import logging import os import sys import datetime import inspect import numpy as np # The following 2 imports *ARE* used. Do not remove either. import xml.dom.minidom from xml.dom.minidom import parseString from lxml import etree from sas.sascalc.data_util.nxsunit import Converter # For saving individual sections of data from ..data_info import Data1D, Data2D, DataInfo, plottable_1D, plottable_2D, \ Collimation, TransmissionSpectrum, Detector, Process, Aperture, \ combine_data_info_with_plottable as combine_data from ..loader_exceptions import FileContentsException, DefaultReaderException, \ DataReaderException from . import xml_reader from .xml_reader import XMLreader from .cansas_constants import CansasConstants, CurrentLevel logger = logging.getLogger(__name__) PREPROCESS = "xmlpreprocess" ENCODING = "encoding" RUN_NAME_DEFAULT = "None" INVALID_SCHEMA_PATH_1_1 = "{0}/sas/sascalc/dataloader/readers/schema/cansas1d_invalid_v1_1.xsd" INVALID_SCHEMA_PATH_1_0 = "{0}/sas/sascalc/dataloader/readers/schema/cansas1d_invalid_v1_0.xsd" INVALID_XML = "\n\nThe loaded xml file, {0} does not fully meet the CanSAS v1.x specification. SasView loaded " + \ "as much of the data as possible.\n\n" CONSTANTS = CansasConstants() CANSAS_FORMAT = CONSTANTS.format CANSAS_NS = CONSTANTS.names ALLOW_ALL = True class Reader(XMLreader): cansas_version = "1.0" base_ns = "{cansas1d/1.0}" cansas_defaults = None type_name = "canSAS" invalid = True frm = "" # Log messages and errors logging = None errors = set() # Namespace hierarchy for current xml_file object names = None ns_list = None # Temporary storage location for loading multiple data sets in a single file current_data1d = None data = None # Wildcards type = ["XML files (*.xml)|*.xml", "SasView Save Files (*.svs)|*.svs"] # List of allowed extensions ext = ['.xml', '.XML', '.svs', '.SVS'] # Flag to bypass extension check allow_all = True def reset_state(self): """ Resets the class state to a base case when loading a new data file so previous data files do not appear a second time """ self.current_datainfo = None self.current_dataset = None self.current_data1d = None self.data = [] self.process = Process() self.transspectrum = TransmissionSpectrum() self.aperture = Aperture() self.collimation = Collimation() self.detector = Detector() self.names = [] self.cansas_defaults = {} self.output = [] self.ns_list = None self.logging = [] self.encoding = None def read(self, xml_file, schema_path="", invalid=True): if schema_path != "" or invalid != True: # read has been called from self.get_file_contents because xml file doens't conform to schema _, self.extension = os.path.splitext(os.path.basename(xml_file)) return self.get_file_contents(xml_file=xml_file, schema_path=schema_path, invalid=invalid) # Otherwise, read has been called by the data loader - file_reader_base_class handles this return super(XMLreader, self).read(xml_file) def get_file_contents(self, xml_file=None, schema_path="", invalid=True): # Reset everything since we're loading a new file self.reset_state() self.invalid = invalid if xml_file is None: xml_file = self.f_open.name # We don't sure f_open since lxml handles opnening/closing files try: # Raises FileContentsException self.load_file_and_schema(xml_file, schema_path) # Parse each SASentry entry_list = self.xmlroot.xpath('/ns:SASroot/ns:SASentry', namespaces={ 'ns': self.cansas_defaults.get( "ns") }) self.is_cansas(self.extension) self.set_processing_instructions() for entry in entry_list: self._parse_entry(entry) self.data_cleanup() except FileContentsException as fc_exc: # File doesn't meet schema - try loading with a less strict schema base_name = xml_reader.__file__ base_name = base_name.replace("\\", "/") base = base_name.split("/sas/")[0] if self.cansas_version == "1.1": invalid_schema = INVALID_SCHEMA_PATH_1_1.format(base, self.cansas_defaults.get("schema")) else: invalid_schema = INVALID_SCHEMA_PATH_1_0.format(base, self.cansas_defaults.get("schema")) self.set_schema(invalid_schema) if self.invalid: try: # Load data with less strict schema self.read(xml_file, invalid_schema, False) # File can still be read but doesn't match schema, so raise exception self.load_file_and_schema(xml_file) # Reload strict schema so we can find where error are in file invalid_xml = self.find_invalid_xml() if invalid_xml != "": basename, _ = os.path.splitext( os.path.basename(self.f_open.name)) invalid_xml = INVALID_XML.format(basename + self.extension) + invalid_xml raise DataReaderException(invalid_xml) # Handled by base class except FileContentsException as fc_exc: msg = "CanSAS Reader could not load the file {}".format(xml_file) if fc_exc.message is not None: # Propagate error messages from earlier msg = fc_exc.message if not self.extension in self.ext: # If the file has no associated loader raise DefaultReaderException(msg) raise FileContentsException(msg) pass else: raise fc_exc except Exception as e: # Convert all other exceptions to FileContentsExceptions raise FileContentsException(str(e)) finally: if not self.f_open.closed: self.f_open.close() def load_file_and_schema(self, xml_file, schema_path=""): base_name = xml_reader.__file__ base_name = base_name.replace("\\", "/") base = base_name.split("/sas/")[0] # Try and parse the XML file try: self.set_xml_file(xml_file) except etree.XMLSyntaxError: # File isn't valid XML so can't be loaded msg = "SasView cannot load {}.\nInvalid XML syntax".format(xml_file) raise FileContentsException(msg) self.cansas_version = self.xmlroot.get("version", "1.0") self.cansas_defaults = CANSAS_NS.get(self.cansas_version, "1.0") if schema_path == "": schema_path = "{}/sas/sascalc/dataloader/readers/schema/{}".format( base, self.cansas_defaults.get("schema").replace("\\", "/") ) self.set_schema(schema_path) def is_cansas(self, ext="xml"): """ Checks to see if the XML file is a CanSAS file :param ext: The file extension of the data file :raises FileContentsException: Raised if XML file isn't valid CanSAS """ if self.validate_xml(): # Check file is valid XML name = "{http://www.w3.org/2001/XMLSchema-instance}schemaLocation" value = self.xmlroot.get(name) # Check schema CanSAS version matches file CanSAS version if CANSAS_NS.get(self.cansas_version).get("ns") == value.rsplit(" ")[0]: return True if ext == "svs": return True # Why is this required? # If we get to this point then file isn't valid CanSAS logger.warning("File doesn't meet CanSAS schema. Trying to load anyway.") raise FileContentsException("The file is not valid CanSAS") def _parse_entry(self, dom, recurse=False): if not self._is_call_local() and not recurse: self.reset_state() if not recurse: self.current_datainfo = DataInfo() # Raises FileContentsException if file doesn't meet CanSAS schema self.invalid = False # Look for a SASentry self.data = [] self.parent_class = "SASentry" self.names.append("SASentry") self.current_datainfo.meta_data["loader"] = "CanSAS XML 1D" self.current_datainfo.meta_data[ PREPROCESS] = self.processing_instructions if self._is_call_local() and not recurse: basename, _ = os.path.splitext(os.path.basename(self.f_open.name)) self.current_datainfo.filename = basename + self.extension # Create an empty dataset if no data has been passed to the reader if self.current_dataset is None: self._initialize_new_data_set(dom) self.base_ns = "{" + CANSAS_NS.get(self.cansas_version).get("ns") + "}" # Loop through each child in the parent element for node in dom: attr = node.attrib name = attr.get("name", "") type = attr.get("type", "") # Get the element name and set the current names level tagname = node.tag.replace(self.base_ns, "") tagname_original = tagname # Skip this iteration when loading in save state information if tagname in ["fitting_plug_in", "pr_inversion", "invariant", "corfunc"]: continue # Get where to store content self.names.append(tagname_original) self.ns_list = CONSTANTS.iterate_namespace(self.names) # If the element is a child element, recurse if len(node.getchildren()) > 0: self.parent_class = tagname_original if tagname == 'SASdata': self._initialize_new_data_set(node) if isinstance(self.current_dataset, plottable_2D): x_bins = attr.get("x_bins", "") y_bins = attr.get("y_bins", "") if x_bins is not "" and y_bins is not "": self.current_dataset.shape = (x_bins, y_bins) else: self.current_dataset.shape = () # Recurse to access data within the group self._parse_entry(node, recurse=True) if tagname == "SASsample": self.current_datainfo.sample.name = name elif tagname == "beam_size": self.current_datainfo.source.beam_size_name = name elif tagname == "SAScollimation": self.collimation.name = name elif tagname == "aperture": self.aperture.name = name self.aperture.type = type self._add_intermediate() else: # TODO: Clean this up to make it faster (fewer if/elifs) if isinstance(self.current_dataset, plottable_2D): data_point = node.text unit = attr.get('unit', '') else: data_point, unit = self._get_node_value(node, tagname) # If this is a dataset, store the data appropriately if tagname == 'Run': self.current_datainfo.run_name[data_point] = name self.current_datainfo.run.append(data_point) elif tagname == 'Title': self.current_datainfo.title = data_point elif tagname == 'SASnote': self.current_datainfo.notes.append(data_point) # I and Q points elif tagname == 'I' and isinstance(self.current_dataset, plottable_1D): self.current_dataset.yaxis("Intensity", unit) self.current_dataset.y = np.append(self.current_dataset.y, data_point) elif tagname == 'Idev' and isinstance(self.current_dataset, plottable_1D): self.current_dataset.dy = np.append(self.current_dataset.dy, data_point) elif tagname == 'Q': self.current_dataset.xaxis("Q", unit) self.current_dataset.x = np.append(self.current_dataset.x, data_point) elif tagname == 'Qdev': self.current_dataset.dx = np.append(self.current_dataset.dx, data_point) elif tagname == 'dQw': self.current_dataset.dxw = np.append(self.current_dataset.dxw, data_point) elif tagname == 'dQl': self.current_dataset.dxl = np.append(self.current_dataset.dxl, data_point) elif tagname == 'Qmean': pass elif tagname == 'Shadowfactor': pass elif tagname == 'Sesans': self.current_datainfo.isSesans = bool(data_point) self.current_dataset.xaxis(attr.get('x_axis'), attr.get('x_unit')) self.current_dataset.yaxis(attr.get('y_axis'), attr.get('y_unit')) elif tagname == 'yacceptance': self.current_datainfo.sample.yacceptance = (data_point, unit) elif tagname == 'zacceptance': self.current_datainfo.sample.zacceptance = (data_point, unit) # I and Qx, Qy - 2D data elif tagname == 'I' and isinstance(self.current_dataset, plottable_2D): self.current_dataset.yaxis("Intensity", unit) self.current_dataset.data = np.fromstring(data_point, dtype=float, sep=",") elif tagname == 'Idev' and isinstance(self.current_dataset, plottable_2D): self.current_dataset.err_data = np.fromstring(data_point, dtype=float, sep=",") elif tagname == 'Qx': self.current_dataset.xaxis("Qx", unit) self.current_dataset.qx_data = np.fromstring(data_point, dtype=float, sep=",") elif tagname == 'Qy': self.current_dataset.yaxis("Qy", unit) self.current_dataset.qy_data = np.fromstring(data_point, dtype=float, sep=",") elif tagname == 'Qxdev': self.current_dataset.xaxis("Qxdev", unit) self.current_dataset.dqx_data = np.fromstring(data_point, dtype=float, sep=",") elif tagname == 'Qydev': self.current_dataset.yaxis("Qydev", unit) self.current_dataset.dqy_data = np.fromstring(data_point, dtype=float, sep=",") elif tagname == 'Mask': inter = [item == "1" for item in data_point.split(",")] self.current_dataset.mask = np.asarray(inter, dtype=bool) # Sample Information elif tagname == 'ID' and self.parent_class == 'SASsample': self.current_datainfo.sample.ID = data_point elif tagname == 'Title' and self.parent_class == 'SASsample': self.current_datainfo.sample.name = data_point elif tagname == 'thickness' and self.parent_class == 'SASsample': self.current_datainfo.sample.thickness = data_point self.current_datainfo.sample.thickness_unit = unit elif tagname == 'transmission' and self.parent_class == 'SASsample': self.current_datainfo.sample.transmission = data_point elif tagname == 'temperature' and self.parent_class == 'SASsample': self.current_datainfo.sample.temperature = data_point self.current_datainfo.sample.temperature_unit = unit elif tagname == 'details' and self.parent_class == 'SASsample': self.current_datainfo.sample.details.append(data_point) elif tagname == 'x' and self.parent_class == 'position': self.current_datainfo.sample.position.x = data_point self.current_datainfo.sample.position_unit = unit elif tagname == 'y' and self.parent_class == 'position': self.current_datainfo.sample.position.y = data_point self.current_datainfo.sample.position_unit = unit elif tagname == 'z' and self.parent_class == 'position': self.current_datainfo.sample.position.z = data_point self.current_datainfo.sample.position_unit = unit elif tagname == 'roll' and self.parent_class == 'orientation' and 'SASsample' in self.names: self.current_datainfo.sample.orientation.x = data_point self.current_datainfo.sample.orientation_unit = unit elif tagname == 'pitch' and self.parent_class == 'orientation' and 'SASsample' in self.names: self.current_datainfo.sample.orientation.y = data_point self.current_datainfo.sample.orientation_unit = unit elif tagname == 'yaw' and self.parent_class == 'orientation' and 'SASsample' in self.names: self.current_datainfo.sample.orientation.z = data_point self.current_datainfo.sample.orientation_unit = unit # Instrumental Information elif tagname == 'name' and self.parent_class == 'SASinstrument': self.current_datainfo.instrument = data_point # Detector Information elif tagname == 'name' and self.parent_class == 'SASdetector': self.detector.name = data_point elif tagname == 'SDD' and self.parent_class == 'SASdetector': self.detector.distance = data_point self.detector.distance_unit = unit elif tagname == 'slit_length' and self.parent_class == 'SASdetector': self.detector.slit_length = data_point self.detector.slit_length_unit = unit elif tagname == 'x' and self.parent_class == 'offset': self.detector.offset.x = data_point self.detector.offset_unit = unit elif tagname == 'y' and self.parent_class == 'offset': self.detector.offset.y = data_point self.detector.offset_unit = unit elif tagname == 'z' and self.parent_class == 'offset': self.detector.offset.z = data_point self.detector.offset_unit = unit elif tagname == 'x' and self.parent_class == 'beam_center': self.detector.beam_center.x = data_point self.detector.beam_center_unit = unit elif tagname == 'y' and self.parent_class == 'beam_center': self.detector.beam_center.y = data_point self.detector.beam_center_unit = unit elif tagname == 'z' and self.parent_class == 'beam_center': self.detector.beam_center.z = data_point self.detector.beam_center_unit = unit elif tagname == 'x' and self.parent_class == 'pixel_size': self.detector.pixel_size.x = data_point self.detector.pixel_size_unit = unit elif tagname == 'y' and self.parent_class == 'pixel_size': self.detector.pixel_size.y = data_point self.detector.pixel_size_unit = unit elif tagname == 'z' and self.parent_class == 'pixel_size': self.detector.pixel_size.z = data_point self.detector.pixel_size_unit = unit elif tagname == 'roll' and self.parent_class == 'orientation' and 'SASdetector' in self.names: self.detector.orientation.x = data_point self.detector.orientation_unit = unit elif tagname == 'pitch' and self.parent_class == 'orientation' and 'SASdetector' in self.names: self.detector.orientation.y = data_point self.detector.orientation_unit = unit elif tagname == 'yaw' and self.parent_class == 'orientation' and 'SASdetector' in self.names: self.detector.orientation.z = data_point self.detector.orientation_unit = unit # Collimation and Aperture elif tagname == 'length' and self.parent_class == 'SAScollimation': self.collimation.length = data_point self.collimation.length_unit = unit elif tagname == 'name' and self.parent_class == 'SAScollimation': self.collimation.name = data_point elif tagname == 'distance' and self.parent_class == 'aperture': self.aperture.distance = data_point self.aperture.distance_unit = unit elif tagname == 'x' and self.parent_class == 'size': self.aperture.size.x = data_point self.collimation.size_unit = unit elif tagname == 'y' and self.parent_class == 'size': self.aperture.size.y = data_point self.collimation.size_unit = unit elif tagname == 'z' and self.parent_class == 'size': self.aperture.size.z = data_point self.collimation.size_unit = unit # Process Information elif tagname == 'name' and self.parent_class == 'SASprocess': self.process.name = data_point elif tagname == 'description' and self.parent_class == 'SASprocess': self.process.description = data_point elif tagname == 'date' and self.parent_class == 'SASprocess': try: self.process.date = datetime.datetime.fromtimestamp(data_point) except: self.process.date = data_point elif tagname == 'SASprocessnote': self.process.notes.append(data_point) elif tagname == 'term' and self.parent_class == 'SASprocess': unit = attr.get("unit", "") dic = { "name": name, "value": data_point, "unit": unit } self.process.term.append(dic) # Transmission Spectrum elif tagname == 'T' and self.parent_class == 'Tdata': self.transspectrum.transmission = np.append(self.transspectrum.transmission, data_point) self.transspectrum.transmission_unit = unit elif tagname == 'Tdev' and self.parent_class == 'Tdata': self.transspectrum.transmission_deviation = np.append(self.transspectrum.transmission_deviation, data_point) self.transspectrum.transmission_deviation_unit = unit elif tagname == 'Lambda' and self.parent_class == 'Tdata': self.transspectrum.wavelength = np.append(self.transspectrum.wavelength, data_point) self.transspectrum.wavelength_unit = unit # Source Information elif tagname == 'wavelength' and (self.parent_class == 'SASsource' or self.parent_class == 'SASData'): self.current_datainfo.source.wavelength = data_point self.current_datainfo.source.wavelength_unit = unit elif tagname == 'wavelength_min' and self.parent_class == 'SASsource': self.current_datainfo.source.wavelength_min = data_point self.current_datainfo.source.wavelength_min_unit = unit elif tagname == 'wavelength_max' and self.parent_class == 'SASsource': self.current_datainfo.source.wavelength_max = data_point self.current_datainfo.source.wavelength_max_unit = unit elif tagname == 'wavelength_spread' and self.parent_class == 'SASsource': self.current_datainfo.source.wavelength_spread = data_point self.current_datainfo.source.wavelength_spread_unit = unit elif tagname == 'x' and self.parent_class == 'beam_size': self.current_datainfo.source.beam_size.x = data_point self.current_datainfo.source.beam_size_unit = unit elif tagname == 'y' and self.parent_class == 'beam_size': self.current_datainfo.source.beam_size.y = data_point self.current_datainfo.source.beam_size_unit = unit elif tagname == 'z' and self.parent_class == 'pixel_size': self.current_datainfo.source.data_point.z = data_point self.current_datainfo.source.beam_size_unit = unit elif tagname == 'radiation' and self.parent_class == 'SASsource': self.current_datainfo.source.radiation = data_point elif tagname == 'beam_shape' and self.parent_class == 'SASsource': self.current_datainfo.source.beam_shape = data_point # Everything else goes in meta_data else: new_key = self._create_unique_key(self.current_datainfo.meta_data, tagname) self.current_datainfo.meta_data[new_key] = data_point self.names.remove(tagname_original) length = 0 if len(self.names) > 1: length = len(self.names) - 1 self.parent_class = self.names[length] if not self._is_call_local() and not recurse: self.frm = "" self.current_datainfo.errors = set() for error in self.errors: self.current_datainfo.errors.add(error) self.data_cleanup() self.sort_one_d_data() self.sort_two_d_data() self.reset_data_list() return self.output[0], None def _is_call_local(self): if self.frm == "": inter = inspect.stack() self.frm = inter[2] mod_name = self.frm[1].replace("\\", "/").replace(".pyc", "") mod_name = mod_name.replace(".py", "") mod = mod_name.split("sas/") mod_name = mod[1] if mod_name != "sascalc/dataloader/readers/cansas_reader": return False return True def _add_intermediate(self): """ This method stores any intermediate objects within the final data set after fully reading the set. """ if self.parent_class == 'SASprocess': self.current_datainfo.process.append(self.process) self.process = Process() elif self.parent_class == 'SASdetector': self.current_datainfo.detector.append(self.detector) self.detector = Detector() elif self.parent_class == 'SAStransmission_spectrum': self.current_datainfo.trans_spectrum.append(self.transspectrum) self.transspectrum = TransmissionSpectrum() elif self.parent_class == 'SAScollimation': self.current_datainfo.collimation.append(self.collimation) self.collimation = Collimation() elif self.parent_class == 'aperture': self.collimation.aperture.append(self.aperture) self.aperture = Aperture() elif self.parent_class == 'SASdata': self.data.append(self.current_dataset) def _get_node_value(self, node, tagname): """ Get the value of a node and any applicable units :param node: The XML node to get the value of :param tagname: The tagname of the node """ #Get the text from the node and convert all whitespace to spaces units = '' node_value = node.text if node_value is not None: node_value = ' '.join(node_value.split()) else: node_value = "" # If the value is a float, compile with units. if self.ns_list.ns_datatype == "float": # If an empty value is given, set as zero. if node_value is None or node_value.isspace() \ or node_value.lower() == "nan": node_value = "0.0" #Convert the value to the base units node_value, units = self._unit_conversion(node, tagname, node_value) # If the value is a timestamp, convert to a datetime object elif self.ns_list.ns_datatype == "timestamp": if node_value is None or node_value.isspace(): pass else: try: node_value = \ datetime.datetime.fromtimestamp(node_value) except ValueError: node_value = None return node_value, units def _unit_conversion(self, node, tagname, node_value): """ A unit converter method used to convert the data included in the file to the default units listed in data_info :param node: XML node :param tagname: name of the node :param node_value: The value of the current dom node """ attr = node.attrib value_unit = '' err_msg = None default_unit = None if not isinstance(node_value, float): node_value = float(node_value) if 'unit' in attr and attr.get('unit') is not None: try: unit = attr['unit'] # Split the units to retain backwards compatibility with # projects, analyses, and saved data from v4.1.0 unit_list = unit.split("|") if len(unit_list) > 1: local_unit = unit_list[1] else: local_unit = unit unitname = self.ns_list.current_level.get("unit", "") if "SASdetector" in self.names: save_in = "detector" elif "aperture" in self.names: save_in = "aperture" elif "SAScollimation" in self.names: save_in = "collimation" elif "SAStransmission_spectrum" in self.names: save_in = "transspectrum" elif "SASdata" in self.names: x = np.zeros(1) y = np.zeros(1) self.current_data1d = Data1D(x, y) save_in = "current_data1d" elif "SASsource" in self.names: save_in = "current_datainfo.source" elif "SASsample" in self.names: save_in = "current_datainfo.sample" elif "SASprocess" in self.names: save_in = "process" else: save_in = "current_datainfo" default_unit = getattrchain(self, '.'.join((save_in, unitname))) if (local_unit and default_unit and local_unit.lower() != default_unit.lower() and local_unit.lower() != "none"): # Check local units - bad units raise KeyError #print("loading", tagname, node_value, local_unit, default_unit) data_conv_q = Converter(local_unit) value_unit = default_unit node_value = data_conv_q(node_value, units=default_unit) else: value_unit = local_unit except KeyError: # Do not throw an error for loading Sesans data in cansas xml # This is a temporary fix. if local_unit != "A" and local_unit != 'pol': err_msg = "CanSAS reader: unexpected " err_msg += "\"{0}\" unit [{1}]; " err_msg = err_msg.format(tagname, local_unit) err_msg += "expecting [{0}]".format(default_unit) value_unit = local_unit except Exception: err_msg = "CanSAS reader: unknown error converting " err_msg += "\"{0}\" unit [{1}]" err_msg = err_msg.format(tagname, local_unit) value_unit = local_unit elif 'unit' in attr: value_unit = attr['unit'] if err_msg: self.errors.add(err_msg) return node_value, value_unit def _initialize_new_data_set(self, node=None): if node is not None: for child in node: if child.tag.replace(self.base_ns, "") == "Idata": for i_child in child: if i_child.tag.replace(self.base_ns, "") == "Qx": self.current_dataset = plottable_2D() return self.current_dataset = plottable_1D(np.array(0), np.array(0)) ## Writing Methods def write(self, filename, datainfo): """ Write the content of a Data1D as a CanSAS XML file :param filename: name of the file to write :param datainfo: Data1D object """ # Create XML document doc, _ = self._to_xml_doc(datainfo) # Write the file file_ref = open(filename, 'wb') if self.encoding is None: self.encoding = "UTF-8" doc.write(file_ref, encoding=self.encoding, pretty_print=True, xml_declaration=True) file_ref.close() def _to_xml_doc(self, datainfo): """ Create an XML document to contain the content of a Data1D :param datainfo: Data1D object """ is_2d = False if issubclass(datainfo.__class__, Data2D): is_2d = True # Get PIs and create root element pi_string = self._get_pi_string() # Define namespaces and create SASroot object main_node = self._create_main_node() # Create ElementTree, append SASroot and apply processing instructions base_string = pi_string + self.to_string(main_node) base_element = self.create_element_from_string(base_string) doc = self.create_tree(base_element) # Create SASentry Element entry_node = self.create_element("SASentry") root = doc.getroot() root.append(entry_node) # Add Title to SASentry self.write_node(entry_node, "Title", datainfo.title) # Add Run to SASentry self._write_run_names(datainfo, entry_node) # Add Data info to SASEntry if is_2d: self._write_data_2d(datainfo, entry_node) else: self._write_data(datainfo, entry_node) # Transmission Spectrum Info # TODO: fix the writer to linearize all data, including T_spectrum # self._write_trans_spectrum(datainfo, entry_node) # Sample info self._write_sample_info(datainfo, entry_node) # Instrument info instr = self._write_instrument(datainfo, entry_node) # Source self._write_source(datainfo, instr) # Collimation self._write_collimation(datainfo, instr) # Detectors self._write_detectors(datainfo, instr) # Processes info self._write_process_notes(datainfo, entry_node) # Note info self._write_notes(datainfo, entry_node) # Return the document, and the SASentry node associated with # the data we just wrote # If the calling function was not the cansas reader, return a minidom # object rather than an lxml object. self.frm = inspect.stack()[1] doc, entry_node = self._check_origin(entry_node, doc) return doc, entry_node def write_node(self, parent, name, value, attr=None): """ :param doc: document DOM :param parent: parent node :param name: tag of the element :param value: value of the child text node :param attr: attribute dictionary :return: True if something was appended, otherwise False """ if value is not None: parent = self.ebuilder(parent, name, value, attr) return True return False def _get_pi_string(self): """ Creates the processing instructions header for writing to file """ pis = self.return_processing_instructions() if len(pis) > 0: pi_tree = self.create_tree(pis[0]) i = 1 for i in range(1, len(pis) - 1): pi_tree = self.append(pis[i], pi_tree) pi_string = self.to_string(pi_tree) else: pi_string = "" return pi_string def _create_main_node(self): """ Creates the primary xml header used when writing to file """ xsi = "http://www.w3.org/2001/XMLSchema-instance" version = self.cansas_version n_s = CANSAS_NS.get(version).get("ns") if version == "1.1": url = "http://www.cansas.org/formats/1.1/" else: url = "http://svn.smallangles.net/svn/canSAS/1dwg/trunk/" schema_location = "{0} {1}cansas1d.xsd".format(n_s, url) attrib = {"{" + xsi + "}schemaLocation" : schema_location, "version" : version} nsmap = {'xsi' : xsi, None: n_s} main_node = self.create_element("{" + n_s + "}SASroot", attrib=attrib, nsmap=nsmap) return main_node def _write_run_names(self, datainfo, entry_node): """ Writes the run names to the XML file :param datainfo: The Data1D object the information is coming from :param entry_node: lxml node ElementTree object to be appended to """ if datainfo.run is None or datainfo.run == []: datainfo.run.append(RUN_NAME_DEFAULT) datainfo.run_name[RUN_NAME_DEFAULT] = RUN_NAME_DEFAULT for item in datainfo.run: runname = {} if item in datainfo.run_name and \ len(str(datainfo.run_name[item])) > 1: runname = {'name': datainfo.run_name[item]} self.write_node(entry_node, "Run", item, runname) def _write_data(self, datainfo, entry_node): """ Writes 1D I and Q data to the XML file :param datainfo: The Data1D object the information is coming from :param entry_node: lxml node ElementTree object to be appended to """ node = self.create_element("SASdata") self.append(node, entry_node) for i in range(len(datainfo.x)): point = self.create_element("Idata") node.append(point) self.write_node(point, "Q", datainfo.x[i], {'unit': datainfo.x_unit}) if len(datainfo.y) >= i: self.write_node(point, "I", datainfo.y[i], {'unit': datainfo.y_unit}) if datainfo.dy is not None and len(datainfo.dy) > i: self.write_node(point, "Idev", datainfo.dy[i], {'unit': datainfo.y_unit}) if datainfo.dx is not None and len(datainfo.dx) > i: self.write_node(point, "Qdev", datainfo.dx[i], {'unit': datainfo.x_unit}) if datainfo.dxw is not None and len(datainfo.dxw) > i: self.write_node(point, "dQw", datainfo.dxw[i], {'unit': datainfo.x_unit}) if datainfo.dxl is not None and len(datainfo.dxl) > i: self.write_node(point, "dQl", datainfo.dxl[i], {'unit': datainfo.x_unit}) if datainfo.isSesans: sesans_attrib = {'x_axis': datainfo._xaxis, 'y_axis': datainfo._yaxis, 'x_unit': datainfo.x_unit, 'y_unit': datainfo.y_unit} sesans = self.create_element("Sesans", attrib=sesans_attrib) sesans.text = str(datainfo.isSesans) entry_node.append(sesans) self.write_node(entry_node, "yacceptance", datainfo.sample.yacceptance[0], {'unit': datainfo.sample.yacceptance[1]}) self.write_node(entry_node, "zacceptance", datainfo.sample.zacceptance[0], {'unit': datainfo.sample.zacceptance[1]}) def _write_data_2d(self, datainfo, entry_node): """ Writes 2D data to the XML file :param datainfo: The Data2D object the information is coming from :param entry_node: lxml node ElementTree object to be appended to """ attr = {} if datainfo.data.shape: attr["x_bins"] = str(len(datainfo.x_bins)) attr["y_bins"] = str(len(datainfo.y_bins)) node = self.create_element("SASdata", attr) self.append(node, entry_node) point = self.create_element("Idata") node.append(point) qx = ','.join(str(v) for v in datainfo.qx_data) qy = ','.join(str(v) for v in datainfo.qy_data) intensity = ','.join(str(v) for v in datainfo.data) self.write_node(point, "Qx", qx, {'unit': datainfo._xunit}) self.write_node(point, "Qy", qy, {'unit': datainfo._yunit}) self.write_node(point, "I", intensity, {'unit': datainfo._zunit}) if datainfo.err_data is not None: err = ','.join(str(v) for v in datainfo.err_data) self.write_node(point, "Idev", err, {'unit': datainfo._zunit}) if datainfo.dqy_data is not None: dqy = ','.join(str(v) for v in datainfo.dqy_data) self.write_node(point, "Qydev", dqy, {'unit': datainfo._yunit}) if datainfo.dqx_data is not None: dqx = ','.join(str(v) for v in datainfo.dqx_data) self.write_node(point, "Qxdev", dqx, {'unit': datainfo._xunit}) if datainfo.mask is not None: mask = ','.join("1" if v else "0" for v in datainfo.mask) self.write_node(point, "Mask", mask) def _write_trans_spectrum(self, datainfo, entry_node): """ Writes the transmission spectrum data to the XML file :param datainfo: The Data1D object the information is coming from :param entry_node: lxml node ElementTree object to be appended to """ for i in range(len(datainfo.trans_spectrum)): spectrum = datainfo.trans_spectrum[i] node = self.create_element("SAStransmission_spectrum", {"name" : spectrum.name}) self.append(node, entry_node) if isinstance(spectrum.timestamp, datetime.datetime): node.setAttribute("timestamp", spectrum.timestamp) for i in range(len(spectrum.wavelength)): point = self.create_element("Tdata") node.append(point) self.write_node(point, "Lambda", spectrum.wavelength[i], {'unit': spectrum.wavelength_unit}) self.write_node(point, "T", spectrum.transmission[i], {'unit': spectrum.transmission_unit}) if spectrum.transmission_deviation is not None \ and len(spectrum.transmission_deviation) >= i: self.write_node(point, "Tdev", spectrum.transmission_deviation[i], {'unit': spectrum.transmission_deviation_unit}) def _write_sample_info(self, datainfo, entry_node): """ Writes the sample information to the XML file :param datainfo: The Data1D object the information is coming from :param entry_node: lxml node ElementTree object to be appended to """ sample = self.create_element("SASsample") if datainfo.sample.name is not None: self.write_attribute(sample, "name", str(datainfo.sample.name)) self.append(sample, entry_node) self.write_node(sample, "ID", str(datainfo.sample.ID)) self.write_node(sample, "thickness", datainfo.sample.thickness, {"unit": datainfo.sample.thickness_unit}) self.write_node(sample, "transmission", datainfo.sample.transmission) self.write_node(sample, "temperature", datainfo.sample.temperature, {"unit": datainfo.sample.temperature_unit}) pos = self.create_element("position") written = self.write_node(pos, "x", datainfo.sample.position.x, {"unit": datainfo.sample.position_unit}) written = written | self.write_node( \ pos, "y", datainfo.sample.position.y, {"unit": datainfo.sample.position_unit}) written = written | self.write_node( \ pos, "z", datainfo.sample.position.z, {"unit": datainfo.sample.position_unit}) if written == True: self.append(pos, sample) ori = self.create_element("orientation") written = self.write_node(ori, "roll", datainfo.sample.orientation.x, {"unit": datainfo.sample.orientation_unit}) written = written | self.write_node( \ ori, "pitch", datainfo.sample.orientation.y, {"unit": datainfo.sample.orientation_unit}) written = written | self.write_node( \ ori, "yaw", datainfo.sample.orientation.z, {"unit": datainfo.sample.orientation_unit}) if written == True: self.append(ori, sample) for item in datainfo.sample.details: self.write_node(sample, "details", item) def _write_instrument(self, datainfo, entry_node): """ Writes the instrumental information to the XML file :param datainfo: The Data1D object the information is coming from :param entry_node: lxml node ElementTree object to be appended to """ instr = self.create_element("SASinstrument") self.append(instr, entry_node) self.write_node(instr, "name", datainfo.instrument) return instr def _write_source(self, datainfo, instr): """ Writes the source information to the XML file :param datainfo: The Data1D object the information is coming from :param instr: instrument node to be appended to """ source = self.create_element("SASsource") if datainfo.source.name is not None: self.write_attribute(source, "name", str(datainfo.source.name)) self.append(source, instr) if datainfo.source.radiation is None or datainfo.source.radiation == '': datainfo.source.radiation = "neutron" self.write_node(source, "radiation", datainfo.source.radiation) size = self.create_element("beam_size") if datainfo.source.beam_size_name is not None: self.write_attribute(size, "name", str(datainfo.source.beam_size_name)) written = self.write_node( \ size, "x", datainfo.source.beam_size.x, {"unit": datainfo.source.beam_size_unit}) written = written | self.write_node( \ size, "y", datainfo.source.beam_size.y, {"unit": datainfo.source.beam_size_unit}) written = written | self.write_node( \ size, "z", datainfo.source.beam_size.z, {"unit": datainfo.source.beam_size_unit}) if written == True: self.append(size, source) self.write_node(source, "beam_shape", datainfo.source.beam_shape) self.write_node(source, "wavelength", datainfo.source.wavelength, {"unit": datainfo.source.wavelength_unit}) self.write_node(source, "wavelength_min", datainfo.source.wavelength_min, {"unit": datainfo.source.wavelength_min_unit}) self.write_node(source, "wavelength_max", datainfo.source.wavelength_max, {"unit": datainfo.source.wavelength_max_unit}) self.write_node(source, "wavelength_spread", datainfo.source.wavelength_spread, {"unit": datainfo.source.wavelength_spread_unit}) def _write_collimation(self, datainfo, instr): """ Writes the collimation information to the XML file :param datainfo: The Data1D object the information is coming from :param instr: lxml node ElementTree object to be appended to """ if datainfo.collimation == [] or datainfo.collimation is None: coll = Collimation() datainfo.collimation.append(coll) for item in datainfo.collimation: coll = self.create_element("SAScollimation") if item.name is not None: self.write_attribute(coll, "name", str(item.name)) self.append(coll, instr) self.write_node(coll, "length", item.length, {"unit": item.length_unit}) for aperture in item.aperture: apert = self.create_element("aperture") if aperture.name is not None: self.write_attribute(apert, "name", str(aperture.name)) if aperture.type is not None: self.write_attribute(apert, "type", str(aperture.type)) self.append(apert, coll) size = self.create_element("size") if aperture.size_name is not None: self.write_attribute(size, "name", str(aperture.size_name)) written = self.write_node(size, "x", aperture.size.x, {"unit": aperture.size_unit}) written = written | self.write_node( \ size, "y", aperture.size.y, {"unit": aperture.size_unit}) written = written | self.write_node( \ size, "z", aperture.size.z, {"unit": aperture.size_unit}) if written == True: self.append(size, apert) self.write_node(apert, "distance", aperture.distance, {"unit": aperture.distance_unit}) def _write_detectors(self, datainfo, instr): """ Writes the detector information to the XML file :param datainfo: The Data1D object the information is coming from :param inst: lxml instrument node to be appended to """ if datainfo.detector is None or datainfo.detector == []: det = Detector() det.name = "" datainfo.detector.append(det) for item in datainfo.detector: det = self.create_element("SASdetector") written = self.write_node(det, "name", item.name) written = written | self.write_node(det, "SDD", item.distance, {"unit": item.distance_unit}) if written == True: self.append(det, instr) off = self.create_element("offset") written = self.write_node(off, "x", item.offset.x, {"unit": item.offset_unit}) written = written | self.write_node(off, "y", item.offset.y, {"unit": item.offset_unit}) written = written | self.write_node(off, "z", item.offset.z, {"unit": item.offset_unit}) if written == True: self.append(off, det) ori = self.create_element("orientation") written = self.write_node(ori, "roll", item.orientation.x, {"unit": item.orientation_unit}) written = written | self.write_node(ori, "pitch", item.orientation.y, {"unit": item.orientation_unit}) written = written | self.write_node(ori, "yaw", item.orientation.z, {"unit": item.orientation_unit}) if written == True: self.append(ori, det) center = self.create_element("beam_center") written = self.write_node(center, "x", item.beam_center.x, {"unit": item.beam_center_unit}) written = written | self.write_node(center, "y", item.beam_center.y, {"unit": item.beam_center_unit}) written = written | self.write_node(center, "z", item.beam_center.z, {"unit": item.beam_center_unit}) if written == True: self.append(center, det) pix = self.create_element("pixel_size") written = self.write_node(pix, "x", item.pixel_size.x, {"unit": item.pixel_size_unit}) written = written | self.write_node(pix, "y", item.pixel_size.y, {"unit": item.pixel_size_unit}) written = written | self.write_node(pix, "z", item.pixel_size.z, {"unit": item.pixel_size_unit}) if written == True: self.append(pix, det) self.write_node(det, "slit_length", item.slit_length, {"unit": item.slit_length_unit}) def _write_process_notes(self, datainfo, entry_node): """ Writes the process notes to the XML file :param datainfo: The Data1D object the information is coming from :param entry_node: lxml node ElementTree object to be appended to """ for item in datainfo.process: node = self.create_element("SASprocess") self.append(node, entry_node) self.write_node(node, "name", item.name) self.write_node(node, "date", item.date) self.write_node(node, "description", item.description) for term in item.term: if isinstance(term, list): value = term['value'] del term['value'] elif isinstance(term, dict): value = term.get("value") del term['value'] else: value = term self.write_node(node, "term", value, term) for note in item.notes: self.write_node(node, "SASprocessnote", note) if len(item.notes) == 0: self.write_node(node, "SASprocessnote", "") def _write_notes(self, datainfo, entry_node): """ Writes the notes to the XML file and creates an empty note if none exist :param datainfo: The Data1D object the information is coming from :param entry_node: lxml node ElementTree object to be appended to """ if len(datainfo.notes) == 0: node = self.create_element("SASnote") self.append(node, entry_node) else: for item in datainfo.notes: node = self.create_element("SASnote") self.write_text(node, item) self.append(node, entry_node) def _check_origin(self, entry_node, doc): """ Return the document, and the SASentry node associated with the data we just wrote. If the calling function was not the cansas reader, return a minidom object rather than an lxml object. :param entry_node: lxml node ElementTree object to be appended to :param doc: entire xml tree """ if not self.frm: self.frm = inspect.stack()[1] mod_name = self.frm[1].replace("\\", "/").replace(".pyc", "") mod_name = mod_name.replace(".py", "") mod = mod_name.split("sas/") mod_name = mod[1] if mod_name != "sascalc/dataloader/readers/cansas_reader": string = self.to_string(doc, pretty_print=False) doc = parseString(string) node_name = entry_node.tag node_list = doc.getElementsByTagName(node_name) entry_node = node_list.item(0) return doc, entry_node # DO NOT REMOVE - used in saving and loading panel states. def _store_float(self, location, node, variable, storage, optional=True): """ Get the content of a xpath location and store the result. Check that the units are compatible with the destination. The value is expected to be a float. The xpath location might or might not exist. If it does not exist, nothing is done :param location: xpath location to fetch :param node: node to read the data from :param variable: name of the data member to store it in [string] :param storage: data object that has the 'variable' data member :param optional: if True, no exception will be raised if unit conversion can't be done :raise ValueError: raised when the units are not recognized """ entry = get_content(location, node) try: value = float(entry.text) except ValueError: value = None if value is not None: # If the entry has units, check to see that they are # compatible with what we currently have in the data object units = entry.get('unit') if units is not None: toks = variable.split('.') local_unit = getattr(storage, toks[0]+"_unit") if local_unit is not None and units.lower() != local_unit.lower(): try: conv = Converter(units) setattrchain(storage, variable, conv(value, units=local_unit)) except Exception: _, exc_value, _ = sys.exc_info() err_mess = "CanSAS reader: could not convert" err_mess += " %s unit [%s]; expecting [%s]\n %s" \ % (variable, units, local_unit, exc_value) self.errors.add(err_mess) if optional: logger.info(err_mess) else: raise ValueError(err_mess) else: setattrchain(storage, variable, value) else: setattrchain(storage, variable, value) # DO NOT REMOVE - used in saving and loading panel states. def _store_content(self, location, node, variable, storage): """ Get the content of a xpath location and store the result. The value is treated as a string. The xpath location might or might not exist. If it does not exist, nothing is done :param location: xpath location to fetch :param node: node to read the data from :param variable: name of the data member to store it in [string] :param storage: data object that has the 'variable' data member :return: return a list of errors """ entry = get_content(location, node) if entry is not None and entry.text is not None: setattrchain(storage, variable, entry.text.strip()) # DO NOT REMOVE Called by outside packages: # sas.sasgui.perspectives.invariant.invariant_state # sas.sasgui.perspectives.fitting.pagestate def get_content(location, node): """ Get the first instance of the content of a xpath location. :param location: xpath location :param node: node to start at :return: Element, or None """ nodes = node.xpath(location, namespaces={'ns': CANSAS_NS.get("1.0").get("ns")}) if len(nodes) > 0: return nodes[0] else: return None # DO NOT REMOVE Called by outside packages: # sas.sasgui.perspectives.fitting.pagestate def write_node(doc, parent, name, value, attr=None): """ :param doc: document DOM :param parent: parent node :param name: tag of the element :param value: value of the child text node :param attr: attribute dictionary :return: True if something was appended, otherwise False """ if attr is None: attr = {} if value is not None: node = doc.createElement(name) node.appendChild(doc.createTextNode(str(value))) for item in attr: node.setAttribute(item, attr[item]) parent.appendChild(node) return True return False def getattrchain(obj, chain, default=None): """Like getattr, but the attr may contain multiple parts separated by '.'""" for part in chain.split('.'): if hasattr(obj, part): obj = getattr(obj, part, None) else: return default return obj def setattrchain(obj, chain, value): """Like setattr, but the attr may contain multiple parts separated by '.'""" parts = list(chain.split('.')) for part in parts[-1]: obj = getattr(obj, part, None) if obj is None: raise ValueError("missing parent object "+part) setattr(obj, value)