Add airframe config parser

This commit is contained in:
Lorenz Meier 2015-07-29 19:13:28 +02:00
parent 6dff0b5ebf
commit 68aa8fed9e
5 changed files with 441 additions and 0 deletions

View File

@ -0,0 +1 @@
This folder contains a python library used by px_process_params.py

View File

@ -0,0 +1 @@
__all__ = ["srcscanner", "srcparser", "xmlout", "dokuwikiout", "dokuwikirpc"]

View File

@ -0,0 +1,342 @@
import sys
import re
class ParameterGroup(object):
"""
Single parameter group
"""
def __init__(self, name):
self.name = name
self.params = []
def AddParameter(self, param):
"""
Add parameter to the group
"""
self.params.append(param)
def GetName(self):
"""
Get parameter group name
"""
return self.name
def GetParams(self):
"""
Returns the parsed list of parameters. Every parameter is a Parameter
object. Note that returned object is not a copy. Modifications affect
state of the parser.
"""
return sorted(self.params,
key=lambda x: x.GetFieldValue("code"))
class Parameter(object):
"""
Single parameter
"""
# Define sorting order of the fields
priority = {
"board": 9,
"short_desc": 8,
"long_desc": 7,
"min": 5,
"max": 4,
"unit": 3,
# all others == 0 (sorted alphabetically)
}
def __init__(self, name, airframe_type, airframe_id, maintainer):
self.fields = {}
self.outputs = {}
self.name = name
self.type = airframe_type
self.id = airframe_id
self.maintainer = maintainer
def GetName(self):
return self.name
def GetType(self):
return self.type
def GetId(self):
return self.id
def GetMaintainer(self):
return self.maintainer
def SetField(self, code, value):
"""
Set named field value
"""
self.fields[code] = value
def SetOutput(self, code, value):
"""
Set named output value
"""
self.outputs[code] = value
def GetFieldCodes(self):
"""
Return list of existing field codes in convenient order
"""
keys = self.fields.keys()
keys = sorted(keys)
keys = sorted(keys, key=lambda x: self.priority.get(x, 0), reverse=True)
return keys
def GetFieldValue(self, code):
"""
Return value of the given field code or None if not found.
"""
fv = self.fields.get(code)
if not fv:
# required because python 3 sorted does not accept None
return ""
return self.fields.get(code)
def GetOutputCodes(self):
"""
Return list of existing output codes in convenient order
"""
keys = self.outputs.keys()
keys = sorted(keys)
keys = sorted(keys, key=lambda x: self.priority.get(x, 0), reverse=True)
return keys
def GetOutputValue(self, code):
"""
Return value of the given output code or None if not found.
"""
fv = self.outputs.get(code)
if not fv:
# required because python 3 sorted does not accept None
return ""
return self.outputs.get(code)
class SourceParser(object):
"""
Parses provided data and stores all found parameters internally.
"""
re_split_lines = re.compile(r'[\r\n]+')
re_comment_start = re.compile(r'^\#\s')
re_comment_content = re.compile(r'^\#\s*(.*)')
re_comment_tag = re.compile(r'@([a-zA-Z][a-zA-Z0-9_]*)\s*(.*)')
re_comment_end = re.compile(r'(.*?)\s*\#\n/')
re_cut_type_specifier = re.compile(r'[a-z]+$')
re_is_a_number = re.compile(r'^-?[0-9\.]')
re_remove_dots = re.compile(r'\.+$')
re_remove_carriage_return = re.compile('\n+')
valid_tags = set(["url", "maintainer", "output", "name", "type"])
# Order of parameter groups
priority = {
# All other groups = 0 (sort alphabetically)
"Miscellaneous": -10
}
def __init__(self):
self.param_groups = {}
def GetSupportedExtensions(self):
"""
Returns list of supported file extensions that can be parsed by this
parser. The parser uses any extension.
"""
return [""]
def Parse(self, path, contents):
"""
Incrementally parse program contents and append all found airframes
to the list.
"""
airframe_id = None
airframe_id = path.rsplit('/',1)[1].split('_',1)[0]
# Skip if not numeric
if (not self.IsNumber(airframe_id)):
return True
# This code is essentially a comment-parsing grammar. "state"
# represents parser state. It contains human-readable state
# names.
state = None
tags = {}
outputs = {}
for line in self.re_split_lines.split(contents):
line = line.strip()
# Ignore empty lines
if line == "":
continue
if state is None and self.re_comment_start.match(line):
state = "wait-short"
short_desc = None
long_desc = None
if state is not None and state != "comment-processed":
m = self.re_comment_end.search(line)
if m:
line = m.group(1)
last_comment_line = True
else:
last_comment_line = False
m = self.re_comment_content.match(line)
if m:
comment_content = m.group(1)
if comment_content == "":
# When short comment ends with empty comment line,
# start waiting for the next part - long comment.
if state == "wait-short-end":
state = "wait-long"
else:
m = self.re_comment_tag.match(comment_content)
if m:
tag, desc = m.group(1, 2)
if (tag == "output"):
key, text = desc.split(' ', 1)
outputs[key] = text;
else:
tags[tag] = desc
current_tag = tag
state = "wait-tag-end"
elif state == "wait-short":
# Store first line of the short description
short_desc = comment_content
state = "wait-short-end"
elif state == "wait-short-end":
# Append comment line to the short description
short_desc += "\n" + comment_content
elif state == "wait-long":
# Store first line of the long description
long_desc = comment_content
state = "wait-long-end"
elif state == "wait-long-end":
# Append comment line to the long description
long_desc += "\n" + comment_content
elif state == "wait-tag-end":
# Append comment line to the tag text
tags[current_tag] += "\n" + comment_content
else:
raise AssertionError(
"Invalid parser state: %s" % state)
elif not last_comment_line:
# Invalid comment line (inside comment, but not starting with
# "*" or "*/". Reset parsed content.
state = None
if last_comment_line:
state = "comment-processed"
else:
state = None
# Process parsed content
airframe_type = None
maintainer = "John Doe <john@example.com>"
airframe_name = None
# Done with file, store
for tag in tags:
if tag == "maintainer":
maintainer = tags[tag]
if tag == "type":
airframe_type = tags[tag]
if tag == "name":
airframe_name = tags[tag]
elif tag not in self.valid_tags:
sys.stderr.write("Aborting due to invalid documentation tag: '%s'\n" % tag)
return False
# Sanity check
if airframe_type == None:
sys.stderr.write("Aborting due to missing @type tag in file: '%s'\n" % path)
return False
if airframe_name == None:
sys.stderr.write("Aborting due to missing @name tag in file: '%s'\n" % path)
return False
# We already know this is an airframe config, so add it
param = Parameter(airframe_name, airframe_type, airframe_id, maintainer)
# Done with file, store
for tag in tags:
if tag == "maintainer":
maintainer = tags[tag]
if tag == "type":
airframe_type = tags[tag]
if tag == "name":
airframe_name = tags[tag]
else:
param.SetField(tag, tags[tag])
# Store outputs
for output in outputs:
param.SetOutput(output, outputs[output])
# Store the parameter
if airframe_type not in self.param_groups:
self.param_groups[airframe_type] = ParameterGroup(airframe_type)
self.param_groups[airframe_type].AddParameter(param)
return True
def IsNumber(self, numberString):
try:
float(numberString)
return True
except ValueError:
return False
def Validate(self):
"""
Validates the airframe meta data.
"""
seenParamNames = []
for group in self.GetParamGroups():
for param in group.GetParams():
name = param.GetName()
board = param.GetFieldValue("board")
# Check for duplicates
name_plus_board = name + "+" + board
for seenParamName in seenParamNames:
if seenParamName == name_plus_board:
sys.stderr.write("Duplicate parameter definition: {0}\n".format(name_plus_board))
return False
seenParamNames.append(name_plus_board)
# Validate values
default = param.GetDefault()
min = param.GetFieldValue("min")
max = param.GetFieldValue("max")
#sys.stderr.write("{0} default:{1} min:{2} max:{3}\n".format(name, default, min, max))
if default != "" and not self.IsNumber(default):
sys.stderr.write("Default value not number: {0} {1}\n".format(name, default))
return False
if min != "":
if not self.IsNumber(min):
sys.stderr.write("Min value not number: {0} {1}\n".format(name, min))
return False
if default != "" and float(default) < float(min):
sys.stderr.write("Default value is smaller than min: {0} default:{1} min:{2}\n".format(name, default, min))
return False
if max != "":
if not self.IsNumber(max):
sys.stderr.write("Max value not number: {0} {1}\n".format(name, max))
return False
if default != "" and float(default) > float(max):
sys.stderr.write("Default value is larger than max: {0} default:{1} max:{2}\n".format(name, default, max))
return False
return True
def GetParamGroups(self):
"""
Returns the parsed list of parameters. Every parameter is a Parameter
object. Note that returned object is not a copy. Modifications affect
state of the parser.
"""
groups = self.param_groups.values()
groups = sorted(groups, key=lambda x: x.GetName())
groups = sorted(groups, key=lambda x: self.priority.get(x.GetName(), 0), reverse=True)
return groups

View File

@ -0,0 +1,37 @@
import os
import re
import codecs
class SourceScanner(object):
"""
Traverses directory tree, reads all source files, and passes their contents
to the Parser.
"""
def ScanDir(self, srcdir, parser):
"""
Scans provided path and passes all found contents to the parser using
parser.Parse method.
"""
extensions = tuple(parser.GetSupportedExtensions())
for dirname, dirnames, filenames in os.walk(srcdir):
for filename in filenames:
if filename.endswith(extensions):
path = os.path.join(dirname, filename)
if not self.ScanFile(path, parser):
return False
return True
def ScanFile(self, path, parser):
"""
Scans provided file and passes its contents to the parser using
parser.Parse method.
"""
with codecs.open(path, 'r', 'utf-8') as f:
try:
contents = f.read()
except:
contents = ''
print('Failed reading file: %s, skipping content.' % path)
pass
return parser.Parse(path, contents)

View File

@ -0,0 +1,60 @@
import xml.etree.ElementTree as ET
import codecs
def indent(elem, level=0):
i = "\n" + level*" "
if len(elem):
if not elem.text or not elem.text.strip():
elem.text = i + " "
if not elem.tail or not elem.tail.strip():
elem.tail = i
for elem in elem:
indent(elem, level+1)
if not elem.tail or not elem.tail.strip():
elem.tail = i
else:
if level and (not elem.tail or not elem.tail.strip()):
elem.tail = i
class XMLOutput():
def __init__(self, groups, board):
xml_parameters = ET.Element("airframes")
xml_version = ET.SubElement(xml_parameters, "version")
xml_version.text = "1"
last_param_name = ""
board_specific_param_set = False
for group in groups:
xml_group = ET.SubElement(xml_parameters, "airframe_group")
xml_group.attrib["name"] = group.GetName()
for param in group.GetParams():
if (last_param_name == param.GetName() and not board_specific_param_set) or last_param_name != param.GetName():
xml_param = ET.SubElement(xml_group, "airframe")
xml_param.attrib["name"] = param.GetName()
xml_param.attrib["id"] = param.GetId()
xml_param.attrib["maintainer"] = param.GetMaintainer()
last_param_name = param.GetName()
for code in param.GetFieldCodes():
value = param.GetFieldValue(code)
if code == "board":
if value == board:
board_specific_param_set = True
xml_field = ET.SubElement(xml_param, code)
xml_field.text = value
else:
xml_group.remove(xml_param)
else:
xml_field = ET.SubElement(xml_param, code)
xml_field.text = value
for code in param.GetOutputCodes():
value = param.GetOutputValue(code)
xml_field = ET.SubElement(xml_param, "output")
xml_field.attrib["name"] = code
xml_field.text = value
if last_param_name != param.GetName():
board_specific_param_set = False
indent(xml_parameters)
self.xml_document = ET.ElementTree(xml_parameters)
def Save(self, filename):
self.xml_document.write(filename, encoding="UTF-8")