flag an error if we have already seen this enum
if enum_name in self.layer_dict[layer_name]['CHECKS']:
print('ERROR : % layer has duplicate error enum: %s' % (layer_name, enum_name))
self.layer_dict[layer_name]['CHECKS'].append(enum_name)
# If the line includes 'enum' and the expected enum name, start capturing enums
if False not in [ex in line for ex in ['enum', layer_inputs[layer_name]['error_enum']]]:
store_enum = True
# For each source file, parse into dicts
for sf in self.source_files:
#print('Parsing source file %s' % sf)
pass
# TODO : In the source file we want to see where checks actually occur
# Need to build function tree of checks so that we know all of the
# checks that occur under a top-level Vulkan API call
# Eventually in the validation we can flag ENUMs that aren't being
# used in the source, and we can document source code lines as well
# as Vulkan API calls where each specific ENUM check is made
def print_structs(self):
print('This is where I print the data structs')
for layer in self.layer_dict:
print('Layer %s has %i checks:\n%s' % (layer, len(self.layer_dict[layer]['CHECKS'])-1, "\n\t".join(self.layer_dict[layer]['CHECKS'])))
# Class to parse hand-written md layer documentation into a dict and then validate its contents
class LayerDoc:
def __init__(self, source_file):
self.layer_doc_filename = source_file
self.txt_color = bcolors()
# Main data struct to store info from layer doc
self.layer_doc_dict = {}
# Comprehensive list of all validation checks recorded in doc
self.enum_list = []
# Parse the contents of doc into data struct
def parse(self):
layer_name = 'INIT'
parse_layer_details = False
detail_trigger = '| Check | '
parse_pending_work = False
pending_trigger = ' Pending Work'
parse_overview = False
overview_trigger = ' Overview'
enum_prefix = ''
with open(self.layer_doc_filename) as f:
for line in f:
if parse_pending_work:
if '.' in line and line.strip()[0].isdigit():
todo_item = line.split('.')[1].strip()
self.layer_doc_dict[layer_name]['pending'].append(todo_item)
if pending_trigger in line and '##' in line:
parse_layer_details = False
parse_pending_work = True
parse_overview = False
self.layer_doc_dict[layer_name]['pending'] = []
if parse_layer_details:
# Grab details but skip the fomat line with a bunch of '-' chars
if '|' in line and line.count('-') < 20:
detail_sections = line.split('|')
#print("Details elements from line %s: %s" % (line, detail_sections))
check_name = '%s%s' % (enum_prefix, detail_sections[3].strip())
self.enum_list.append(check_name)
self.layer_doc_dict[layer_name][check_name] = {}
self.layer_doc_dict[layer_name][check_name]['summary_txt'] = detail_sections[1].strip()
self.layer_doc_dict[layer_name][check_name]['details_txt'] = detail_sections[2].strip()
self.layer_doc_dict[layer_name][check_name]['api_list'] = detail_sections[4].split()
self.layer_doc_dict[layer_name][check_name]['tests'] = detail_sections[5].split()
self.layer_doc_dict[layer_name][check_name]['notes'] = detail_sections[6].strip()
# strip any unwanted commas from api and test names
self.layer_doc_dict[layer_name][check_name]['api_list'] = [a.strip(',') for a in self.layer_doc_dict[layer_name][check_name]['api_list']]
test_list = [a.strip(',') for a in self.layer_doc_dict[layer_name][check_name]['tests']]
self.layer_doc_dict[layer_name][check_name]['tests'] = [a.split('.')[-1] for a in test_list]
# Trigger details parsing when we have table header
if detail_trigger in line:
parse_layer_details = True
parse_pending_work = False
parse_overview = False
enum_txt = line.split('|')[3]
if '*' in enum_txt:
enum_prefix = enum_txt.split()[-1].strip('*').strip()
#print('prefix: %s' % enum_prefix)
if parse_overview:
self.layer_doc_dict[layer_name]['overview'] += line
if overview_trigger in line and '##' in line:
parse_layer_details = False
parse_pending_work = False
parse_overview = True
layer_name = line.split()[1]
self.layer_doc_dict[layer_name] = {}
self.layer_doc_dict[layer_name]['overview'] = ''
# Verify that checks, tests and api references in layer doc match reality
# Report API calls from doc that are not found in API
# Report checks from doc that are not in actual layers
# Report checks from layers that are not captured in doc
# Report checks from doc that do not have a valid test
def validate(self, layer_dict, tests_set):
#print("tests_set: %s" % (tests_set))
# Count number of errors found and return it
errors_found = 0
warnings_found = 0
# A few checks that are allowed to not have tests
no_test_checks = ['DRAWSTATE_INTERNAL_ERROR', 'DRAWSTATE_OUT_OF_MEMORY', 'MEMTRACK_INTERNAL_ERROR', 'OBJTRACK_INTERNAL_ERROR']
# First we'll go through the doc datastructures and flag any issues
for chk in self.enum_list:
doc_layer_found = False
for real_layer in layer_dict:
if chk in layer_dict[real_layer]['CHECKS']:
#print('Found actual layer check %s in doc' % (chk))
doc_layer_found = True
continue
if not doc_layer_found:
print(self.txt_color.red() + 'Actual layers do not contain documented check: %s' % (chk) + self.txt_color.endc())
errors_found += 1
# Now go through API names in doc and verify they're real
# First we're going to transform proto names from vulkan.py into single list
core_api_names = [p.name for p in vulkan.core.protos]
wsi_s_names = [p.name for p in vulkan.ext_khr_surface.protos]
wsi_ds_names = [p.name for p in vulkan.ext_khr_device_swapchain.protos]
dbg_rpt_names = [p.name for p in vulkan.ext_debug_report.protos]
api_names = core_api_names + wsi_s_names + wsi_ds_names + dbg_rpt_names
for ln in self.layer_doc_dict:
for chk in self.layer_doc_dict[ln]:
if chk in ['overview', 'pending']:
continue
for api in self.layer_doc_dict[ln][chk]['api_list']:
if api[2:] not in api_names and api not in layer_extension_functions:
print(self.txt_color.red() + 'Doc references invalid function: %s' % (api) + self.txt_color.endc())
errors_found += 1
# For now warn on missing or invalid tests
for test in self.layer_doc_dict[ln][chk]['tests']:
if '*' in test:
# naive way to handle wildcards, just make sure we have matches on parts
test_parts = test.split('*')
for part in test_parts:
part_found = False
for t in tests_set:
if part in t:
part_found = True
break
if not part_found:
print(self.txt_color.red() + 'Validation check %s has missing or invalid test : %s' % (chk, test))
errors_found += 1
break
elif test not in tests_set and not chk.endswith('_NONE'):
if test == 'TODO':
if chk not in no_test_checks:
warnings_found += 1
else:
print(self.txt_color.red() + 'Validation check %s has missing or invalid test : %s' % (chk, test))
errors_found += 1
# Now go through all of the actual checks in the layers and make sure they're covered in the doc
for ln in layer_dict:
for chk in layer_dict[ln]['CHECKS']:
if chk not in self.enum_list:
print(self.txt_color.red() + 'Doc is missing check: %s' % (chk) + self.txt_color.endc())
errors_found += 1
return (errors_found, warnings_found)
# Print all of the checks captured in the doc
def print_checks(self):
print('Checks captured in doc:\n%s' % ('\n\t'.join(self.enum_list)))
def main(argv=None):
# Parse args
opts = handle_args()
# Create parser for layer files
layer_parser = LayerParser(opts.in_headers, opts.in_source)
# Parse files into internal data structs
layer_parser.parse()
# Parse test files
test_parser = TestParser(opts.test_source)
test_parser.parse()
# Generate requested types of output
if opts.print_structs: # Print details of internal data structs
layer_parser.print_structs()
layer_doc = LayerDoc(opts.layer_doc)
layer_doc.parse()
if opts.print_doc_checks:
layer_doc.print_checks()
if opts.validate:
(num_errors, num_warnings) = layer_doc.validate(layer_parser.layer_dict, test_parser.tests_set)
txt_color = bcolors()
if (0 == num_warnings):
print(txt_color.green() + 'No warning cases found between %s and implementation' % (os.path.basename(opts.layer_doc)) + txt_color.endc())
else:
print(txt_color.yellow() + 'Found %s warnings due to missing tests. Missing tests are labeled as "TODO" in "%s."' % (num_warnings, opts.layer_doc))
if (0 == num_errors):
print(txt_color.green() + 'No mismatches found between %s and implementation' % (os.path.basename(opts.layer_doc)) + txt_color.endc())
else:
return num_errors
return 0
if __name__ == "__main__":
sys.exit(main())