Esempio n. 1
0
    def getSRS(self, srsname, typename):
        """Returns None or Crs object for given name

        @param typename:  feature name
        @type typename: String
        """
        if not isinstance(srsname, Crs):
            srs = Crs(srsname)
        else:
            srs = srsname

        try:
            index = self.contents[typename].crsOptions.index(srs)
            # Return the Crs string that was pulled directly from the
            # GetCaps document (the 'id' attribute in the Crs object).
            return self.contents[typename].crsOptions[index]
        except ValueError:
            options = ", ".join(
                [crs.id for crs in self.contents[typename].crsOptions])
            log.warning(
                "Requested srsName %r is not declared as being "
                "allowed for requested typename %r. "
                "Options are: %r.",
                srs.getcode(),
                typename,
                options,
            )
            return None
Esempio n. 2
0
def axml_display(d):
    # <owc:display>
    # 		<owc:pixelWidth>
    if is_empty(d):
        return None
    else:
        try:
            creator_display = etree.Element(ns_elem("owc", "display"),
                                            nsmap=ns)
            pixel_width = try_int(extract_p('pixelWidth', d, None))
            if pixel_width is not None:
                etree.SubElement(creator_display,
                                 ns_elem("owc",
                                         "pixelWidth")).text = str(pixel_width)
            pixel_height = try_int(extract_p('pixelHeight', d, None))
            if pixel_height is not None:
                etree.SubElement(creator_display, ns_elem(
                    "owc", "pixelHeight")).text = str(pixel_height)
            mm_per_pixel = try_float(extract_p('mmPerPixel', d, None))
            if mm_per_pixel is not None:
                etree.SubElement(creator_display, ns_elem(
                    "owc", "mmPerPixel")).text = str(mm_per_pixel)
            return creator_display
        except Exception as ex:
            log.warning('could encode creator_display', ex)
            return None
Esempio n. 3
0
def axml_link(d):
    # < link rel = "via" type = "application/xml" href = "..." title = "..."
    if is_empty(d):
        return None
    else:
        try:
            link = etree.Element("link", nsmap=ns)
            href = extract_p('href', d, None)
            if href is not None:
                link.set("href", href)
            rel = extract_p('rel', d, None)
            if rel is not None:
                link.set("rel", rel)
            mimetype = extract_p('type', d, None)
            if mimetype is not None:
                link.set("type", mimetype)
            lang = extract_p('lang', d, None)
            if lang is not None:
                link.set("lang", lang)
            title = extract_p('title', d, None)
            if title is not None:
                link.set("title", title)
            length = try_int(extract_p('length', d, None))
            if length is not None:
                link.set("length", str(length))
            return link
        except Exception as ex:
            log.warning('could not encode link', ex)
            return None
Esempio n. 4
0
def axml_offering(d):
    # <owc:offering code="http://www.opengis.net/spec/owc-at...">
    #   <owc:offering code="http://www.opengis.net/spec....l">
    # 			<owc:content type="application/gml+xml">
    if is_empty(d):
        return None
    else:
        try:
            offering_code = extract_p('code', d, None)
            offering = etree.Element(ns_elem("owc", "offering"),
                                     attrib={"code": offering_code},
                                     nsmap=ns)

            # use axml_operation here
            operations = [
                axml_operation(do) for do in extract_p('operations', d, [])
            ]
            [offering.append(el) for el in operations if el is not None]
            # use axml_content here
            contents = [
                axml_content(do) for do in extract_p('contents', d, [])
            ]
            [offering.append(el) for el in contents if el is not None]
            # use axml_styeset here
            styles = [axml_styleset(do) for do in extract_p('styles', d, [])]
            [offering.append(el) for el in styles if el is not None]
            return offering
        except Exception as ex:
            log.warning('could encode offering', ex)
            return None
Esempio n. 5
0
def axml_content(d):
    """
    OwcContent dict to Atom XML
    :param d:
    :return:
    """
    #   <owc:content type="image/tiff" href=".."
    if is_empty(d):
        return None
    else:
        try:
            content_elem = etree.Element(ns_elem("owc", "content"), nsmap=ns)

            mimetype = extract_p('type', d, None)
            if mimetype is not None:
                content_elem.set("type", mimetype)
            url = extract_p('url', d, None)
            if url is not None:
                content_elem.set("href", url)
            title = extract_p('title', d, None)
            if title is not None:
                content_elem.set("title", title)

            content = extract_p('content', d, None)
            if content is None:
                content_elem.text = content
            return content_elem
        except Exception as ex:
            log.warning('could encode content', ex)
            return None
Esempio n. 6
0
def _executeRequest(processid,
                    inputs,
                    output,
                    verbose=True,
                    outputFilePath=None,
                    sleepSecs=10):
    """
    This function makes a call to the Web Processing Service with
    the specified user inputs.
    """
    wps = WebProcessingService(WPS_URL)

    execution = wps.execute(processid, inputs, output)

    sleepSecs = sleepSecs
    err_count = 1

    while execution.isComplete() == False:
        try:
            monitorExecution(execution, sleepSecs,
                             download=False)  # monitors for success
            err_count = 1
        except Exception:
            log.warning(
                'An error occurred while checking status, checking again. Sleeping %d seconds...'
                % sleepSecs)
            err_count += 1
            if err_count > WPS_attempts:
                raise Exception(
                    'The status document failed to return, status checking has aborted. There has been a network or server issue preventing the status document from being retrieved, the request may still be running. For more information, check the status url %s'
                    % execution.statusLocation)
            sleep(sleepSecs)

    if outputFilePath == None:
        outputFilePath = 'gdp_' + processid.replace(
            '.', '-') + '_' + strftime("%Y-%m-%dT%H-%M-%S-%Z")

    done = False
    err_count = 1
    while done == False:
        try:
            monitorExecution(execution, download=True, filepath=outputFilePath)
            done = True
        except Exception:
            log.warning(
                'An error occurred while trying to download the result file, trying again.'
            )
            err_count += 1
            sleep(sleepSecs)
            if err_count > WPS_attempts:
                raise Exception(
                    "The process completed successfully, but an error occurred while downloading the result. You may be able to download the file using the link at the bottom of the status document: %s"
                    % execution.statusLocation)

    _check_for_execution_errors(execution)

    return outputFilePath
Esempio n. 7
0
def _executeRequest(processid,
                    inputs,
                    output,
                    verbose=True,
                    outputFilePath=None,
                    sleepSecs=10):
    """
    This function makes a call to the Web Processing Service with
    the specified user inputs.
    """
    wps = WebProcessingService(WPS_URL)

    execution = wps.execute(processid, inputs, output)

    sleepSecs = sleepSecs
    err_count = 1

    while execution.isComplete() == False:
        try:
            monitorExecution(
                execution, sleepSecs, download=False)  # monitors for success
            err_count = 1
        except Exception:
            log.warning(
                'An error occurred while checking status, checking again. Sleeping %d seconds...'
                % sleepSecs)
            err_count += 1
            if err_count > WPS_attempts:
                raise Exception(
                    'The status document failed to return, status checking has aborted. There has been a network or server issue preventing the status document from being retrieved, the request may still be running. For more information, check the status url %s'
                    % execution.statusLocation)
            sleep(sleepSecs)

    if outputFilePath == None:
        outputFilePath = 'gdp_' + processid.replace(
            '.', '-') + '_' + strftime("%Y-%m-%dT%H-%M-%S-%Z")

    done = False
    err_count = 1
    while done == False:
        try:
            monitorExecution(execution, download=True, filepath=outputFilePath)
            done = True
        except Exception:
            log.warning(
                'An error occurred while trying to download the result file, trying again.'
            )
            err_count += 1
            sleep(sleepSecs)
            if err_count > WPS_attempts:
                raise Exception(
                    "The process completed successfully, but an error occurred while downloading the result. You may be able to download the file using the link at the bottom of the status document: %s"
                    % execution.statusLocation)

    _check_for_execution_errors(execution)

    return outputFilePath
Esempio n. 8
0
def axml_styleset(d):
    #     <owc:styleSet>
    #       <owc:name>raster</owc:name>
    #       <owc:title>Default Raster</owc:title>
    #       <owc:abstract>A sample style that draws a </owc:abstract>
    #       <owc:legendURL href="h...." type="image/png"/>
    #     </owc:styleSet>
    if is_empty(d):
        return None
    else:
        try:
            styleset = etree.Element(ns_elem("owc", "styleSet"), nsmap=ns)

            name = extract_p('name', d, None)
            if name is not None:
                etree.SubElement(styleset, ns_elem("owc", "name")).text = name
            title = extract_p('title', d, None)
            if title is not None:
                etree.SubElement(styleset, ns_elem("owc",
                                                   "title")).text = title
            subtitle = extract_p('abstract', d, None)
            if subtitle is not None:
                etree.SubElement(styleset, ns_elem("owc",
                                                   "abstract")).text = subtitle
            is_default = extract_p('default', d, None)
            # TODO no example for default setting on style set
            if is_default is not None:
                etree.SubElement(styleset,
                                 ns_elem("owc", "default")).text = is_default
            legend_url = extract_p('legendURL', d, None)
            if legend_url is not None:
                etree.SubElement(styleset,
                                 ns_elem("owc", "legendURL")).text = legend_url
            # TODO no example for content on style set
            content = extract_p('content', d, None)
            content_enc = None if content is None else axml_content(content)
            if content_enc is not None:
                styleset.append(content_enc)
            return styleset
        except Exception as ex:
            log.warning('could encode styleset', ex)
            return None
Esempio n. 9
0
def axml_creator_app(d):
    # <generator uri="http://w.." version="1.0">MiraMon</generator>
    if is_empty(d):
        return None
    else:
        try:
            creator_app = etree.Element("generator", nsmap=ns)
            title = extract_p('title', d, None)
            if title is not None:
                creator_app.text = title
            uri = extract_p('uri', d, None)
            if uri is not None:
                creator_app.set("uri", uri)
            version = extract_p('version', d, None)
            if version is not None:
                creator_app.set("version", version)
            return creator_app
        except Exception as ex:
            log.warning('could encode creator_app', ex)
            return None
Esempio n. 10
0
def axml_category(d):
    # <category term="maps" label="This file contains maps"/>
    if is_empty(d):
        return None
    else:
        try:
            category = etree.Element("category", nsmap=ns)
            term = extract_p('term', d, None)
            if term is not None:
                category.set("term", term)
            scheme = extract_p('scheme', d, None)
            if scheme is not None:
                category.set("scheme", scheme)
            label = extract_p('label', d, None)
            if label is not None:
                category.set("label", label)
            return category
        except Exception as ex:
            log.warning('could encode category', ex)
            return None
Esempio n. 11
0
    def getSRS(self, srsname, typename):
        """Returns None or Crs object for given name

        @param typename:  feature name 
        @type typename: String
        """
        if not isinstance(srsname, Crs):
            srs = Crs(srsname)
        else:
            srs = srsname

        try:
            index = self.contents[typename].crsOptions.index(srs)
            # Return the Crs string that was pulled directly from the
            # GetCaps document (the 'id' attribute in the Crs object).
            return self.contents[typename].crsOptions[index]
        except ValueError:
            options = ", ".join(map(lambda x: x.id, self.contents[typename].crsOptions))
            log.warning("Requested srsName '%s' not available for requested typename '%s'. \
                         Options are: %s. " % (srs.getcode(), typename, options))
            return None
Esempio n. 12
0
    def getSRS(self, srsname, typename):
        """Returns None or Crs object for given name

        @param typename:  feature name 
        @type typename: String
        """
        if not isinstance(srsname, Crs):
            srs = Crs(srsname)
        else:
            srs = srsname

        try:
            index = self.contents[typename].crsOptions.index(srs)
            # Return the Crs string that was pulled directly from the
            # GetCaps document (the 'id' attribute in the Crs object).
            return self.contents[typename].crsOptions[index]
        except ValueError:
            options = ", ".join(map(lambda x: x.id, self.contents[typename].crsOptions))
            log.warning("Requested srsName '%s' not available for requested typename '%s'. \
                         Options are: %s. " % (srs.getcode(), typename, options))
            return None
Esempio n. 13
0
def axml_author(d):
    # <author> ..
    # 		<name>
    #       <email>
    if is_empty(d):
        return None
    else:
        try:
            author = etree.Element("author", nsmap=ns)
            name = extract_p('name', d, None)
            if name is not None:
                etree.SubElement(author, "name").text = name
            email = extract_p('email', d, None)
            if email is not None:
                etree.SubElement(author, "email").text = email
            uri = extract_p('uri', d, None)
            if uri is not None:
                etree.SubElement(author, "uri").text = uri
            return author
        except Exception as ex:
            log.warning('could encode author', ex)
            return None
Esempio n. 14
0
def axml_operation(d):
    #   <owc:operation code="GetCapabilities" method="GET"
    #       type="applica..." href="..."
    #       <owc:request type="application/xml"> ..
    # etree.SubElement(entry, ns_elem("owc", "offering"), name="blah").text = "some value1"
    if is_empty(d):
        return None
    else:
        try:
            operation = etree.Element(ns_elem("owc", "operation"), nsmap=ns)

            operations_code = extract_p('code', d, None)
            if operations_code is not None:
                operation.set("code", operations_code)
            http_method = extract_p('method', d, None)
            if http_method is not None:
                operation.set("method", http_method)
            mimetype = extract_p('type', d, None)
            if mimetype is not None:
                operation.set("type", mimetype)
            request_url = extract_p('href', d, None)
            if request_url is not None:
                operation.set("href", request_url)

            # use axml_content here
            request = extract_p('request', d, None)
            request_enc = None if request is None else axml_content(request)
            if request_enc is not None:
                operation.append(request_enc)
            # use axml_content here
            result = extract_p('result', d, None)
            result_enc = None if result is None else axml_content(result)
            if result_enc is not None:
                operation.append(result_enc)
            return operation
        except Exception as ex:
            log.warning('could encode operation', ex)
            return None
Esempio n. 15
0
    def getPOSTGetFeatureRequest(
        self,
        typename=None,
        filter=None,
        bbox=None,
        featureid=None,
        featureversion=None,
        propertyname=None,
        maxfeatures=None,
        storedQueryID=None,
        storedQueryParams=None,
        outputFormat=None,
        method="Post",
        startindex=None,
        sortby=None,
    ):
        """Formulate proper GetFeature request using KVP encoding
        ----------
        typename : list
            List of typenames (string)
        filter : string
            XML-encoded OGC filter expression.
        bbox : tuple
            (left, bottom, right, top) in the feature type's coordinates == (minx, miny, maxx, maxy)
        featureid : list
            List of unique feature ids (string)
        featureversion : string
            Default is most recent feature version.
        propertyname : list
            List of feature property names. Leave blank (None) to get all properties.
        maxfeatures : int
            Maximum number of features to be returned.
        method : string
            Qualified name of the HTTP DCP method to use.
        outputFormat: string (optional)
            Requested response format of the request.
        startindex: int (optional)
            Start position to return feature set (paging in combination with maxfeatures)
        storedQueryID : string
            A name identifying a prepared set available in WFS-service.
            WFS version 2.0.0 and above only.
        storedQueryParams : dict
            Variable amount of extra information sent to server related to
            storedQueryID to further define the requested data.
            WFS version 2.0.0 and above only.
            {'parameter_name': parameter_value}
        sortby: list (optional)
            List of property names whose values should be used to order
            (upon presentation) the set of feature instances that
            satify the query.

        There are 5 different modes of use

        1) typename and bbox (simple spatial query)
        2) typename and filter (==query) (more expressive)
        3) featureid (direct access to known features)
        4) storedQueryID and optional storedQueryParams
        5) filter only via Post method
        """

        try:
            base_url = next(
                (
                    m.get("url")
                    for m in self.getOperationByName("GetFeature").methods
                    if m.get("type").lower() == method.lower()
                )
            )
        except StopIteration:
            base_url = self.url

        if not typename and filter:
            return base_url, filter

        request = self.create_post_request()

        if storedQueryID:
            if self.version in ["1.0.0", "1.1.0"]:
                log.warning("Stored queries are only supported in version 2.0.0 and above.")
                return None

            storedQueryParams = storedQueryParams or {}
            request.create_storedquery(storedQueryID, storedQueryParams)
            data = request.to_string()
            return base_url, data

        typename = (
            [typename] if isinstance(typename, str) else typename
        )  # noqa: E721
        typenames = ",".join(typename)

        request.create_query(typenames)

        if featureid:
            featureid = (
                [featureid] if isinstance(featureid, str) else featureid
            )
            request.set_featureid(featureid)
        elif bbox:
            request.set_bbox(self.getBBOXPost(bbox, typename))
        elif filter:
            request.set_filter(filter)

        if featureversion:
            request.set_featureversion(str(featureversion))
        if maxfeatures:
            request.set_maxfeatures(maxfeatures)
        if outputFormat:
            request.set_outputformat(outputFormat)
        if propertyname:
            propertyname = (
                [propertyname] if isinstance(propertyname, str) else propertyname
            )
            request.set_propertyname(propertyname)
        if sortby:
            sortby = (
                [sortby] if isinstance(sortby, str) else sortby
            )
            request.set_sortby(sortby)
        if startindex:
            request.set_startindex(startindex)

        data = request.to_string()
        return base_url, data
Esempio n. 16
0
def decode_atomxml(xml_string):
    """
    here parse atom xml to a dict for instanciating of OWC:Context
    :param xmlstring:
    :return: OwcContext-ready dict
    """
    context_base_dict = {
        "type": "FeatureCollection",
        "id": None,
        "bbox": None,
        "properties": {
            "lang": None,
            "links": {
                "profiles": [],
                "via": [],
            },
            'title': None,
            'abstract': None,
            'updated': None,
            'authors': [],
            'publisher': None,
            'generator': None,
            'display': None,
            'rights': None,
            'date': None,
            'categories': [],
        },
        'features': []
    }
    feed_root = etree.fromstring(xml_string)
    # feed_root = etree.parse(xml_bytes)
    # feed_root xml lang use?
    # # log.debug(feed_root)
    # feed xml:lang=en
    # lang = feed_root.get('{http://www.w3.org/XML/1998/namespace}lang')
    lang = util.testXMLAttribute(feed_root,
                                 '{http://www.w3.org/XML/1998/namespace}lang')
    # log.debug("lang %s ", lang)
    context_base_dict['properties'].update({"lang": lang})

    # <id>
    val = feed_root.find(util.nspath_eval('atom:id', ns))
    id = util.testXMLValue(val)
    # log.debug("id %s :: %s", id, val)
    context_base_dict.update({"id": id})

    # <link rel="profile"
    #   href="http://www.opengis.net/spec/owc-atom/1.0/req/core"
    #   title="compliant bla bla"
    # < link rel = "via" type = "application/xml" href = "..." title = "..."
    vals = feed_root.findall(util.nspath_eval('atom:link', ns))
    links_profile = []
    links_via = []
    for val in vals:
        rel = util.testXMLAttribute(val, 'rel')
        href = util.testXMLAttribute(val, 'href')
        mimetype = util.testXMLAttribute(val, 'type')
        lang = util.testXMLAttribute(val, 'lang')
        title = util.testXMLAttribute(val, 'title')
        length = util.testXMLAttribute(val, 'length')
        link = {
            "href": href,
            "type": mimetype,
            "length": length,
            "lang": lang,
            "title": title,
            "rel": rel
        }
        # log.debug("link %s :: %s", link, vals)
        if link.get("rel") == "profile" and not is_empty(link):
            links_profile.append(link)
        elif link.get("rel") == "via" and not is_empty(link):
            links_via.append(link)
        else:
            log.warning("unknown link type in Ows Context section: %r", link)

    context_base_dict['properties']['links'].update(
        {"profiles": links_profile})
    context_base_dict['properties']['links'].update({"via": links_via})

    # <title>
    val = feed_root.find(util.nspath_eval('atom:title', ns))
    title = util.testXMLValue(val)
    # log.debug("title %s :: %s", title, val)
    context_base_dict['properties'].update({"title": title})

    # <subtitle type = "html"
    val = feed_root.find(util.nspath_eval('atom:subtitle', ns))
    subtitle = util.testXMLValue(val)
    # log.debug("subtitle %s :: %s", subtitle, val)
    context_base_dict['properties'].update({"abstract": subtitle})

    # <author> ..
    # 		<name>
    #       <email>
    vals = feed_root.findall(util.nspath_eval('atom:author', ns))
    authors = []
    for val in vals:
        val_name = val.find(util.nspath_eval('atom:name', ns))
        val_email = val.find(util.nspath_eval('atom:email', ns))
        val_uri = val.find(util.nspath_eval('atom:uri', ns))
        name = util.testXMLValue(val_name)
        email = util.testXMLValue(val_email)
        uri = util.testXMLValue(val_uri)
        author = {"name": name, "email": email, "uri": uri}
        # log.debug("author %s :: %s", author, vals)
        if not is_empty(author):
            authors.append(author)

    context_base_dict['properties'].update({"authors": authors})

    # <georss:where>
    val = feed_root.find(util.nspath_eval('georss:where', ns))
    if val is not None:
        if len(list(val)) > 0:
            xmltxt = element_to_string(list(val)[0], False)
            # log.debug("geometry %s :: %s", xmltxt, val)
            context_base_dict['properties'].update(
                {"bbox": xmltxt.decode('utf-8')})

    # <updated>2012-11-04T17:26:23Z</updated>
    val = feed_root.find(util.nspath_eval('atom:updated', ns))
    update_date = util.testXMLValue(val)
    # log.debug("updated %s :: %s", update_date, val)
    context_base_dict['properties'].update({"updated": update_date})

    # <dc:date>2009-01-23T09:08:56.000Z/2009-01-23T09:14:08.000Z</dc:date>
    val = feed_root.find(util.nspath_eval('dc:date', ns))
    time_interval_of_interest = util.testXMLValue(val)
    # log.debug("dc:date %s :: %s", time_interval_of_interest, val)
    context_base_dict['properties'].update({"date": time_interval_of_interest})

    # <rights>
    val = feed_root.find(util.nspath_eval('atom:rights', ns))
    rights = util.testXMLValue(val)
    # log.debug("rights %s :: %s", rights, val)
    context_base_dict['properties'].update({"rights": rights})

    # <dc:publisher>
    val = feed_root.find(util.nspath_eval('dc:publisher', ns))
    publisher = util.testXMLValue(val)
    # log.debug("dc:publisher %s :: %s", publisher, val)
    context_base_dict['properties'].update({"publisher": publisher})

    # <owc:display>
    # 		<owc:pixelWidth>
    val_display = feed_root.find(util.nspath_eval('owc:display', ns))
    val_pixel_width = None if val_display is None \
        else val_display.find(util.nspath_eval('owc:pixelWidth', ns))
    val_pixel_height = None if val_display is None \
        else val_display.find(util.nspath_eval('owc:pixelHeight', ns))
    val_mm_per_pixel = None if val_display is None \
        else val_display.find(util.nspath_eval('owc:mmPerPixel', ns))
    pixel_width = util.testXMLValue(val_pixel_width)
    pixel_height = util.testXMLValue(val_pixel_height)
    mm_per_pixel = util.testXMLValue(val_mm_per_pixel)
    owc_display = {
        "pixelWidth": pixel_width,
        "pixelHeight": pixel_height,
        "mmPerPixel": mm_per_pixel
    }
    # log.debug("display %s :: %s", owc_display, val_display)
    if not is_empty(owc_display):
        context_base_dict['properties'].update({"display": owc_display})

    # <generator uri="http://w.." version="1.0">MiraMon</generator>
    val = feed_root.find(util.nspath_eval('atom:generator', ns))
    name = util.testXMLValue(val)
    version = util.testXMLAttribute(val, 'version')
    uri = util.testXMLAttribute(val, 'uri')
    owc_generator = {"name": name, "version": version, "uri": uri}
    # log.debug("generator %s :: %s", owc_generator, val)
    if not is_empty(owc_generator):
        context_base_dict['properties'].update({"generator": owc_generator})

    # <category term="maps" label="This file contains maps"/>
    vals = feed_root.findall(util.nspath_eval('atom:category', ns))
    categories = []
    for val in vals:
        term = util.testXMLAttribute(val, 'term')
        scheme = util.testXMLAttribute(val, 'scheme')
        label = util.testXMLAttribute(val, 'label')
        category = {"term": term, "scheme": scheme, "label": label}
        # log.debug("category %s :: %s", category, vals)
        if not is_empty(category):
            categories.append(category)

    context_base_dict['properties'].update({"categories": categories})

    # <entry> ...
    entries = feed_root.findall(util.nspath_eval('atom:entry', ns))
    resources = []
    for entry in entries:
        entry_dict = parse_entry(entry)

        if entry_dict.get("id") is not None:
            resources.append(entry_dict)
        else:
            log.warning("feature entry has no id, not allowed: skipping!")

    context_base_dict.update({"features": resources})
    return context_base_dict
Esempio n. 17
0
def axml_context(d):
    """
    encodes base OwcContext as dict to atom xml tree
    :param d:
    :return:
    """
    xml = etree.Element("feed", nsmap=ns)
    etree.SubElement(xml, "id").text = d['id']

    spec_reference = [
        axml_link(do) for do in extract_p('properties.links.profiles', d, [])
    ]
    [xml.append(el) for el in spec_reference if el is not None]

    area_of_interest = extract_p('bbox', d, None)
    if area_of_interest is not None:
        try:
            gml = etree.fromstring(area_of_interest)
            georss = etree.SubElement(xml, ns_elem("georss", "where"))
            georss.append(gml)
        except Exception as ex:
            log.warning('could encode bbox into georss:where', ex)
            pass

    context_metadata = [
        axml_link(do) for do in extract_p('properties.links.via', d, [])
    ]
    [xml.append(el) for el in context_metadata if el is not None]

    language = extract_p('properties.lang', d, None)
    if language is not None:
        xml.set(ns_elem("xml", "lang"), language)

    title = extract_p('properties.title', d, None)
    if title is not None:
        etree.SubElement(xml, "title").text = title

    # <subtitle type = "html"
    subtitle = extract_p('properties.abstract', d, None)
    if subtitle is not None:
        etree.SubElement(xml, "subtitle").text = subtitle

    update_date = extract_p('properties.updated', d, None)
    if update_date is not None:
        etree.SubElement(xml, "updated").text = update_date

    authors = [
        axml_author(do) for do in extract_p('properties.authors', d, [])
    ]
    [xml.append(el) for el in authors if el is not None]

    publisher = extract_p('properties.publisher', d, None)
    if publisher is not None:
        etree.SubElement(xml, ns_elem("dc", "publisher")).text = publisher

    creator_application = axml_creator_app(
        extract_p('properties.generator', d, None))
    if creator_application is not None and not is_empty(creator_application):
        xml.append(creator_application)

    creator_display = axml_display(extract_p('properties.display', d, None))
    if creator_display is not None:
        xml.append(creator_display)

    rights = extract_p('properties.rights', d, None)
    if rights is not None:
        etree.SubElement(xml, "rights").text = rights

    time_interval_of_interest = extract_p('properties.date', d, None)
    if time_interval_of_interest is not None:
        etree.SubElement(xml, ns_elem("dc",
                                      "date")).text = time_interval_of_interest

    keywords = [
        axml_category(do) for do in extract_p('properties.categories', d, [])
    ]
    [xml.append(el) for el in keywords if el is not None]

    # here we generate single elements and attach them
    resources = [axml_resource(do) for do in extract_p('features', d, [])]
    [xml.append(el) for el in resources if el is not None]

    return xml
Esempio n. 18
0
def axml_resource(d):
    """
    encodes an OwcResource as dict into atom xml tree
    :param d:
    :return:
    """
    entry = etree.Element("entry", nsmap=ns)

    etree.SubElement(entry, "id").text = d['id']

    geospatial_extent = extract_p('geometry', d, None)
    if geospatial_extent is not None:
        try:
            gml = etree.fromstring(geospatial_extent)
            georss = etree.SubElement(entry, ns_elem("georss", "where"))
            georss.append(gml)
        except Exception as ex:
            log.warning('could encode geometry into georss:where', ex)
            pass

    title = d['properties']['title']
    if title is not None:
        etree.SubElement(entry, "title").text = title

    subtitle = extract_p('properties.abstract', d, None)
    # <content type = "text" >
    if subtitle is not None:
        etree.SubElement(entry, "content").text = subtitle

    update_date = extract_p('properties.updated', d, None)
    if update_date is not None:
        etree.SubElement(entry, "updated").text = update_date

    authors = [
        axml_author(do) for do in extract_p('properties.authors', d, [])
    ]
    [entry.append(el) for el in authors if el is not None]

    publisher = extract_p('properties.publisher', d, None)
    if update_date is not None:
        etree.SubElement(entry, ns_elem("dc", "publisher")).text = publisher

    rights = extract_p('properties.rights', d, None)
    if update_date is not None:
        etree.SubElement(entry, ns_elem("dc", "rights")).text = rights

    temporal_extent = extract_p('properties.date', d, None)
    if temporal_extent is not None:
        etree.SubElement(entry, "date").text = temporal_extent

    keywords = [
        axml_category(do) for do in extract_p('properties.categories', d, [])
    ]
    [entry.append(el) for el in keywords if el is not None]

    resource_metadata = [
        axml_link(do) for do in extract_p('properties.links.via', d, [])
    ]
    [entry.append(el) for el in resource_metadata if el is not None]

    content_description = [
        axml_content(do)
        for do in extract_p('properties.links.alternates', d, [])
    ]
    [entry.append(el) for el in content_description if el is not None]

    preview = [
        axml_link(do) for do in extract_p('properties.links.preview', d, [])
    ]
    [entry.append(el) for el in preview if el is not None]

    content_by_ref = [
        axml_link(do) for do in extract_p('properties.links.data', d, [])
    ]
    [entry.append(el) for el in content_by_ref if el is not None]

    offerings = [
        axml_offering(do) for do in extract_p('properties.offerings', d, [])
    ]
    [entry.append(el) for el in offerings if el is not None]

    # TODO no examples for active attribute
    active = extract_p('properties.active', d, None)
    if active is not None:
        etree.SubElement(entry, "active").text = active

    min_scale_denominator = try_float(
        extract_p('properties.minscaledenominator', d, None))
    # <owc:minScaleDenominator>2500</owc:minScaleDenominator>
    if min_scale_denominator is not None:
        etree.SubElement(entry, ns_elem(
            "owc", "minScaleDenominator")).text = str(min_scale_denominator)

    max_scale_denominator = try_float(
        extract_p('properties.maxscaledenominator', d, None))
    # <owc:maxScaleDenominator>25000</owc:maxScaleDenominator>
    if max_scale_denominator is not None:
        etree.SubElement(entry, ns_elem(
            "owc", "maxScaleDenominator")).text = str(max_scale_denominator)

    # TODO no examples for folder attribute
    folder = extract_p('properties.folder', d, None)
    if folder is not None:
        etree.SubElement(entry, "folder").text = folder

    # xml.append(entry)
    return entry
Esempio n. 19
0
def parse_entry(entry_node):
    """
    parse an aotm entry into a feature/resource dict to build OwcResource from

    :param entry_node: xml element root node of the atom:entry
    :return: dictionary for OwcResource.from_dict()
    """
    resource_base_dict = {
        "type": "Feature",
        "id": None,
        "geometry": None,
        "properties": {
            'title': None,
            'abstract': None,
            'updated': None,
            'date': None,
            'authors': [],
            'publisher': None,
            'rights': None,
            'categories': [],
            "links": {
                "alternates": [],
                "previews": [],
                "data": [],
                "via": [],
            },
            'offerings': [],
            'active': None,
            'minscaledenominator': None,
            'maxscaledenominator': None,
            'folder': None
        }
    }

    # <id>ftp://ftp.remotesensing.org/pub/geotiff/samples/gdal_eg/cea.txt</id>
    val = entry_node.find(util.nspath_eval('atom:id', ns))
    id = util.testXMLValue(val)
    # log.debug("entry :id %s :: %s", id, val)
    resource_base_dict.update({"id": id})

    # <title>GeoTIFF Example</title>
    val = entry_node.find(util.nspath_eval('atom:title', ns))
    title = util.testXMLValue(val)
    # log.debug("entry: title %s :: %s", id, val)
    resource_base_dict['properties'].update({"title": title})

    # <updated>2011-11-01T00:00:00Z</updated>
    val = entry_node.find(util.nspath_eval('atom:updated', ns))
    update_date = util.testXMLValue(val)
    # log.debug("entry: updated %s :: %s", update_date, val)
    resource_base_dict['properties'].update({"updated": update_date})

    # <dc:publisher>
    val = entry_node.find(util.nspath_eval('dc:publisher', ns))
    publisher = util.testXMLValue(val)
    # log.debug("entry: dc:publisher %s :: %s", publisher, val)
    resource_base_dict['properties'].update({"publisher": publisher})

    # <dc:rights>
    val = entry_node.find(util.nspath_eval('dc:rights', ns))
    rights = util.testXMLValue(val)
    # log.debug("entry: rights %s :: %s", rights, val)
    resource_base_dict['properties'].update({"rights": rights})

    # <georss:where>
    val = entry_node.find(util.nspath_eval('georss:where', ns))
    if val is not None:
        if len(list(val)) > 0:
            # xmltxt = etree.tostring(
            #     list(val)[0], encoding='utf8', method='xml')
            xmltxt = element_to_string(list(val)[0], False)
            # TODO here parse geometry??
            # log.debug("entry: geometry %s :: %s", xmltxt, val)
            resource_base_dict.update({"geometry": xmltxt.decode('utf-8')})

    # <content type = "text" > aka subtitle, aka abstract
    val = entry_node.find(util.nspath_eval('atom:content', ns))
    subtitle = util.testXMLValue(val)
    # log.debug("entry: subtitle %s :: %s", subtitle, val)
    resource_base_dict['properties'].update({"abstract": subtitle})

    # <author> ..
    # 		<name>
    #       <email>
    vals = entry_node.findall(util.nspath_eval('atom:author', ns))
    authors = []
    for val in vals:
        val_name = val.find(util.nspath_eval('atom:name', ns))
        val_email = val.find(util.nspath_eval('atom:email', ns))
        val_uri = val.find(util.nspath_eval('atom:uri', ns))
        name = util.testXMLValue(val_name)
        email = util.testXMLValue(val_email)
        uri = util.testXMLValue(val_uri)
        author = {"name": name, "email": email, "uri": uri}
        # log.debug("entry: author %s :: %s", author, vals)
        if not is_empty(author):
            authors.append(author)

    resource_base_dict['properties'].update({"authors": authors})

    # <link rel="enclosure" type="image/png"
    #   length="12345" title="..." href="http://o..."/>
    # <link rel="icon" type="image/png" title="Preview f..."
    #   href="http://..."/>
    # <link rel="via" type="application/vnd.ogc.wms_xml"
    #   title="Original .." href="...."/>
    vals = entry_node.findall(util.nspath_eval('atom:link', ns))
    links_alternates = []
    links_previews = []
    links_data = []
    links_via = []
    for val in vals:
        rel = util.testXMLAttribute(val, 'rel')
        href = util.testXMLAttribute(val, 'href')
        mimetype = util.testXMLAttribute(val, 'type')
        lang = util.testXMLAttribute(val, 'lang')
        title = util.testXMLAttribute(val, 'title')
        length = util.testXMLAttribute(val, 'length')
        link = {
            "href": href,
            "type": mimetype,
            "length": length,
            "lang": lang,
            "title": title,
            "rel": rel
        }
        # log.debug("entry: link %s :: %s", link, vals)
        if link.get("rel") == "alternate" and not is_empty(link):
            links_alternates.append(link)
        elif link.get("rel") == "icon" and not is_empty(link):
            links_previews.append(link)
        elif link.get("rel") == "enclosure" and not is_empty(link):
            links_data.append(link)
        elif link.get("rel") == "via" and not is_empty(link):
            links_via.append(link)
        else:
            log.warning("unknown link type in Ows Resource entry section: %r",
                        link)

    resource_base_dict['properties']['links'].update(
        {"alternates": links_alternates})
    resource_base_dict['properties']['links'].update(
        {"previews": links_previews})
    resource_base_dict['properties']['links'].update({"data": links_data})
    resource_base_dict['properties']['links'].update({"via": links_via})

    # <owc:offering code="http://www.opengis.net/spec/owc-at...">
    #   <owc:content type="image/tiff" href=".."
    #   <owc:offering code="http://www.opengis.net/spec....l">
    # 			<owc:content type="application/gml+xml">
    #   <owc:operation code="GetCapabilities" method="GET"
    #       type="applica..." href="..."
    #       <owc:request type="application/xml"> ..
    #     <owc:styleSet>
    #       <owc:name>raster</owc:name>
    #       <owc:title>Default Raster</owc:title>
    #       <owc:abstract>A sample style that draws a </owc:abstract>
    #       <owc:legendURL href="h...." type="image/png"/>
    #     </owc:styleSet>
    offering_nodes = entry_node.findall(util.nspath_eval('owc:offering', ns))
    offerings = []
    for offering_node in offering_nodes:
        offering_code = util.testXMLAttribute(offering_node, 'code')
        operations = []
        contents = []
        styles = []
        operation_nodes = offering_node.findall(
            util.nspath_eval('owc:operation', ns))
        for op_val in operation_nodes:
            operations_code = util.testXMLAttribute(op_val, 'code')
            http_method = util.testXMLAttribute(op_val, 'method')
            mimetype = util.testXMLAttribute(op_val, 'type')
            request_url = util.testXMLAttribute(op_val, 'href')
            req_content_val = val.find(util.nspath_eval('owc:request', ns))
            req_content = None
            if req_content_val is not None:
                req_content = parse_owc_content(req_content_val)

            # TODO no example for result/response
            op_dict = {
                "code": operations_code,
                "method": http_method,
                "type": mimetype,
                "href": request_url,
                "request": None if is_empty(req_content) else req_content,
                "result": None
            }
            # log.debug("entry: operation %s :: %s", op_dict, vals)
            if not is_empty(op_dict):
                operations.append(op_dict)

        content_nodes = offering_node.findall(
            util.nspath_eval('owc:content', ns))
        for cont_val in content_nodes:
            content_dict = parse_owc_content(cont_val)
            # log.debug("entry: content_dict %s :: %s", content_dict, vals)
            if not is_empty(content_dict):
                contents.append(content_dict)

        style_nodes = offering_node.findall(
            util.nspath_eval('owc:styleSet', ns))
        for style_val in style_nodes:
            val_name = style_val.find(util.nspath_eval('owc:name', ns))
            val_title = style_val.find(util.nspath_eval('owc:title', ns))
            val_abstr = style_val.find(util.nspath_eval('owc:abstract', ns))
            val_uri = style_val.find(util.nspath_eval('owc:legendURL', ns))
            name = util.testXMLValue(val_name)
            title = util.testXMLValue(val_title)
            abstr = util.testXMLValue(val_abstr)
            legend_url = util.testXMLAttribute(val_uri, 'href')
            style_set = {
                "name": name,
                "title": title,
                "abstract": abstr,
                "default": None,
                "legendURL": legend_url,
                "content": None
            }
            # log.debug("entry: style_set %s :: %s", style_set, vals)
            if not is_empty(style_set):
                styles.append(style_set)

        offering_dict = {
            "code": offering_code,
            "operations": operations,
            "contents": contents,
            "styles": styles
        }
        if offering_code is not None:
            offerings.append(offering_dict)

    resource_base_dict['properties'].update({"offerings": offerings})

    # TODO no examples for active attribute
    # <owc:minScaleDenominator>2500</owc:minScaleDenominator>
    val = entry_node.find(util.nspath_eval('owc:minScaleDenominator', ns))
    min_scale_denominator = util.testXMLValue(val)
    # log.debug("entry: min-scale-... %s :: %s", min_scale_denominator, val)
    resource_base_dict['properties'].update(
        {"minscaledenominator": min_scale_denominator})

    # <owc:maxScaleDenominator>25000</owc:maxScaleDenominator>
    val = entry_node.find(util.nspath_eval('owc:maxScaleDenominator', ns))
    max_scale_denominator = util.testXMLValue(val)
    # log.debug("entry: max_scale_... %s :: %s", max_scale_denominator, val)
    resource_base_dict['properties'].update(
        {"maxscaledenominator": max_scale_denominator})
    # TODO no examples for folder attribute

    return resource_base_dict
Esempio n. 20
0
    if async:
        # Return the execution object. This can be monitored via user-supplied code
        return wps.execute(processid, inputs, output)
    else:
        # Blocks waiting for the execution to complete and then returns the file path to the results
        execution = wps.execute(processid, inputs, output)

        # sleepSecs = sleepSecs
        err_count = 1

        while not execution.isComplete():
            try:
                monitorExecution(execution, sleepSecs, download=False)  # monitors for success
                err_count = 1
            except Exception:
                log.warning('An error occurred while checking status, checking again. ' +
                            'Sleeping {} seconds...'.format(sleepSecs))
                err_count += 1

                if err_count > WPS_attempts:
                    raise Exception('The status document failed to return, status checking has aborted. ' +
                                    'There has been a network or server issue preventing the status document ' +
                                    'from being retrieved, the request may still be running. For more information, ' +
                                    'check the status url {}'.format(execution.statusLocation))
                # sleep(sleepSecs)    # monitorExecution calls checkStatus which handles the sleeping

        # Check for any problems with the execution result
        _check_for_execution_errors(execution)

        # Attempt to retrieve the file
        if outputFilePath is None:
            outputFilePath = 'gdp_' + processid.replace('.', '-') + '_' + strftime("%Y-%m-%dT%H-%M-%S-%Z")