Source code for gavo.protocols.oaiclient

"""
A simple client of OAI-http.

This includes both some high-level functions and rudimentary parsers
that can serve as bases for more specialized parsers.
"""

#c Copyright 2008-2023, the GAVO project <gavo@ari.uni-heidelberg.de>
#c
#c This program is free software, covered by the GNU GPL.  See the
#c COPYING file in the source distribution.


import io
import hashlib
import os
import pickle
import re
import urllib.parse
from xml import sax
from xml.sax import saxutils

from gavo import base
from gavo import utils


[docs]class FailedQuery(Exception): def __init__(self, msg, code="?", value="?"): Exception.__init__(self, msg) self.code, self.value = code, value
[docs]class NoRecordsMatch(Exception): pass
[docs]class PrefixIsTaken(Exception): pass
# Canonical prefixes, i.e., essentially fixed prefixes for certain # namespaces. This is all an ugly nightmare, but this is what you # get for having namespace prefixes in attributes.
[docs]class CanonicalPrefixes(object): """a self-persisting dictionary of the prefixes we use in our OAI interface. CanonicalPrefixes objects are constructed with the name of a pickle file containing a list of (prefix, uri) pairs. This reproduces some code from stanxml.NSRegistry, but we want that stuff as instance method here, not as class method. """ def __init__(self, pickleName): self.pickleName = pickleName self._registry = {} self._reverseRegistry = {} self._loadData()
[docs] def registerPrefix(self, prefix, ns, save=True): if prefix in self._registry: if ns!=self._registry[prefix]: raise PrefixIsTaken(prefix) return self._registry[prefix] = ns if ns in self._reverseRegistry and self._reverseRegistry[ns]!=prefix: raise ValueError("Namespace %s already has prefix %s, will" " not clobber with %s"%(ns, self._reverseRegistry[ns], prefix)) self._reverseRegistry[ns] = prefix if save: self._saveData()
[docs] def registerPrefixOrMakeUp(self, prefix, ns): """registers prefix for ns or, if prefix is already taken, makes up a new prefix for the namespace URI ns. """ try: self.registerPrefix(prefix, ns) except PrefixIsTaken: origPrefix, uniquer = prefix, 0 while True: try: prefix = origPrefix+str(uniquer) self.registerPrefix(prefix, ns) except PrefixIsTaken: uniquer += 1 else: break
[docs] def getPrefixForNS(self, ns): try: return self._reverseRegistry[ns] except KeyError: raise base.NotFoundError(ns, "XML namespace", "registry of XML namespaces.")
[docs] def haveNS(self, ns): return ns in self._reverseRegistry
[docs] def getNSForPrefix(self, prefix): try: return self._registry[prefix] except KeyError: raise base.NotFoundError(prefix, "XML namespace prefix", "registry of prefixes.")
[docs] def iterNS(self): return iter(self._registry.items())
def _fillFromPairs(self, pairs): """fills the instance from a list of prefix, uri pairs. Pairs is what is stored in the pickle. """ for prefix, uri in pairs: self.registerPrefix(prefix, uri, save=False) def _bootstrap(self): """sets up our canonical prefixes from DaCHS' (stanxml) namespace registry. """ from gavo import api #noflake: hope most prefixes are registered after that from gavo.utils import stanxml self._fillFromPairs(iter(stanxml.NSRegistry._registry.items())) self._saveData() def _loadData(self): try: with open(self.pickleName, "rb") as f: self._fillFromPairs(pickle.load(f)) except IOError: # most likely, the file does not exist yet base.ui.notifyWarning("Starting new canonical prefixes") self._bootstrap() def _saveData(self): toPersist = list(sorted(self._registry.items())) try: with open(self.pickleName+".tmp", "wb") as f: pickle.dump(toPersist, f) os.rename(self.pickleName+".tmp", self.pickleName) except IOError as msg: base.ui.notifyWarning("Could not persist canonical prefixes: %s"% msg)
[docs]def getCanonicalPrefixes(): return CanonicalPrefixes(os.path.join(base.getConfig("cacheDir"), "rrOaiPrefixes.pickle"))
[docs]class OAIErrorMixin(object): def _end_error(self, name, attrs, content): if attrs["code"]=="noRecordsMatch": raise NoRecordsMatch() raise FailedQuery("Registry bailed with code %s, value %s"%( attrs["code"], content), attrs["code"], content)
[docs]class IdParser(utils.StartEndHandler, OAIErrorMixin): """A parser for simple OAI-PMH headers. Records end up as a list of dictionaries in the recs attribute. """ resumptionToken = None def __init__(self, initRecs=None): utils.StartEndHandler.__init__(self) if initRecs is None: self.recs = [] else: self.recs = initRecs
[docs] def getResult(self): return self.recs
def _end_identifier(self, name, attrs, content): self.recs[-1]["id"] = content.strip() def _end_datestamp(self, name, attrs, content): try: self.recs[-1]["date"] = utils.parseISODT(content) except ValueError: # don't fail just because of a broken date self.recs[-1]["date"] = None def _start_header(self, name, attrs): self.recs.append({}) def _end_resumptionToken(self, name, attrs, content): if content.strip(): self.resumptionToken = content
[docs]class RecordParser(IdParser, OAIErrorMixin): """A simple parser for ivo_vor records. This only pulls out a number of the most salient items; more will probably follow as needed. """ def _end_title(self, name, attrs, content): if self.getParentTag()=="Resource": self.recs[-1][name] = content def _end_email(self, name, attrs, content): if self.getParentTag()=="contact": self.recs[-1]["contact.email"] = content def _end_name(self, name, attrs, content): if self.getParentTag()=="creator": self.recs[-1].setdefault(name, []).append(content) def _end_subject(self, name, attrs, content): self.recs[-1].setdefault(name, []).append(content) def _handleContentChild(self, name, attrs, content): if self.getParentTag()=="content": self.recs[-1][name] = content _end_description = _end_source = _end_referenceURL = \ _handleContentChild def _end_datestamp(self, name, attrs, content): # nuke IdParser implementation, we take our date from ri:Resource pass def _startResource(self, name, attrs): self.recs.append({}) def _end_Resource(self, name, attrs, content): self.recs[-1]["date"] = utils.parseISODT(attrs["updated"]) def _end_accessURL(self, name, attrs, content): self.recs[-1].setdefault(name, []).append(content)
[docs]class OAIRecordsParser(sax.ContentHandler, OAIErrorMixin): """a SAX ContentHandler generating tuples of some record-level metadata and pre-formatted XML of simple implementation of the OAI interface. canonicalPrefixes is a CanonicalPrefixesInstance built from res/canonicalPrefixes.pickle Note that we *require* that records actually carry ivo_vor metadata. Note that this is nothing people should need in normal operation. GAVO Heidelberg needs this for infrastructure services (including OAI on RegTAP, but we needed it beyond that). """ # attribute names the values of which should be disambiguated to # reduce the likelihood of clashes when ids are reused between documents. # (see _normalizeAttrs) _referringAttributeNames = set(["id", "ref", "coord_system_id"]) resumptionToken = None def __init__(self, canonicalPrefixes=None): self.canonicalPrefixes = canonicalPrefixes or getCanonicalPrefixes() sax.ContentHandler.__init__(self) self.buffer = None self.writer = None self.rowdicts = [] self.prefixMap = {} self.prefixesToTranslate = {}
[docs] def startPrefixMapping(self, prefix, uri): self.prefixMap.setdefault(prefix, []).append(uri) if prefix: # Here, we make sure we find a globally unique prefix for every # namespace URI. canonicalPrefixes makes sure this unique prefix # is persistent and later available to the OAI interface if not self.canonicalPrefixes.haveNS(uri): self.canonicalPrefixes.registerPrefixOrMakeUp(prefix, uri) canonPrefix = self.canonicalPrefixes.getPrefixForNS(uri) if prefix!=canonPrefix or prefix in self.prefixesToTranslate: self.prefixesToTranslate.setdefault(prefix, []).append(canonPrefix)
[docs] def endPrefixMapping(self, prefix): self.prefixMap[prefix].pop() if prefix in self.prefixesToTranslate: self.prefixesToTranslate[prefix].pop() if not self.prefixesToTranslate[prefix]: del self.prefixesToTranslate[prefix]
[docs] def startElementNS(self, namePair, ignored, attrs): ns, name = namePair if ns is not None: name = self.canonicalPrefixes.getPrefixForNS(ns)+":"+name if attrs: attrs = self._normalizeAttrs(attrs) if name in self.startHandlers: self.startHandlers[name](self, name, attrs) if self.writer: self.writer.startElement(name, attrs) self._lastChars = []
[docs] def endElementNS(self, namePair, name): ns, name = namePair if ns is not None: name = self.canonicalPrefixes.getPrefixForNS(ns)+":"+name if self.writer: self.writer.endElement(name) if name in self.endHandlers: self.endHandlers[name](self, name)
[docs] def characters(self, stuff): if self.writer: self.writer.characters(stuff) # Hack, see _getLastContent self._lastChars.append(stuff)
[docs] def normalizeNamespace(self, name): """fixes the namespace prefix of name if necessary. name must be a qualified name, i.e., contain exactly one colon. "normalize" here means make sure the prefix matches our canonical prefix and change it to the canonical one if necessary. """ prefix, base = name.split(":") if prefix not in self.prefixesToTranslate: return name return self.prefixesToTranslate[prefix][-1]+":"+base
def _normalizeAttrs(self, attrs): """fixes attribute name and attribute value namespaces if necessary. It also always checks for xsi:type and fixes namespaced attribute values as necessary. See also normalizeNamespace. """ newAttrs = {} for ns, name in list(attrs.keys()): value = attrs[(ns, name)] if ns is None: newName = name else: newName = self.canonicalPrefixes.getPrefixForNS(ns)+":"+name if newName=="xsi:type": if ":" in value: value = self.normalizeNamespace(value) # to uniqueify id/ref-pairs, prepend an md5-digest of the ivoid # to selected ids. This isn't guaranteed to always work, but # if someone is devious enough to cause collisions here, they # deserve no better. if newName in self._referringAttributeNames: value = value+hashlib.md5(self.ivoid.encode("utf-8")).hexdigest() newAttrs[newName] = value return newAttrs def _getLastContent(self): """returns the entire character content since the last XML event. """ return "".join(self._lastChars)
[docs] def notifyError(self, err): self._errorOccurred = True
[docs] def shipout(self, role, record): # see _end_identifier for an explanation of the following condition if self.ivoid is None: return # see our docstring on why we need the following if not self.metadataSeen: return if self._errorOccurred: return # _start_header sets _isDeleted if self._isDeleted: return self.rowdicts.append((role, record))
def _start_oai_header(self, name, attrs): self._isDeleted = attrs.get("status", "").lower()=="deleted" def _start_oai_record(self, name, attrs): self._errorOccurred = False self.curXML = io.BytesIO() self.writer = saxutils.XMLGenerator(self.curXML, "utf-8") self.writer.startDocument() self.ivoid, self.updated = None, None self.metadataSeen = False self.oaiSets = set() def _start_ri_Resource(self, anme, attrs): self.metadataSeen = True def _end_oai_record(self, name): if self.writer is not None: self.writer.endDocument() # yeah, we decode the serialized result right away; it's easier # to store character streams in the DB the way I'm doing things. oaixml = self.curXML.getvalue().decode("utf-8") # unfortunately, XMLGenerator insists on adding an XML declaration, # which I can't have here. I remove it manually if oaixml.startswith("<?xml"): oaixml = oaixml[oaixml.index("?>")+2:] self.shipout("oairecs", { "ivoid": self.ivoid, "updated": self.updated, "oaixml": oaixml}) self.writer = None self.curXML = None def _end_oai_setSpec(self, name): self.oaiSets.add(self._getLastContent()) def _end_oai_identifier(self, name): self.ivoid = self._getLastContent().lower().strip() def _end_oai_resumptionToken(self, name): self.resumptionToken = self._getLastContent() def _start_oai_error(self, name, attrs): self._errorAttrs = attrs def _end_oai_error(self, name): self._end_error(name, self._errorAttrs, self._getLastContent())
[docs] def getResult(self): return self.rowdicts
startHandlers = { "oai:record": _start_oai_record, "oai:header": _start_oai_header, "ri:Resource": _start_ri_Resource, "oai:error": _start_oai_error, } endHandlers = { "oai:record": _end_oai_record, "oai:setSpec": _end_oai_setSpec, "oai:resumptionToken": _end_oai_resumptionToken, "oai:identifier": _end_oai_identifier, "oai:error": _end_oai_error, }
[docs]class ServerProperties(object): """A container for what an OAI-PMH server gives in response to identify. """ repositoryName = None baseURL = None protocolVersion = None adminEmails = () earliestDatestamp = None deletedRecord = None granularity = None repositoryName = None compressions = () def __init__(self): self.adminEmails = [] self.compressions = [] self.descriptions = []
[docs] def set(self, name, value): setattr(self, name, value)
[docs] def add(self, name, value): getattr(self, name).append(value)
[docs]class IdentifyParser(utils.StartEndHandler, OAIErrorMixin): """A parser for the result of the identify operation. The result (an instance of ServerProperties) is in the serverProperties attribute. """ resumptionToken = None serverProperties = None
[docs] def getResult(self): if not self.serverProperties: raise FailedQuery("Identify request did not contain an Identify" " element") return self.serverProperties
def _start_Identify(self, name, attrs): self.serverProperties = ServerProperties() def _endListThing(self, name, attrs, content): self.serverProperties.add(name+"s", content.strip()) _end_adminEmail = _end_compression \ = _endListThing def _endStringThing(self, name, attrs, content): self.serverProperties.set(name, content.strip()) _end_repositoryName = _end_baseURL = _end_protocolVersion \ = _end_granularity = _end_deletedRecord = _end_earliestDatestamp \ = _end_repositoryName = _endStringThing
[docs]class OAIQuery(object): """A container for queries to OAI interfaces. Construct it with the oai endpoint and the OAI verb, plus some optional query attributes. If you want to retain or access the raw responses of the server, pass a contentCallback function -- it will be called with a byte string containing the payload of the server response if it was parsed successfully. Error responses cannot be obtained in this way. The OAIQuery is constructed with OAI-PMH parameters (verb, startDate, endDate, set, metadataPrefix; see the OAI-PMH docs for what they mean, only verb is mandatory). In addition, you can pass granularity, which is the granularity """ startDate = None endDate = None set = None registry = None metadataPrefix = None # maxRecords is mainly used in test_oai; that's why there's no # constructor parameter for it maxRecords = None # a timeout on HTTP operations timeout = 100 def __init__(self, registry, verb, startDate=None, endDate=None, set=None, metadataPrefix="ivo_vor", identifier=None, contentCallback=None, granularity=None): self.registry = registry self.verb, self.set = verb, set self.startDate, self.endDate = startDate, endDate self.identifier = identifier self.metadataPrefix = metadataPrefix self.contentCallback = contentCallback self.granularity = granularity if not self.granularity: self.granularity = "YYYY-MM-DD"
[docs] def getKWs(self, **moreArgs): """returns a dictionary containing query keywords for OAI interfaces from what's specified on the command line. """ kws = {"verb": self.verb} if self.metadataPrefix: kws["metadataPrefix"] = self.metadataPrefix kws.update(moreArgs) if self.granularity=='YY-MM-DD': dateFormat = "%Y-%m-%d" else: dateFormat = "%Y-%m-%dT%H:%M:%SZ" if self.startDate: kws["from"] = self.startDate.strftime(dateFormat) if self.endDate: kws["until"] = self.endDate.strftime(dateFormat) if self.set: kws["set"] = self.set if self.maxRecords: kws["maxRecords"] = str(self.maxRecords) if self.identifier: kws["identifier"] = self.identifier if "resumptionToken" in kws: kws = {"resumptionToken": kws["resumptionToken"], "verb": kws["verb"]} return kws
[docs] def doHTTP(self, **moreArgs): """returns the result of parsing the current query plus moreArgs to the current registry. The result is returned as a string. """ srcURL = self.registry.rstrip("?" )+"?"+self._getOpQS(**self.getKWs(**moreArgs)) base.ui.notifyInfo("OAI query %s"%srcURL) f = utils.urlopenRemote(srcURL, timeout=self.timeout) res = f.read() f.close() return res
def _getOpQS(self, **args): """returns a properly quoted HTTP query part from its (keyword) arguments. """ # we don't use urllib.urlencode to not encode empty values like a=&b=val qString = "&".join("%s=%s"%(k, urllib.parse.quote(v)) for k, v in args.items() if v) return "%s"%(qString)
[docs] def talkOAI(self, parserClass): """processes an OAI dialogue for verb using the IdParser-derived parserClass. """ res = self.doHTTP(verb=self.verb) if not res.strip(): # empty reply is not admissible XML here raise FailedQuery("Empty HTTP response") handler = parserClass() try: xmlReader = sax.make_parser() xmlReader.setFeature(sax.handler.feature_namespaces, True) xmlReader.setContentHandler(handler) xmlReader.parse(io.BytesIO(res)) if self.contentCallback: self.contentCallback(res) except NoRecordsMatch: return [] oaiResult = handler.getResult() while handler.resumptionToken is not None: resumptionToken = handler.resumptionToken handler = parserClass(oaiResult) try: res = self.doHTTP(verb=self.verb, resumptionToken=resumptionToken) sax.parseString(res, handler) if self.contentCallback: self.contentCallback(res) except NoRecordsMatch: break return oaiResult
[docs]def getIdentifiers(registry, startDate=None, endDate=None, set=None, granularity=None): """returns a list of "short" records for what's in the registry specified by args. """ q = OAIQuery(registry, verb="ListIdentifiers", startDate=startDate, endDate=endDate, set=set) return q.talkOAI(IdParser)
[docs]def getRecords(registry, startDate=None, endDate=None, set=None, granularity=None): """returns a list of "long" records for what's in the registry specified by args. parser should be a subclass of RecordParser; otherwise, you'll miss resumption and possibly other features. """ q = OAIQuery(registry, verb="ListRecords", startDate=startDate, endDate=endDate, set=set, granularity=granularity) return q.talkOAI(RecordParser)
def _addCanonicalNSDecls(xmlLiteral): """adds XML namespace declarations for namespace prefixes we suspect in xmlLiteral. This is an ugly hack based on REs necessary because in the OAIRecordsParser we discard the namespace declarations. It won't work with CDATA sections, and it'll make a hash of things if namespace declarations are already present. However, for the use case of making the mutilated resource records coming out of the OAIRecordsParser valid, it will just do. Without an XML schema and a full parse (which of course is impossible without the necessary declarations), this is, really, not possible. But the whole idea of canonical namespace prefixes is a mess, and so we hack along; in particular, we accept any string of the form \w+: within what looks like an XML tag as a namespace. Oh my. """ prefixesUsed = set() for elementContent in re.finditer("<[^>]+>", xmlLiteral): prefixesUsed |= set(re.findall("([a-zA-Z_-][\w-]*):[a-zA-Z_]", elementContent.group())) cp = getCanonicalPrefixes() nsDecls = " ".join('xmlns:%s=%s'%( pref, utils.escapeAttrVal(cp.getNSForPrefix(pref))) for pref in prefixesUsed) return re.sub("<([\w:-]+)", r"<\1 "+nsDecls, xmlLiteral, 1)
[docs]def getRecord(registry, identifier): """returns the XML form of an OAI-PMH record for identifier from the OAI-PMH endpoint at URL registry. This uses the OAIRecordsParser which enforces canonical prefixes, and the function will add their declarations as necessary. This also means that evil registry records could be broken by us. """ q = OAIQuery(registry, verb="GetRecord", identifier=identifier) res = q.talkOAI(OAIRecordsParser) dest, row = res[0] assert dest=='oairecs' return _addCanonicalNSDecls(row["oaixml"])
[docs]def parseRecord(recordXML): """returns some main properties from an XML-encoded VOResource record. recordXML can be an OAI-PMH response or just a naked record. If multiple records are contained in recordXML, only the first will be returned. What's coming back is a dictionary as produced by RecordParser. """ handler = RecordParser() sax.parseString(recordXML, handler) return handler.recs[0]
[docs]def getServerProperties(registry): """returns a ServerProperties instance for registry. In particular, you can retrieve the granularity argument that actually matches the registry from the result's granularity attribute. """ q = OAIQuery(registry, verb="Identify", metadataPrefix=None) return q.talkOAI(IdentifyParser)