コード例 #1
0
def Prjpopupeditpreset(PrjId):
    sql = """select p.projid,title,initclassiflist
                  ,(select string_agg(pts.id::varchar,',') objtaxon from projects_taxo_stat pts where pts.projid=p.projid)
            from projects p
            """
    if not current_user.has_role(database.AdministratorLabel):
        sql += "  Join projectspriv pp on p.projid = pp.projid and pp.member=%d" % (
            current_user.id, )
    sql += " order by upper(title) "
    Prj = GetAll(sql, cursor_factory=psycopg2.extras.RealDictCursor)
    # Prj2=Prj[:]
    # for i in range(200):Prj.extend(Prj2) # for test on bigger list
    TaxonList = {-1}
    for P in Prj:
        lst = ntcv(P['initclassiflist']) + "," + ntcv(P['objtaxon'])
        for t in lst.split(','):
            if t.isdecimal():
                TaxonList.add(int(t))
    # txt=",".join((str(x) for x in TaxonList))
    TaxoMap = GetAssoc2Col(
        "select id,display_name  from taxonomy where id = any (%s)",
        [[x for x in TaxonList]])
    # txt = str(TaxoMap)
    txt = ""
    for P in Prj:
        result = []
        initclassiflist = {
            int(x.strip())
            for x in ntcv(P['initclassiflist']).split(',') if x.isdecimal()
        }
        objtaxon = {
            int(x.strip())
            for x in ntcv(P['objtaxon']).split(',') if x.isdecimal()
        }
        objtaxon.difference_update(initclassiflist)
        for t in initclassiflist:
            resolved = TaxoMap.get(int(t), None)
            if resolved:
                result.append(resolved)
        P['presetids'] = ",".join((str(x) for x in initclassiflist))
        P['preset'] = ", ".join(sorted(result))
        result = []
        for t in objtaxon:
            resolved = TaxoMap.get(int(t), None)
            if resolved:
                result.append(resolved)
        P['objtaxonnotinpreset'] = ", ".join(sorted(result))
        P['objtaxonids'] = ",".join((str(x) for x in objtaxon))

    # construction de la liste distincte tous les taxon
    return render_template('project/popupeditpreset.html', Prj=Prj, txt=txt)
コード例 #2
0
ファイル: browsetaxo.py プロジェクト: ecotaxa/ecotaxoserver
def browsetaxotsvexport():
    sql = """select t.id,t.parent_id,t.name,t.taxotype,t.taxostatus
      ,t.id_source,t.source_url,t.source_desc
      ,t.creator_email,to_char(t.creation_datetime,'yyyy-mm-dd hh24:mi,ss') creation_datetime
      ,t.id_instance,t.rename_to
      ,to_char(t.lastupdate_datetime,'yyyy-mm-dd hh24:mi:ss') lastupdate_datetime
      ,i.name instance_name,tr.display_name rename_to_name,t.nbrobj,t.nbrobjcum
      ,t.display_name,{}
    from taxonomy t 
    LEFT JOIN ecotaxainst i on t.id_instance=i.id
    LEFT JOIN taxonomy tr on tr.id=t.rename_to
    {}
    order by 1
    """.format(SQLTreeSelect, SQLTreeJoin)
    lst = GetAll(sql)
    t = []
    t.append(
        "id\tparent_id\tname\ttaxotype\ttaxostatus\tid_source\tsource_url\tsource_desc\tcreator_email\tcreation_datetime"
        +
        "\tid_instance\trename_to\tlastupdate_datetime\tinstance_name\trename_to_name\tnbrobj\tnbrobjcum\tdisplay_name\tlineage"
    )
    for l in lst:
        # t.append("{id},{parent_id}".format(l))
        t.append("\t".join(
            (str(ntcv(x)).replace("\n",
                                  "\\n").replace("\t",
                                                 "\\t").replace("\r", "")
             for x in l[0:99])))
    return Response("\n".join(t),
                    mimetype="text/tsv",
                    headers={
                        "Content-Disposition":
                        "attachment; filename=taxoexport_%s.tsv" %
                        datetime.datetime.utcnow().strftime('%Y%m%d_%H%M%S')
                    })
コード例 #3
0
def searchtaxo():
    term=gvg("q")
    if len(term)<=2:
        # return "[]"
        if not current_user.is_authenticated:
            return "[]"
        # current_user.id
        with app.MRUClassif_lock:
            # app.MRUClassif[current_user.id]=[{"id": 2904, "pr": 0, "text": "Teranympha (Eucomonymphidae-Teranymphidae)"},
         # {"id": 12488, "pr": 0, "text": "Teranympha mirabilis "},
         # {"id": 76677, "pr": 0, "text": "Terasakiella (Methylocystaceae)"},
         # {"id": 82969, "pr": 0, "text": "Terasakiella pusilla "}]
            return json.dumps(app.MRUClassif.get(current_user.id,[])) # gère les MRU en utilisant les classif
    ltfound=term.find('<')>0
    SQLWith="""
    """
    # * et espace comme %
    terms=[x.lower().replace("*","%").replace(" ","%")+R"%" for x in term.split('<')]
    param={'term':terms[0]} # le premier term est toujours appliqué sur le display name
    ExtraWhere=ExtraFrom=""
    if len(terms)>1:
        ExtraFrom = SQLTreeJoin
        terms = ['%%<'+x.replace("%","%%").replace("*","%%").replace(" ","%%")  for x in terms[1:]]
        termsSQL=QuotedString("".join(terms)).getquoted().decode('iso-8859-15','strict')
        ExtraWhere= ' and '+SQLTreeExp+" ilike "+termsSQL
    sql="""SELECT tf.id, tf.display_name as name
          ,0 FROM taxonomy tf
          {0}
          WHERE  lower(tf.display_name) LIKE %(term)s  {1}
          order by lower(tf.display_name) limit 200""".format(ExtraFrom,ExtraWhere)

    PrjId=gvg("projid")
    if PrjId!="":
        PrjId=int(PrjId)
        Prj=database.Projects.query.filter_by(projid=PrjId).first()
        if ntcv(Prj.initclassiflist) != "":
            InitClassif=Prj.initclassiflist
            InitClassif=", ".join(["("+x.strip()+")" for x in InitClassif.split(",") if x.strip()!=""])
            #             ,tf.name||case when p1.name is not null and tf.name not like '%% %%'  then ' ('||p1.name||')' else ' ' end as name

            sql="""
            SELECT tf.id
            ,tf.display_name as name
            , case when id2 is null then 0 else 1 end inpreset FROM taxonomy tf
            join (select t.id id1,c.id id2 FROM taxonomy t
            full JOIN (VALUES """+InitClassif+""") c(id) ON t.id = c.id
                 WHERE  lower(display_name) LIKE %(term)s) tl2
            on tf.id=coalesce(id1,id2)
            """+ExtraFrom+"""
              WHERE  lower(tf.display_name) LIKE %(term)s """+ExtraWhere+"""
            order by inpreset desc,lower(tf.display_name),name limit 200 """
    res = GetAll(sql, param,debug=False)
    return json.dumps([dict(id=r[0],text=r[1],pr=r[2]) for r in res])
コード例 #4
0
ファイル: taskimportupdate.py プロジェクト: obs6/ecotaxa_dev
    def SPStep1(self):
        logging.info("Input Param = %s" % (self.param.__dict__))
        logging.info("Start Step 1")
        Prj = database.Projects.query.filter_by(
            projid=self.param.ProjectId).first()
        WarnMessages = []
        if getattr(self.param, 'IntraStep', 0) == 0:
            #Sous tache 1 On dezippe ou on pointe sur le repertoire source.
            if self.param.InData.lower().endswith("zip"):
                logging.info("SubTask0 : Unzip File on temporary folder")
                self.UpdateProgress(1, "Unzip File on temporary folder")
                self.param.SourceDir = os.path.normpath(
                    os.path.join(
                        os.path.dirname(os.path.realpath(__file__)),
                        "../../temptask/task%06d/data" % (int(self.task.id))))
                if not os.path.exists(self.param.SourceDir):
                    os.mkdir(self.param.SourceDir)
                with zipfile.ZipFile(self.param.InData, 'r') as z:
                    z.extractall(self.param.SourceDir)
            else:
                self.param.SourceDir = self.param.InData
            self.param.IntraStep = 1

        if self.param.IntraStep == 1:
            self.param.Mapping = {}  # Reset à chaque Tentative
            # Import du mapping existant dans le projet
            for k, v in DecodeEqualList(Prj.mappingobj).items():
                self.param.Mapping['object_' + v] = {
                    'table': 'obj_field',
                    'title': v,
                    'type': k[0],
                    'field': k
                }
            for k, v in DecodeEqualList(Prj.mappingsample).items():
                self.param.Mapping['sample_' + v] = {
                    'table': 'sample',
                    'title': v,
                    'type': k[0],
                    'field': k
                }
            for k, v in DecodeEqualList(Prj.mappingacq).items():
                self.param.Mapping['acq_' + v] = {
                    'table': 'acq',
                    'title': v,
                    'type': k[0],
                    'field': k
                }
            for k, v in DecodeEqualList(Prj.mappingprocess).items():
                self.param.Mapping['process_' + v] = {
                    'table': 'process',
                    'title': v,
                    'type': k[0],
                    'field': k
                }
            ProjectWasEmpty = len(self.param.Mapping) == 0
            self.param.TaxoFound = {}  # Reset à chaque Tentative
            self.param.UserFound = {}  # Reset à chaque Tentative
            self.param.steperrors = []  # Reset des erreurs
            # La version 1.1 normalise le champ 'acq_instrument' donc s'il etait mappé sur un txx on le renome pour que la MAJ aille dans le champ instrument
            if 'acq_instrument' in self.param.Mapping:
                self.param.Mapping['acq_instrument_old'] = self.param.Mapping[
                    'acq_instrument']
                self.param.Mapping['acq_instrument_old'][
                    'title'] = 'instrument_old'
                del self.param.Mapping['acq_instrument']
            # recuperation de toutes les paire objet/Images du projet
            logging.info("SubTask1 : Analyze TSV Files")
            self.UpdateProgress(2, "Analyze TSV Files")
            self.LastNum = {x: {'n': 0, 't': 0} for x in PredefinedTables}
            # Extraction des Max des champs
            for m in self.param.Mapping.values():
                v = int(m['field'][1:])
                if v > self.LastNum[m['table']][m['field'][0]]:
                    self.LastNum[m['table']][m['field'][0]] = v
            sd = Path(self.param.SourceDir)
            self.param.TotalRowCount = 0
            Seen = set(
            )  # Memorise les champs pour lesquels il y a des valeurs
            ClassifIDSeen = set()
            for filter in ("**/ecotaxa*.txt", "**/ecotaxa*.tsv"):
                for CsvFile in sd.glob(filter):
                    relname = CsvFile.relative_to(
                        sd)  # Nom relatif à des fins d'affichage uniquement
                    logging.info("Analyzing file %s" % (relname.as_posix()))
                    with open(CsvFile.as_posix(),
                              encoding='latin_1') as csvfile:
                        # lecture en mode dictionnaire basé sur la premiere ligne
                        rdr = csv.DictReader(csvfile,
                                             delimiter='\t',
                                             quotechar='"')
                        #lecture la la ligne des types (2nd ligne du fichier
                        LType = {
                            champ.strip(" \t").lower(): v
                            for champ, v in rdr.__next__().items()
                        }
                        # Fabrication du mapping
                        ListeChamps = [
                            champ.strip(" \t").lower()
                            for champ in rdr.fieldnames
                        ]
                        for ColName in ListeChamps:
                            if ColName in self.param.Mapping or ColName in (
                                    'object_annotation_parent_category',
                                    'object_annotation_hierarchy'):
                                continue  # Le champ à déjà été détecté OU un des champs Bani de L'importation car fruit de l'export.
                            ColSplitted = ColName.split("_", 1)
                            if len(ColSplitted) != 2:
                                self.LogErrorForUser(
                                    "Invalid Header '%s' in file %s. Format must be Table_Field. Field ignored"
                                    % (ColName, relname.as_posix()))
                                continue
                            Table = ColSplitted[
                                0]  # On isole la partie table avant le premier _
                            if ColName in PredefinedFields:
                                Table = PredefinedFields[ColName]['table']
                                self.param.Mapping[ColName] = PredefinedFields[
                                    ColName]
                            else:  # champs non predefinis donc dans nXX ou tXX
                                if Table == "object":
                                    Table = "obj_field"
                                if not Table in PredefinedTables:
                                    self.LogErrorForUser(
                                        "Invalid Header '%s' in file %s. Table Incorrect. Field ignored"
                                        % (ColName, relname.as_posix()))
                                    continue
                                if Table != 'obj_head' and Table != 'obj_field':  # Dans les autres tables les types sont forcés à texte
                                    SelType = 't'
                                else:
                                    if LType[ColName] not in PredefinedTypes:
                                        self.LogErrorForUser(
                                            "Invalid Type '%s' for Field '%s' in file %s. Incorrect Type. Field ignored"
                                            % (LType[champ], ColName,
                                               relname.as_posix()))
                                        continue
                                    SelType = PredefinedTypes[LType[ColName]]
                                self.LastNum[Table][SelType] += 1
                                self.param.Mapping[ColName] = {
                                    'table':
                                    Table,
                                    'field':
                                    SelType +
                                    "%02d" % self.LastNum[Table][SelType],
                                    'type':
                                    SelType,
                                    'title':
                                    ColSplitted[1]
                                }
                                logging.info("New field %s found in file %s",
                                             ColName, relname.as_posix())
                                if not ProjectWasEmpty:
                                    WarnMessages.append(
                                        "New field %s found in file %s" %
                                        (ColName, relname.as_posix()))
                        # Test du contenu du fichier
                        RowCount = 0
                        for lig in rdr:
                            RowCount += 1
                            for champ in rdr.fieldnames:
                                ColName = champ.strip(" \t").lower()
                                m = self.param.Mapping.get(ColName, None)
                                if m is None:
                                    continue  # Le champ n'est pas considéré
                                v = CleanValue(lig[champ])
                                Seen.add(
                                    ColName
                                )  # V2.0 (1.1 sur main import) si la colonne est présente c'est considéré Seen, avant il fallait avoir vu une valeur.
                                if v != "":  # si pas de valeurs, pas de controle
                                    if m['type'] == 'n':
                                        vf = ToFloat(v)
                                        if vf is None:
                                            self.LogErrorForUser(
                                                "Invalid float value '%s' for Field '%s' in file %s."
                                                %
                                                (v, champ, relname.as_posix()))
                                        elif ColName == 'object_lat':
                                            if vf < -90 or vf > 90:
                                                self.LogErrorForUser(
                                                    "Invalid Lat. value '%s' for Field '%s' in file %s. Incorrect range -90/+90°."
                                                    % (v, champ,
                                                       relname.as_posix()))
                                        elif ColName == 'object_long':
                                            if vf < -180 or vf > 180:
                                                self.LogErrorForUser(
                                                    "Invalid Long. value '%s' for Field '%s' in file %s. Incorrect range -180/+180°."
                                                    % (v, champ,
                                                       relname.as_posix()))
                                        elif ColName == 'object_annotation_category_id':
                                            if self.param.updateclassif == "Y":
                                                ClassifIDSeen.add(int(v))
                                    elif ColName == 'object_date':
                                        try:
                                            datetime.date(
                                                int(v[0:4]), int(v[4:6]),
                                                int(v[6:8]))
                                        except ValueError:
                                            self.LogErrorForUser(
                                                "Invalid Date value '%s' for Field '%s' in file %s."
                                                %
                                                (v, champ, relname.as_posix()))
                                    elif ColName == 'object_time':
                                        try:
                                            v = v.zfill(6)
                                            datetime.time(
                                                int(v[0:2]), int(v[2:4]),
                                                int(v[4:6]))
                                        except ValueError:
                                            self.LogErrorForUser(
                                                "Invalid Time value '%s' for Field '%s' in file %s."
                                                %
                                                (v, champ, relname.as_posix()))
                                    elif ColName == 'object_annotation_category':
                                        if self.param.updateclassif == "Y":
                                            if CleanValue(
                                                    lig.get(
                                                        'object_annotation_category_id',
                                                        '')
                                            ) == '':  # traité que si un ID numérique non spécifié
                                                v = self.param.TaxoMap.get(
                                                    v,
                                                    v)  # Applique le mapping
                                                self.param.TaxoFound[v.lower(
                                                )] = None  #creation d'une entrée dans le dictionnaire.
                                    elif ColName == 'object_annotation_person_name':
                                        if self.param.updateclassif == "Y":
                                            self.param.UserFound[v.lower()] = {
                                                'email':
                                                CleanValue(
                                                    lig.get(
                                                        'object_annotation_person_email',
                                                        ''))
                                            }
                                    elif ColName == 'object_annotation_status':
                                        if self.param.updateclassif == "Y":
                                            if v != 'noid' and v.lower(
                                            ) not in database.ClassifQualRevert:
                                                self.LogErrorForUser(
                                                    "Invalid Annotation Status '%s' for Field '%s' in file %s."
                                                    % (v, champ,
                                                       relname.as_posix()))

                            #Analyse l'existance du fichier Image
                            ObjectId = CleanValue(lig.get('object_id', ''))
                            if ObjectId == '':
                                self.LogErrorForUser(
                                    "Missing object_id on line '%s' in file %s. "
                                    % (RowCount, relname.as_posix()))
                        logging.info("File %s : %d row analysed",
                                     relname.as_posix(), RowCount)
                        self.param.TotalRowCount += RowCount
            if self.param.TotalRowCount == 0:
                self.LogErrorForUser("No object found")
            if len(ClassifIDSeen) > 0:
                ClassifIDFoundInDB = GetAll(
                    "select id from taxonomy where id = any (%s)",
                    [list(ClassifIDSeen)])
                ClassifIDFoundInDB = {int(r['id']) for r in ClassifIDFoundInDB}
                ClassifIDNotFoundInDB = ClassifIDSeen.difference(
                    ClassifIDFoundInDB)
                if len(ClassifIDNotFoundInDB) > 0:
                    msg = "Some specified classif_id doesn't exists, correct them prior to reload %s" % (
                        ",".join([str(x) for x in ClassifIDNotFoundInDB]))
                    self.param.steperrors.append(msg)
                    logging.error(msg)
            self.UpdateProgress(15, "TSV File Parsed" % ())
            # print(self.param.Mapping)
            logging.info("Taxo Found = %s", self.param.TaxoFound)
            logging.info("Users Found = %s", self.param.UserFound)
            logging.info("For Information Not Seen Fields %s",
                         [k for k in self.param.Mapping if k not in Seen])
            if len(self.param.steperrors) > 0:
                self.task.taskstate = "Error"
                self.task.progressmsg = "Some errors founds during file parsing "
                db.session.commit()
                return
            self.param.IntraStep = 2
        if self.param.IntraStep == 2:
            logging.info("Start Sub Step 1.2")
            self.pgcur.execute(
                "select id,lower(name),lower(email) from users where lower(name) = any(%s) or email= any(%s) ",
                ([x for x in self.param.UserFound.keys()
                  ], [x.get('email') for x in self.param.UserFound.values()]))
            # Résolution des noms à partir du nom ou de l'email
            for rec in self.pgcur:
                for u in self.param.UserFound:
                    if u == rec[1] or ntcv(self.param.UserFound[u].get(
                            'email')).lower() == rec[2]:
                        self.param.UserFound[u]['id'] = rec[0]
            logging.info("Users Found = %s", self.param.UserFound)
            NotFoundUser = [
                k for k, v in self.param.UserFound.items()
                if v.get("id") == None
            ]
            if len(NotFoundUser) > 0:
                logging.info("Some Users Not Found = %s", NotFoundUser)
            # récuperation des ID des taxo trouvées
            NotFoundTaxo = []
            ResolveTaxoFound(self.param.TaxoFound, NotFoundTaxo)
            if len(NotFoundTaxo) > 0:
                logging.info("Some Taxo Not Found = %s", NotFoundTaxo)
            if len(NotFoundUser) == 0 and len(NotFoundTaxo) == 0 and len(
                    WarnMessages
            ) == 0:  # si tout est déjà résolue on enchaine sur la phase 2
                self.SPStep2()
            else:
                self.task.taskstate = "Question"
            if len(WarnMessages) > 0:
                self.UpdateProgress(
                    20,
                    "Taxo automatic resolution Done, <span style='color:red;font-weight:bold;'>Some Warning :\n-%s </span>"
                    % ("\n-".join(WarnMessages)))
            else:
                self.UpdateProgress(20, "Taxo automatic resolution Done" % ())
コード例 #5
0
ファイル: taskimportupdate.py プロジェクト: obs6/ecotaxa_dev
    def SPStep2(self):
        logging.info("Start Step 2 : Effective data import")
        logging.info("Taxo Mapping = %s", self.param.TaxoFound)
        logging.info("Users Mapping = %s", self.param.UserFound)
        AstralCache = {
            'date': None,
            'time': None,
            'long': None,
            'lat': None,
            'r': ''
        }
        # Mise à jour du mapping en base
        Prj = database.Projects.query.filter_by(
            projid=self.param.ProjectId).first()
        Prj.mappingobj = EncodeEqualList({
            v['field']: v.get('title')
            for k, v in self.param.Mapping.items()
            if v['table'] == 'obj_field' and v['field'][0] in (
                't', 'n') and v.get('title') != None
        })
        Prj.mappingsample = EncodeEqualList({
            v['field']: v.get('title')
            for k, v in self.param.Mapping.items()
            if v['table'] == 'sample' and v['field'][0] in (
                't', 'n') and v.get('title') != None
        })
        Prj.mappingacq = EncodeEqualList({
            v['field']: v.get('title')
            for k, v in self.param.Mapping.items()
            if v['table'] == 'acq' and v['field'][0] in (
                't', 'n') and v.get('title') != None
        })
        Prj.mappingprocess = EncodeEqualList({
            v['field']: v.get('title')
            for k, v in self.param.Mapping.items()
            if v['table'] == 'process' and v['field'][0] in (
                't', 'n') and v.get('title') != None
        })
        db.session.commit()
        Ids = {
            "acq": {
                "tbl": "acquisitions",
                "pk": "acquisid"
            },
            "sample": {
                "tbl": "samples",
                "pk": "sampleid"
            },
            "process": {
                "tbl": "process",
                "pk": "processid"
            }
        }
        #recupération des orig_id des acq,sample,process
        for i in Ids:
            sql = "select orig_id," + Ids[i]['pk'] + " from " + Ids[i][
                'tbl'] + " where projid=" + str(self.param.ProjectId)
            Ids[i]["ID"] = {}
            for r in GetAll(sql):
                Ids[i]["ID"][r[0]] = int(r[1])
        # recuperation des ID des objet du projet
        self.ExistingObject = {}
        self.pgcur.execute(
            "SELECT o.orig_id,o.objid from objects o where o.projid=" +
            str(self.param.ProjectId))
        for rec in self.pgcur:
            self.ExistingObject[rec[0]] = rec[1]
        #logging.info("Ids = %s",Ids)
        random.seed()
        RealTableName = {
            "acq": 'acquisitions',
            "sample": 'samples',
            "process": 'process',
            "obj_head": "obj_head",
            "obj_field": "obj_field"
        }
        sd = Path(self.param.SourceDir)
        TotalRowCount = 0
        for filter in ("**/ecotaxa*.txt", "**/ecotaxa*.tsv"):
            for CsvFile in sd.glob(filter):
                relname = CsvFile.relative_to(
                    sd)  # Nom relatif à des fins d'affichage uniquement
                logging.info("Analyzing file %s" % (relname.as_posix()))
                with open(CsvFile.as_posix(), encoding='latin_1') as csvfile:
                    # lecture en mode dictionnaire basé sur la premiere ligne
                    rdr = csv.DictReader(csvfile,
                                         delimiter='\t',
                                         quotechar='"')
                    #lecture la la ligne des types (2nd ligne du fichier
                    LType = {
                        champ.strip(" \t").lower(): v
                        for champ, v in rdr.__next__().items()
                    }
                    ListeChamps = [
                        champ.strip(" \t").lower() for champ in rdr.fieldnames
                    ]
                    # Chargement du contenu du fichier
                    RowCount = 0
                    for rawlig in rdr:
                        lig = {
                            champ.strip(" \t").lower(): v
                            for champ, v in rawlig.items()
                        }
                        Objs = {
                            "acq": self.EmptyObject(),
                            "sample": self.EmptyObject(),
                            "process": self.EmptyObject(),
                            "obj_head": self.EmptyObject(),
                            "obj_field": self.EmptyObject()
                        }
                        ObjsNew = {
                            "acq": database.Acquisitions(),
                            "sample": database.Samples(),
                            "process": database.Process()
                        }
                        RowCount += 1
                        TotalRowCount += 1
                        if 'object_annotation_category_id' in ListeChamps and 'object_annotation_category' in ListeChamps:
                            if CleanValue(
                                    lig.get('object_annotation_category_id',
                                            '')) != '':
                                del lig[
                                    'object_annotation_category']  # s'il y a un ID on ignore le texte
                        for champ in ListeChamps:
                            ColName = champ.strip(" \t").lower()
                            m = self.param.Mapping.get(ColName, None)
                            if m is None:
                                continue  # Le champ n'est pas considéré
                            FieldName = m.get("field", None)
                            FieldTable = m.get("table", None)
                            FieldValue = None
                            v = CleanValue(lig.get(champ))
                            if v != "":  # si pas de valeurs, on laisse le champ null
                                if m['type'] == 'n':
                                    if champ == 'object_annotation_category_id':
                                        if self.param.updateclassif != "Y":
                                            continue
                                    FieldValue = ToFloat(v)
                                elif champ == 'object_date':
                                    FieldValue = datetime.date(
                                        int(v[0:4]), int(v[4:6]), int(v[6:8]))
                                elif champ == 'object_time':
                                    v = v.zfill(6)
                                    FieldValue = datetime.time(
                                        int(v[0:2]), int(v[2:4]), int(v[4:6]))
                                elif FieldName == 'classif_when':
                                    if self.param.updateclassif != "Y":
                                        continue
                                    v2 = CleanValue(
                                        lig.get('object_annotation_time',
                                                '000000')).zfill(6)
                                    FieldValue = datetime.datetime(
                                        int(v[0:4]), int(v[4:6]), int(v[6:8]),
                                        int(v2[0:2]), int(v2[2:4]),
                                        int(v2[4:6]))
                                elif FieldName == 'classif_id':
                                    if self.param.updateclassif != "Y":
                                        continue
                                    FieldValue = self.param.TaxoFound[ntcv(
                                        v
                                    ).lower(
                                    )]  # pour la version numerique, c'est traité par if type=n
                                elif FieldName == 'classif_who':
                                    if self.param.updateclassif != "Y":
                                        continue
                                    FieldValue = self.param.UserFound[ntcv(
                                        v).lower()].get('id', None)
                                elif FieldName == 'classif_qual':
                                    if self.param.updateclassif != "Y":
                                        continue
                                    FieldValue = database.ClassifQualRevert.get(
                                        v.lower())
                                else:  # c'est un champ texte sans rien de special
                                    FieldValue = v
                                # if FieldTable in ('image', 'sample'):
                                if FieldTable in ('image', ):
                                    pass  # les champs images sont ignorés lors d'un update
                                elif FieldTable in Objs:
                                    if FieldName in db.metadata.tables[
                                            RealTableName[FieldTable]].columns:
                                        setattr(Objs[FieldTable], FieldName,
                                                FieldValue)
                                        # logging.info("setattr %s %s %s",FieldTable,FieldName,FieldValue)
                                    # else:
                                    # logging.info("skip F %s %s %s",FieldTable,FieldName,FieldValue)
                                else:
                                    logging.info("skip T %s %s %s", FieldTable,
                                                 FieldName, FieldValue)
                        # Calcul de la position du soleil
                        if (hasattr(Objs["obj_head"], 'objdate')
                                and hasattr(Objs["obj_head"], 'objtime')
                                and hasattr(Objs["obj_head"], 'longitude')
                                and hasattr(Objs["obj_head"], 'latitude')):
                            if not (AstralCache['date']==Objs["obj_head"].objdate and AstralCache['time']==Objs["obj_head"].objtime \
                                and AstralCache['long']==Objs["obj_head"].longitude and AstralCache['lat']==Objs["obj_head"].latitude) :
                                AstralCache = {
                                    'date': Objs["obj_head"].objdate,
                                    'time': Objs["obj_head"].objtime,
                                    'long': Objs["obj_head"].longitude,
                                    'lat': Objs["obj_head"].latitude,
                                    'r': ''
                                }
                                from astral import AstralError
                                try:
                                    AstralCache['r'] = appli.CalcAstralDayTime(
                                        AstralCache['date'],
                                        AstralCache['time'],
                                        AstralCache['lat'],
                                        AstralCache['long'])
                                except AstralError as e:  # dans certains endoit du globe il n'y a jamais de changement nuit/jour certains jours, ca provoque une erreur
                                    app.logger.error(
                                        "Astral error : %s for %s", e,
                                        AstralCache)
                                except Exception as e:  # autre erreurs par exemple si l'heure n'est pas valide;
                                    app.logger.error(
                                        "Astral error : %s for %s", e,
                                        AstralCache)
                            Objs["obj_head"].sunpos = AstralCache['r']
                        # Affectation des ID Sample, Acq & Process et creation de ces dernier si necessaire
                        for t in Ids:
                            if hasattr(Objs[t], 'orig_id') and Objs[
                                    t].orig_id is not None:  # si des colonnes sont definis sur les tables auxilaires
                                if Objs[t].orig_id in Ids[t][
                                        "ID"]:  # si on référence un auxiliaire existant
                                    if t == 'sample':  # on ramene l'enregistrement
                                        ObjAux = database.Samples.query.filter_by(
                                            sampleid=Ids[t]["ID"][
                                                Objs[t].orig_id]).first()
                                    elif t == 'acq':
                                        ObjAux = database.Acquisitions.query.filter_by(
                                            acquisid=Ids[t]["ID"][
                                                Objs[t].orig_id]).first()
                                    elif t == 'process':
                                        ObjAux = database.Process.query.filter_by(
                                            processid=Ids[t]["ID"][
                                                Objs[t].orig_id]).first()
                                    else:
                                        ObjAux = None
                                    if ObjAux is not None:  # si on a bien trouvé l'enregistrement
                                        for attr, value in Objs[t].__dict__.items(
                                        ):  # Pour les champs présents dans l'objet temporaire (dans le TSV donc)
                                            if hasattr(
                                                    ObjAux, attr
                                            ) and getattr(
                                                    ObjAux, attr
                                            ) != value:  # si la valeur change
                                                setattr(
                                                    ObjAux, attr, value
                                                )  # On met à jour l'enregistrement (qui sera commité avec l'objet plus bas)

                                    setattr(Objs["obj_head"], Ids[t]["pk"],
                                            Ids[t]["ID"][Objs[t].orig_id])
                                else:
                                    # on copie les données de l'objet dans le nouvel enregistrement
                                    for attr, value in Objs[t].__dict__.items(
                                    ):
                                        if hasattr(ObjsNew[t], attr):
                                            setattr(ObjsNew[t], attr, value)
                                    Objs[t] = ObjsNew[t]
                                    Objs[t].projid = self.param.ProjectId
                                    db.session.add(Objs[t])
                                    db.session.commit()
                                    Ids[t]["ID"][Objs[t].orig_id] = getattr(
                                        Objs[t], Ids[t]["pk"])
                                    setattr(Objs["obj_head"], Ids[t]["pk"],
                                            Ids[t]["ID"][Objs[t].orig_id])
                                    logging.info("IDS %s %s", t, Ids[t])
                        # Recherche de l'objet si c'est une images complementaire
                        if Objs["obj_field"].orig_id in self.ExistingObject:
                            # Traitement des champs
                            ObjField = database.ObjectsFields.query.filter_by(
                                objfid=self.ExistingObject[
                                    Objs["obj_field"].orig_id]).first()
                            for attr, value in Objs[
                                    "obj_field"].__dict__.items():
                                if hasattr(ObjField, attr):
                                    setattr(ObjField, attr, value)
                            #traitement du header
                            ObjHead = database.Objects.query.filter_by(
                                objid=self.ExistingObject[
                                    Objs["obj_field"].orig_id]).first()
                            for attr, value in Objs["obj_head"].__dict__.items(
                            ):
                                if hasattr(ObjHead, attr):
                                    setattr(ObjHead, attr, value)
                            db.session.commit()

                        else:  # ou Creation de l'objet
                            logging.warning(
                                "Object '%s' not found, can't be updated, row skipped",
                                Objs["obj_field"].orig_id)
                            continue
                            # Objs["obj_head"].projid=self.param.ProjectId
                            # Objs["obj_head"].random_value=random.randint(1,99999999)
                            # Objs["obj_head"].img0id=Objs["image"].imgid
                            # db.session.add(Objs["obj_head"])
                            # db.session.commit()
                            # Objs["obj_field"].objfid=Objs["obj_head"].objid
                            # db.session.add(Objs["obj_field"])
                            # db.session.commit()
                            # self.ExistingObject[Objs["obj_field"].orig_id]=Objs["obj_head"].objid # Provoque un select object sauf si 'expire_on_commit':False
                        if (TotalRowCount % 100) == 0:
                            self.UpdateProgress(
                                100 * TotalRowCount / self.param.TotalRowCount,
                                "Processing files %d/%d" %
                                (TotalRowCount, self.param.TotalRowCount))
                    logging.info("File %s : %d row Processed",
                                 relname.as_posix(), RowCount)
                    db.session.commit()
        self.pgcur.execute("""update obj_head o
                            set imgcount=(select count(*) from images where objid=o.objid)
                            ,img0id=(select imgid from images where objid=o.objid order by imgrank asc limit 1 )
                            where projid=""" + str(self.param.ProjectId))
        self.pgcur.connection.commit()
        self.pgcur.execute(
            """update samples s set latitude=sll.latitude,longitude=sll.longitude
              from (select o.sampleid,min(o.latitude) latitude,min(o.longitude) longitude
              from obj_head o
              where projid=%(projid)s and o.latitude is not null and o.longitude is not null
              group by o.sampleid) sll where s.sampleid=sll.sampleid and projid=%(projid)s """,
            {'projid': self.param.ProjectId})
        self.pgcur.connection.commit()
        appli.project.main.RecalcProjectTaxoStat(Prj.projid)
        appli.project.main.UpdateProjectStat(Prj.projid)
        self.task.taskstate = "Done"
        self.UpdateProgress(100, "Processing done")
コード例 #6
0
def ExploreLoadRightPane():
    # récupération des parametres d'affichage
    Filt = {}
    for k, v in FilterList.items():
        Filt[k] = gvp(k, v)
    ipp = int(Filt["ipp"])
    zoom = int(Filt["zoom"])
    t = ["<a name='toppage'/>"]
    sqlparam = {'projid': gvp("projid")}
    #     sql="""select o.objid,t.name taxoname,o.classif_qual,u.name classifwhoname,i.file_name
    #   ,i.height,i.width,i.thumb_file_name,i.thumb_height,i.thumb_width
    #   ,o.depth_min,o.depth_max,s.orig_id samplename,o.objdate,to_char(o.objtime,'HH24:MI') objtime
    #   ,o.latitude,o.orig_id,o.imgcount
    #    from objects o
    # left Join images i on o.img0id=i.imgid
    # left JOIN taxonomy t on o.classif_id=t.id
    # LEFT JOIN users u on o.classif_who=u.id
    # LEFT JOIN  samples s on o.sampleid=s.sampleid
    # where o.classif_qual='V'
    # """
    whereclause = ""
    sql = """select o.objid,o.classif_qual  ,o.objdate,to_char(o.objtime,'HH24:MI') objtime
  ,o.imgcount,o.img0id,o.classif_id,o.classif_who,o.sampleid,random_value,o.projid
   from obj_head o
where o.classif_qual='V'
"""
    if gvp("taxo[]"):
        taxoids = ",".join(
            (str(int(x)) for x in request.form.getlist("taxo[]")))
        if gvp("taxochild") == "1":
            # whereclause+=""" and o.classif_id in (WITH RECURSIVE rq(id) as ( select id FROM taxonomy where id in(%s)
            #                         union
            #                         SELECT t.id FROM rq JOIN taxonomy t ON rq.id = t.parent_id and (t.nbrobjcum>0 or t.nbrobj>0)
            #                         ) select id from rq)"""%(taxoids,)
            # Sur les petits nombres de ref le in est plus performant que la sous requete
            taxoids = ",".join((str(int(x[0])) for x in GetAll(
                """WITH RECURSIVE rq(id) as ( select id FROM taxonomy where id in(%s)
                                    union
                                    SELECT t.id FROM rq JOIN taxonomy t ON rq.id = t.parent_id 
                                    ) select id from rq """ % (taxoids, ))))
            whereclause += " and o.classif_id in (" + taxoids + ")"  # optimisation qui provoque de faux résultats : and (t.nbrobjcum>0 or t.nbrobj>0)
        else:
            whereclause += " and o.classif_id in (" + taxoids + ")"

    if gvp("MapN") != "" and gvp("MapW") != "" and gvp("MapE") != "" and gvp(
            "MapS") != "":
        whereclause += " and o.latitude between %(MapS)s and %(MapN)s and o.longitude between %(MapW)s and %(MapE)s  "
        sqlparam['MapN'] = gvp("MapN")
        sqlparam['MapW'] = gvp("MapW")
        sqlparam['MapE'] = gvp("MapE")
        sqlparam['MapS'] = gvp("MapS")

    if gvp("depthmin") != "" and gvp("depthmax") != "":
        whereclause += " and o.depth_min between %(depthmin)s and %(depthmax)s and o.depth_max between %(depthmin)s and %(depthmax)s  "
        sqlparam['depthmin'] = gvp("depthmin")
        sqlparam['depthmax'] = gvp("depthmax")

    if gvp("samples") != "":
        whereclause += " and o.sampleid= any (%(samples)s) "
        sqlparam['samples'] = [int(x) for x in gvp("samples").split(',')]

    if gvp("instrum") != "":
        whereclause += " and o.acquisid in (select acquisid  from acquisitions  where instrument ilike %(instrum)s " + (
            "and projid= any (%(projid)s)"
            if gvp("projid") != "" else "") + " )"
        sqlparam['instrum'] = '%' + gvp("instrum") + '%'
    PageTopProjectLink = MapForProject = None
    if gvp("projid") != "":
        whereclause += " and o.projid= any (%(projid)s) "
        sqlparam['projid'] = [int(x) for x in gvp("projid").split(',')]
        if len(sqlparam['projid']) == 1:
            PageTopProjectLink = "<p class='bg-info' style='font-size: larger;font-weight: bold;'>You can explore this project in more details on its <a href='/prj/{0}'>dedicated page</a></p>".format(
                str(sqlparam['projid'][0]))
    if gvp("fromdate") != "":
        whereclause += " and o.objdate>= to_date(%(fromdate)s,'YYYY-MM-DD') "
        sqlparam['fromdate'] = gvp("fromdate")
    if gvp("todate") != "":
        whereclause += " and o.objdate<= to_date(%(todate)s,'YYYY-MM-DD') "
        sqlparam['todate'] = gvp("todate")
    if gvp("month") != "":
        whereclause += " and extract(month from o.objdate) = any (%(month)s) "
        sqlparam['month'] = [int(x) for x in gvp("month").split(',')]
    if gvp("daytime") != "":
        whereclause += " and o.sunpos= any (%(daytime)s) "
        sqlparam['daytime'] = [x for x in gvp("daytime").split(',')]

    if gvp("inverttime") == "1":
        if gvp("fromtime") != "" and gvp("totime") != "":
            whereclause += " and (o.objtime<= time %(fromtime)s or o.objtime>= time %(totime)s)"
            sqlparam['fromtime'] = gvp("fromtime")
            sqlparam['totime'] = gvp("totime")
    else:
        if gvp("fromtime") != "":
            whereclause += " and o.objtime>= time %(fromtime)s "
            sqlparam['fromtime'] = gvp("fromtime")
        if gvp("totime") != "":
            whereclause += " and objtime<= time %(totime)s "
            sqlparam['totime'] = gvp("totime")
    sql += whereclause
    ExtraEndScript = ""
    if whereclause == "":  # si aucune clause, on prend un projet au hasard
        randomproject = GetAll(
            "select projid,title from projects where visible=true and pctvalidated>1 order by random() limit 1"
        )
        if randomproject:
            randomproject = randomproject[0]
            MapForProject = str(randomproject[0])
            sql += " and o.projid= %s " % (randomproject[0])
            ExtraEndScript = """$('#headersubtitle').html('Randomly selected project : <a href="?projid={0}">{1}</a>');""".format(
                randomproject['projid'], XSSEscape(randomproject['title']))
    sql += "  order by random_value Limit %d" % (2 * ipp, )
    # pour de meilleure perf plus de random ici et du coup on prend 20xipp pour créer un peu d'aléa
    # sql+="  Limit %d"%(20*ipp,) # desactivé suite à split table objects mais pourrait devoir revenir.

    #filt_fromdate,#filt_todate
    sql = """select o.*,t.display_name taxoname,u.name classifwhoname,i.file_name,s.orig_id samplename
                  ,i.height,i.width,i.thumb_file_name,i.thumb_height,i.thumb_width,ofi.orig_id
                  from (""" + sql + """)o
left Join images i on o.img0id=i.imgid
left JOIN taxonomy t on o.classif_id=t.id
LEFT JOIN users u on o.classif_who=u.id
LEFT JOIN  samples s on o.sampleid=s.sampleid
left Join obj_field ofi on ofi.objfid=o.objid
where o.projid in (select projid from projects where visible=true)"""
    # if whereclause!="": # on ne tri pas en random global s'il n'y a aucune criteres, impact de perf
    sql += " order by random_value  "
    # sql+=" order by random()  "
    sql += " Limit %d  " % (ipp, )
    res = GetAll(sql, sqlparam, debug=False, doXSSEscape=True)
    trcount = 1
    LineStart = ""
    if (PageTopProjectLink):
        t.append(PageTopProjectLink)
    if MapForProject:
        t.append(
            render_template("search/explore_inserted_popup.html",
                            Projid=MapForProject))
    t.append("<table class=imgtab><tr id=tr1>" + LineStart)
    WidthOnRow = 0
    #récuperation et ajustement des dimensions de la zone d'affichage
    try:
        PageWidth = int(
            gvp("resultwidth"
                )) - 40  # on laisse un peu de marge à droite et la scroolbar
        if PageWidth < 200: PageWidth = 200
    except:
        PageWidth = 200
    try:
        WindowHeight = int(
            gvp("windowheight")) - 100  # on enleve le bandeau du haut
        if WindowHeight < 200: WindowHeight = 200
    except:
        WindowHeight = 200
    #print("PageWidth=%s, WindowHeight=%s"%(PageWidth,WindowHeight))
    # Calcul des dimmensions et affichage des images
    for r in res:
        filename = r['file_name']
        origwidth = r['width']
        origheight = r['height']
        thumbfilename = r['thumb_file_name']
        thumbwidth = r['thumb_width']
        if origwidth is None:  # pas d'image associé, pas trés normal mais arrive pour les subset sans images
            width = 80
            height = 40
        else:
            width = origwidth * zoom // 100
            height = origheight * zoom // 100
        if max(
                width, height
        ) < 120:  # en dessous de 120 px de coté on ne fait plus le scaling
            if max(origwidth, origheight) < 120:
                width = origwidth  # si l'image originale est petite on l'affiche telle quelle
                height = origheight
            elif max(origwidth, origheight) == origwidth:
                width = 120
                height = origheight * 120 // origwidth
                if height < 1: height = 1
            else:
                height = 120
                width = origwidth * 120 // origheight
                if width < 1: width = 1

        # On limite les images pour qu'elles tiennent toujours dans l'écran
        if width > PageWidth:
            width = PageWidth
            height = math.trunc(r['height'] * width / r['width'])
            if height == 0: height = 1
        if height > WindowHeight:
            height = WindowHeight
            width = math.trunc(r['width'] * height / r['height'])
            if width == 0: width = 1
        if WidthOnRow != 0 and (WidthOnRow + width) > PageWidth:
            trcount += 1
            t.append("</tr></table><table class=imgtab><tr id=tr%d>%s" %
                     (trcount, LineStart))
            WidthOnRow = 0
        cellwidth = width + 22
        if cellwidth < 80:
            cellwidth = 80  # on considère au moins 80 car avec les label c'est rarement moins
        # Met la fenetre de zoon la ou il y plus de place, sachant qu'elle fait 400px et ne peut donc pas être callée à gauche des premieres images.
        if (WidthOnRow + cellwidth) > (PageWidth / 2):
            pos = 'left'
        else:
            pos = 'right'
        #Si l'image affiché est plus petite que la miniature, afficher la miniature.
        if thumbwidth is None or thumbwidth < width or thumbfilename is None:  # sinon (si la miniature est plus petite que l'image à afficher )
            thumbfilename = filename  # la miniature est l'image elle même
        txt = "<td width={0}>".format(cellwidth)
        if filename:
            txt+="<img class='lazy' id=I{3} data-src='/vault/{5}' data-zoom-image='{0}' width={1} height={2} pos={4}>"\
                .format(filename,width,height,r['objid'],pos,thumbfilename)
        else:
            txt += "No Image"
        # Génération de la popover qui apparait pour donner quelques détails sur l'image
        poptitletxt = "%s" % (r['orig_id'], )
        # poptxt="%s"%(r['taxoname'],)
        poptxt = ""
        if r['classifwhoname'] != "":
            poptxt += "<em>by</em> %s" % (r['classifwhoname'])
        poptxt += "<br><em>in</em> " + ntcv(r['samplename'])
        popattribute = "data-title=\"{0}\" data-content=\"{1}\" data-placement='{2}'".format(
            poptitletxt, poptxt, 'left' if WidthOnRow > 500 else 'right')

        txt+="""<div class='subimg {1}' {2}>
<div class='taxo'>{0}</div> </div>
<div class=ddet><span class=ddets><span class='glyphicon glyphicon-eye-open'></span> {3}</div>"""\
            .format(r['taxoname'],"",popattribute
                    ,"(%d)"%(r['imgcount'],) if r['imgcount'] is not None and r['imgcount']>1 else "")
        txt += "</td>"

        # WidthOnRow+=max(cellwidth,80) # on ajoute au moins 80 car avec les label c'est rarement moins
        WidthOnRow += cellwidth + 5  # 5 = border-spacing = espace inter image
        t.append(txt)

    t.append("</tr></table>")
    if len(res) == 0:
        t.append(
            """<div class='alert alert-warning' role='alert' style='margin: 10px;font-size:larger;' >
        <span class='glyphicon glyphicon-remove-sign' style='color:#DD2222;'></span>
        Your selection does not return any object. <br>It may be too limited by the selected filters. <br>Try again releasing the selection criteria</div>"""
        )
    t.append("""
    <script>
        PostAddImages();
        %s
    </script>""" % ExtraEndScript)
    return "\n".join(t)
コード例 #7
0
def doimportmassupdate():
    # test avec D:\temp\Downloads\taxoexport_20181228_101007.tsv
    txt = ""
    uploadfile = request.files.get("fichier")
    if uploadfile is None:
        return PrintInCharte(FormatError("You must send a file"))
    app.logger.info('Load file {} by {}'.format(uploadfile.filename,
                                                current_user))
    #creation d'un fichier temporaire qui s'efface automatiquement
    tmpfile = tempfile.TemporaryFile(mode='w+b')
    # app.logger.info('TMP file is {}'.format(tmpfile.name))
    uploadfile.save(tmpfile)  # on copie le contenu dedants
    tmpfile.seek(0)  # on se remet au debut
    f = TextIOWrapper(
        tmpfile, encoding='ascii',
        errors='replace')  # conversion du format binaire au format texte
    rdr = csv.DictReader(
        f,
        delimiter='\t',
        quotechar='"',
    )  # ouverture sous forme de reader dictionnaire
    champs = rdr.fieldnames
    app.logger.info("Loading file with this columns : %s" % str(champs))
    if 'id' not in champs:
        return PrintInCharte(FormatError("A column named 'id' is required"))
    IdList = {x['id'] for x in database.GetAll("select id from taxonomy")}
    InstanceList = {
        x['id']
        for x in database.GetAll("select id from ecotaxainst")
    }
    rowcount = 0
    Errors = []
    UpdatedTaxon = []
    UpdatableCols = [
        'parent_id', 'name', 'taxotype', 'taxostatus', 'id_source',
        'source_url', 'source_desc', 'creator_email', 'creation_datetime',
        'id_instance', 'rename_to'
    ]
    # 'lastupdate_datetime', 'display_name',
    for r in rdr:
        rowcount += 1
        id = int(r['id'])
        taxon = database.Taxonomy.query.filter_by(id=id).first()
        if taxon is None:
            Errors.append("id {} does not exists in the database".format(id))
            continue
        valueschanged = False
        SkipRow = False
        for c in champs:
            if c in UpdatableCols:
                oldvalue = str(ntcv(getattr(taxon, c))).replace('\r', '')
                newvalue = r[c].replace("\\n", "\n").strip()
                if c in ('parent_id', 'rename_to') and newvalue != '':
                    if int(newvalue) not in IdList:
                        Errors.append(
                            "id {} : {} {} does not exists in the database".
                            format(id, c, newvalue))
                        SkipRow = True
                        continue
                if c == 'taxotype':
                    if newvalue not in ('P', 'M'):
                        Errors.append("id {} : Invalid taxotype {} ".format(
                            id, newvalue))
                        SkipRow = True
                        continue
                if c == 'taxostatus':
                    if newvalue not in database.TaxoStatus:
                        Errors.append("id {} : Invalid status {} ".format(
                            id, newvalue))
                        SkipRow = True
                        continue
                if c == 'id_instance' and newvalue != '':
                    if int(newvalue) not in InstanceList:
                        Errors.append(
                            "id {} : {} is not a valid instance id".format(
                                id, newvalue))
                        SkipRow = True
                        continue
                if oldvalue != newvalue:
                    valueschanged = True
                    setattr(taxon, c, newvalue)
                    app.logger.info("id {} : update {} to {}".format(
                        id, oldvalue, newvalue))
        if SkipRow:
            continue
        if valueschanged:
            # db.session.add(taxon)
            UpdatedTaxon.append(id)
            taxon.lastupdate_datetime = datetime.datetime.utcnow()
            db.session.commit()
        # app.logger.info(str(r))
    if len(UpdatedTaxon) > 0:
        ComputeDisplayName(UpdatedTaxon)
    txt += "<p style='color: green'> %s taxon updated </p> " % (
        len(UpdatedTaxon))
    if len(Errors):
        txt += "<p style='color: red'> %s errors <ul> " % (len(Errors))
        txt += "\n".join("<li>%s</li>" % x for x in Errors)
        txt += "</ul></p> "
    txt += "<a href='/browsetaxo/' class='btn btn-primary'><i class='fas fa-arrow-left'></i> Back to Browse Taxonomy</a>"
    g.bodydivmargin = "10px"
    return PrintInCharte(txt)
コード例 #8
0
ファイル: services.py プロジェクト: obs6/ecotaxoserver
def ComputeDisplayName(TaxoList):
    """
    Compute display_name column in database, for the list of provided id
    :param TaxoList: 
    :return: None 
    """
    #     sql="""with duplicate as (select name from taxonomy GROUP BY name HAVING count(*)>1),
    #               nt as (select t.id,case when t.name like '%% %%' or p.id is null then t.name
    #                     when t.taxotype='M' and P.taxotype='M' then concat(p2.name||'>',p.name||'>',t.name)
    #                     when t.name in (select name from duplicate) then concat(p.name||'>',t.name)
    #                     else t.name end
    #                     ||case when t.taxostatus='D' then ' (Deprecated)' else '' end
    #                      newname
    #   from taxonomy t
    #   left JOIN taxonomy p on t.parent_id=p.id
    #   left JOIN taxonomy p2 on p.parent_id=p2.id
    #   where (t.id = any ( %(taxo)s ) or p.id = any ( %(taxo)s ) or p2.id = any ( %(taxo)s ))
    # )
    # update public.taxonomy t set display_name=newname,lastupdate_datetime=to_timestamp(%(ts)s,'YYYY-MM-DD HH24:MI:SS')
    # from nt
    # where nt.id=t.id and display_name IS DISTINCT FROM newname """
    #     # Pour chaque nom on cherche à determiner à quelle hauteur il n'y a plus de doublons
    #     # quand plus de 2 doublons ça peut conduire à une inflation car on va prendre le plus long pour tous
    #     # alors que par forcement necessaire ex : A<B , A<C<D , A<C<E  A<B sera A<B<X inutilement rallongé
    #     sql="""
    #     with duplicate as (
    #     select t.name, count(distinct t.id) cid,
    #       count(distinct concat(t.name,'<'||p.name)) c2,
    #       count(distinct concat(t.name,'<'||p.name,'<'||p2.name)) c3,
    #       count(distinct concat(t.name,'<'||p.name,'<'||p2.name,'<'||p3.name)) c4
    #       from taxonomy t
    #       left JOIN taxonomy p on t.parent_id=p.id
    #       left JOIN taxonomy p2 on p.parent_id=p2.id
    #       left JOIN taxonomy p3 on p2.parent_id=p3.id
    #       group by t.name
    #     having count(distinct t.id)>1 )
    #         ,nt as (select t.id,case when d.name is null then t.name
    #                                  when cid=c2 then concat(t.name,'<'||p.name)
    #                                  when cid=c3 then concat(t.name,'<'||p.name,'<'||p2.name)
    #                                  else  concat(t.name,'<'||p.name,'<'||p2.name,'<'||p3.name)
    #                              end newname
    #       from taxonomy t
    #       left JOIN duplicate d on t.name=d.name
    #       left JOIN taxonomy p on t.parent_id=p.id
    #       left JOIN taxonomy p2 on p.parent_id=p2.id
    #       left JOIN taxonomy p3 on p2.parent_id=p3.id
    #       where (t.id = any ( %(taxo)s ) or p.id = any ( %(taxo)s ) or p2.id = any(%(taxo)s) or p3.id = any(%(taxo)s)  )
    #     )
    #     update public.taxonomy t set display_name=newname
    #     from nt
    #       where nt.id=t.id
    #
    #     """
    #     database.ExecSQL(sql,{'taxo':TaxoList,'ts':datetime.datetime.utcnow().strftime('%Y-%m-%d %H:%M:%S')})
    # on recalcule tous les doublons + ceux qui n'ont pas de noms + ceux ayant le même nom que ceux demandés dans leur lineage 3.
    sql = """with duplicate as (select lower(name) as name from taxonomy GROUP BY lower(name) HAVING count(*)>1)
          select t.id,t.name tname,p.name pname,p2.name p2name,p3.name p3name,t.display_name,t.taxostatus
          from taxonomy t
          left JOIN duplicate d on lower(t.name)=d.name
          left JOIN taxonomy p on t.parent_id=p.id
          left JOIN taxonomy p2 on p.parent_id=p2.id
          left JOIN taxonomy p3 on p2.parent_id=p3.id
          where d.name is not null or t.display_name is null 
          or lower(t.name) in (select lower(st.name) 
                                  from taxonomy st
                                  left JOIN taxonomy sp on st.parent_id=sp.id
                                  left JOIN taxonomy sp2 on sp.parent_id=sp2.id
                                  left JOIN taxonomy sp3 on sp2.parent_id=sp3.id
                                where (st.id=any(%(taxo)s) or sp.id=any(%(taxo)s) or sp2.id=any(%(taxo)s) or sp3.id=any(%(taxo)s)  )  
                )
          """
    Duplicates = database.GetAll(sql, {'taxo': TaxoList},
                                 cursor_factory=psycopg2.extras.RealDictCursor)
    starttime = datetime.datetime.now()
    DStats = {}

    def AddToDefStat(clestat):
        clestat = clestat.lower()
        if clestat in DStats:
            DStats[clestat] += 1
        else:
            DStats[clestat] = 1

    for D in Duplicates:
        cle = ntcv(D['tname'])
        AddToDefStat(cle)
        cle += '<' + ntcv(D['pname'])
        AddToDefStat(cle)
        cle += '<' + ntcv(D['p2name'])
        AddToDefStat(cle)
        cle += '<' + ntcv(D['p3name'])
        AddToDefStat(cle)

    for i, D in enumerate(Duplicates):
        cle = ntcv(D['tname'])
        if DStats[cle.lower()] == 1:
            Duplicates[i]['newname'] = cle
        else:
            cle += '<' + ntcv(D['pname'])
            if DStats[cle.lower()] == 1:
                Duplicates[i]['newname'] = cle
            else:
                cle += '<' + ntcv(D['p2name'])
                if DStats[cle.lower()] == 1:
                    Duplicates[i]['newname'] = cle
                else:
                    cle += '<' + ntcv(D['p3name'])
                    Duplicates[i]['newname'] = cle
        if D['taxostatus'] == 'D':
            Duplicates[i]['newname'] += " (Deprecated)"

    app.logger.debug("Compute time %s ",
                     (datetime.datetime.now() - starttime).total_seconds())
    starttime = datetime.datetime.now()
    UpdateParam = []
    for D in Duplicates:
        if D['display_name'] != D['newname']:
            UpdateParam.append((int(D['id']), D['newname']))
    if len(UpdateParam) > 0:
        cur = g.db.cursor()
        psycopg2.extras.execute_values(
            cur,
            """UPDATE taxonomy SET display_name = data.pdisplay_name,lastupdate_datetime=to_timestamp('{}','YYYY-MM-DD HH24:MI:SS') 
               FROM (VALUES %s) AS data (pid, pdisplay_name)
               WHERE id = data.pid""".format(
                datetime.datetime.utcnow().strftime('%Y-%m-%d %H:%M:%S')),
            UpdateParam)
        cur.connection.commit()
        cur.close()

    app.logger.debug("Update time %s for %d rows",
                     (datetime.datetime.now() - starttime).total_seconds(),
                     len(UpdateParam))
コード例 #9
0
ファイル: taxomain.py プロジェクト: obs6/ecotaxa_dev
def DoFullSync():
    txt = ""
    try:
        UpdatableCols = [
            'parent_id', 'name', 'taxotype', 'taxostatus', 'id_source',
            'id_instance', 'rename_to', 'display_name', 'source_desc',
            'source_url', 'creation_datetime', 'creator_email'
        ]
        MaxUpdate = database.GetAll(
            "select coalesce(max(lastupdate_datetime),to_timestamp('2000-01-01','YYYY-MM-DD')) lastupdate from taxonomy"
        )
        MaxUpdateDate = MaxUpdate[0]['lastupdate']

        j = request_withinstanceinfo("/gettaxon/", {
            'filtertype': 'since',
            'startdate': MaxUpdateDate
        })
        if 'msg' in j:
            return appli.ErrorFormat("Sync Error :" + j['msg'])
        NbrRow = len(j)
        NbrUpdate = NbrInsert = 0
        txt += "Received {} rows<br>".format(NbrRow)
        if (NbrRow > 0):
            txt += "Taxo 0 = {}<br>".format(j[0])
        for jtaxon in j:
            taxon = database.Taxonomy.query.filter_by(
                id=int(jtaxon['id'])).first()
            lastupdate_datetime = datetime.datetime.strptime(
                jtaxon['lastupdate_datetime'], '%Y-%m-%d %H:%M:%S')
            if taxon:
                if taxon.lastupdate_datetime == lastupdate_datetime:
                    continue  #already up to date
                NbrUpdate += 1
            else:
                if ntcv(jtaxon['rename_to']) != '':
                    continue  # don't insert taxon that should be renamed
                if ntcv(jtaxon['taxostatus']) == 'D':
                    continue  # don't insert taxon that are deprecated and planned to be deleted
                NbrInsert += 1
                taxon = database.Taxonomy()
                taxon.id = int(jtaxon['id'])
                db.session.add(taxon)

            for c in UpdatableCols:
                setattr(taxon, c, jtaxon[c])
            taxon.lastupdate_datetime = lastupdate_datetime
            db.session.commit()
        # Manage rename_to
        sqlbase = "with taxorename as (select id,rename_to from taxonomy where rename_to is not null) "
        sql = sqlbase + """select distinct projid from obj_head o join taxorename tr  on o.classif_id=tr.id """
        ProjetsToRecalc = database.GetAll(sql)
        sql = sqlbase + """update obj_head o set classif_id=tr.rename_to 
              from taxorename tr  where o.classif_id=tr.id """
        NbrRenamedObjects = ExecSQL(sql)
        sql = sqlbase + """update obj_head o set classif_auto_id=tr.rename_to 
              from taxorename tr  where o.classif_auto_id=tr.id """
        ExecSQL(sql)
        sql = sqlbase + """update objectsclassifhisto o set classif_id=tr.rename_to 
              from taxorename tr  where o.classif_id=tr.id """
        ExecSQL(sql)
        # on efface les taxon qui doivent être renomé car ils l'ont normalement été
        sql = """delete from taxonomy where rename_to is not null """
        ExecSQL(sql)
        sql = """delete from taxonomy t where taxostatus='D' 
                  and not exists(select 1 from projects_taxo_stat where id=t.id) """
        ExecSQL(sql)
        # il faut recalculer projects_taxo_stat et part_histocat,part_histocat_lst pour ceux qui referencaient un
        # taxon renomé et donc disparu
        if NbrRenamedObjects > 0:
            # cron.RefreshTaxoStat() operation trés longue (env 5 minutes en prod, il faut être plus selectif)
            # permet de recalculer projects_taxo_stat
            for Projet in ProjetsToRecalc:
                appli.project.main.RecalcProjectTaxoStat(Projet['projid'])
            #recalcul part_histocat,part_histocat_lst
            appli.part.prj.GlobalTaxoCompute()

        flash(
            "Received {} rows,Insertion : {} Update :{}".format(
                NbrRow, NbrInsert, NbrUpdate), "success")
        if gvp('updatestat') == 'Y':
            msg = DoSyncStatUpdate()
            flash("Taxon statistics update : " + msg,
                  "success" if msg == 'ok' else 'error')

        # txt="<script>location.reload(true);</script>" # non car ça reprovoque le post de l'arrivée initiale
        txt = "<script>window.location=window.location;</script>"
    except:
        msg = "Error while syncing {}".format(sys.exc_info())
        app.logger.error(msg)
        txt += appli.ErrorFormat(msg)

    return txt
コード例 #10
0
ファイル: taskimportdb.py プロジェクト: obs6/ecotaxa_dev
    def QuestionProcess(self):
        if not (current_user.has_role(database.AdministratorLabel)
                or current_user.has_role(database.ProjectCreatorLabel)):
            return PrintInCharte(
                "ACCESS DENIED for this feature, Admin or privilege creation Required"
            )
        ServerRoot = Path(app.config['SERVERLOADAREA'])
        txt = "<h1>Database Importation Task</h1>"
        errors = []
        if gvg("restart") == 'Y':  # force redemarrage
            self.task.taskstep = 1
            self.UpdateParam()
        if self.task.taskstep == 0:  # ################## Question Creation
            txt += "<h3>Task Creation</h3>"

            if gvp('starttask') == "Y":
                FileToSave = None
                FileToSaveFileName = None
                # Verifier la coherence des données
                uploadfile = request.files.get("uploadfile")
                if uploadfile is not None and uploadfile.filename != '':  # import d'un fichier par HTTP
                    FileToSave = uploadfile  # La copie est faite plus tard, car à ce moment là, le repertoire de la tache n'est pas encore créé
                    FileToSaveFileName = "uploaded.zip"
                    self.param.InData = "uploaded.zip"
                elif len(gvp("ServerPath")) < 2:
                    errors.append("Input Folder/File Too Short")
                else:
                    sp = ServerRoot.joinpath(Path(gvp("ServerPath")))
                    if not sp.exists():  #verifie que le repertoire existe
                        errors.append("Input Folder/File Invalid")
                    else:
                        self.param.InData = sp.as_posix()
                if len(errors) > 0:
                    for e in errors:
                        flash(e, "error")
                else:
                    return self.StartTask(
                        self.param,
                        FileToSave=FileToSave,
                        FileToSaveFileName=FileToSaveFileName)
            return render_template('task/importdb_create.html',
                                   header=txt,
                                   data=self.param,
                                   ServerPath=gvp("ServerPath"))

        newschema = self.GetWorkingSchema()
        # self.task.taskstep=1
        if self.task.taskstep == 1:  # ################## Question Post Import DB
            if gvg("src") == "":  # il faut choisir le projet source
                txt += "<h3>Select project to import</h3>"
                PrjList = GetAll(
                    "select projid,title,status from {0}.projects order by lower(title)"
                    .format(newschema),
                    cursor_factory=None)
                txt += """<table class='table table-condensed table-bordered' style='width:600px;'>"""
                for r in PrjList:
                    txt += "<tr><td><a class='btn btn-primary' href='?src={0}'>Select {0}</a></td><td>{1}</td><td>{2}</td></tr>".format(
                        *r)
                txt += """</table>"""
                return PrintInCharte(txt)
            if gvg("dest") == "":  # il faut choisir le projet destination
                if current_user.has_role(database.AdministratorLabel):
                    PrjList = GetAll(
                        "select projid,title,status from projects order by lower(title)",
                        cursor_factory=None)
                else:
                    PrjList = GetAll("""select p.projid,title,status
                                          from projects p
                                          join projectspriv pp on p.projid = pp.projid
                                          where member=%s and privilege='Manage'
                                          order by lower(title)""",
                                     [current_user.id],
                                     cursor_factory=None)
                txt += "<h3>Select project destination project</h3> or <a class='btn btn-primary' href='?src={0}&dest=new'>New project</a>".format(
                    gvg("src"))
                txt += """<table class='table table-condensed table-bordered' style='width:600px;'>"""
                for r in PrjList:
                    txt += "<tr><td><a class='btn btn-primary' href='?src={0}&dest={1}'>Select {1}</a></td><td>{2}</td><td>{3}</td></tr>".format(
                        gvg("src"), *r)
                txt += """</table>"""
                return PrintInCharte(txt)

            if gvg(
                    "prjok"
            ) == "":  # Creation du projet destination ou MAJ des attributs & detection des Taxo&Users Founds
                SrcPrj = GetAll(
                    "select * from {0}.projects where projid={1}".format(
                        newschema, gvg("src")),
                    cursor_factory=RealDictCursor)
                if gvg("dest") == "new":  # Creation du projet destination
                    Prj = database.Projects()
                    Prj.title = "IMPORT " + SrcPrj[0]['title']
                    db.session.add(Prj)
                    db.session.commit()
                    PrjPriv = database.ProjectsPriv()
                    PrjPriv.projid = Prj.projid
                    PrjPriv.member = current_user.id
                    PrjPriv.privilege = "Manage"
                    db.session.add(PrjPriv)

                else:  # MAJ du projet
                    Prj = database.Projects.query.filter_by(
                        projid=int(gvg("dest"))).first()
                    NbrObj = GetAll(
                        "select count(*) from obj_head WHERE projid=" +
                        gvg("dest"))[0][0]
                    if NbrObj > 0:
                        flash("Destination project must be empty", 'error')
                        return PrintInCharte(
                            "<a href='#' onclick='history.back();'>Back</a>")
                for k, v in SrcPrj[0].items():
                    if k not in ("projid", "title"):
                        setattr(Prj, k, v)
                db.session.commit()
                flash(
                    "Project %s:%s created or updated successfuly" %
                    (Prj.projid, Prj.title), 'success')
                # Controle du mapping Taxo
                sql = """select DISTINCT t.id,lower(t.name) as name,t.parent_id,lower(t.name)||' ('||coalesce(lower(t2.name),'No Parent')||')' as namefull
                          from {0}.obj_head o join {0}.taxonomy t on o.classif_id=t.id
                          left join {0}.taxonomy t2 on t.parent_id=t2.id
                        where o.projid={1} and t.newid is null
                        union select DISTINCT t.id,lower(t.name) as name,t.parent_id,lower(t.name)||' ('||coalesce(lower(t2.name),'No Parent')||')' as namefull
                          from {0}.obj_head o
                          join {0}.taxonomy t on o.classif_auto_id=t.id
                          left join {0}.taxonomy t2 on t.parent_id=t2.id
                          left join taxonomy tt on t.newid =tt.id
                        where o.projid={1}  and tt.id is null
                        order by 2""".format(newschema, gvg("src"))
                self.param.TaxoFound = GetAssoc(sql,
                                                cursor_factory=RealDictCursor,
                                                keyid='id',
                                                debug=False)
                app.logger.info("TaxoFound=%s", self.param.TaxoFound)
                TaxoInDest = GetAssoc2Col(
                    """select t.id,lower(t.name)||' ('||coalesce(lower(t2.name),'No Parent')||')' as name
                                            from taxonomy t
                                            left join taxonomy t2 on t.parent_id=t2.id
                                            where t.id = any (%s)""",
                    (list(self.param.TaxoFound.keys()), ))
                # print(TaxoInDest)
                for id, v in self.param.TaxoFound.items():
                    self.param.TaxoFound[id][
                        'newid'] = None  # par defaut pas de correspondance
                    if id in TaxoInDest:
                        if TaxoInDest[id].lower() == v['namefull']:
                            self.param.TaxoFound[id][
                                'newid'] = id  # ID inchangé
                lst = [
                    t["name"] for t in self.param.TaxoFound.values()
                    if t["newid"] is None
                ]  # liste des Taxon sans mapping
                TaxoInDest = GetAssoc2Col(
                    """select lower(t.name)||' ('||coalesce(lower(t2.name),'No Parent')||')' as name,t.id
                                            from taxonomy t
                                            left join taxonomy t2 on t.parent_id=t2.id
                                            where lower(t.name) = any (%s)""",
                    (lst, ))
                for id, v in self.param.TaxoFound.items():
                    if v['newid'] is None:
                        if v["namefull"] in TaxoInDest:
                            self.param.TaxoFound[id]['newid'] = TaxoInDest[v[
                                "namefull"]]  # ID du même nom dans la base de destination
                NotFoundTaxo = [
                    t["name"] for t in self.param.TaxoFound.values()
                    if t["newid"] is None
                ]  # liste des Taxon sans mapping restant
                app.logger.info("NotFoundTaxo=%s", NotFoundTaxo)
                # Controle du mapping utilisateur
                sql = """select DISTINCT lower(t.name) as name,lower(email) as email,t.password,t.organisation
                          from {0}.obj_head o join {0}.users t on o.classif_who=t.id
                        where o.projid={1}  and t.newid is null
                        union
                        select DISTINCT lower(t.name) as name,lower(email) as email,t.password,t.organisation
                          from {0}.projectspriv pp join {0}.users t on pp.member=t.id
                        where pp.projid={1}  and t.newid is null
                        """.format(newschema, gvg("src"))
                self.param.UserFound = GetAssoc(sql,
                                                cursor_factory=RealDictCursor,
                                                keyid='name')
                self.pgcur = db.engine.raw_connection().cursor()
                self.pgcur.execute(
                    "select id,lower(name),lower(email) from users where lower(name) = any(%s) or email= any(%s) ",
                    (list(self.param.UserFound.keys()),
                     [x.get('email') for x in self.param.UserFound.values()]))
                # Résolution des noms à partir du nom ou de l'email
                for rec in self.pgcur:
                    for u in self.param.UserFound:
                        if u == rec[1] or self.param.UserFound[u].get(
                                'email') == rec[2]:
                            self.param.UserFound[u]['id'] = rec[0]
                logging.info("Users Found = %s", self.param.UserFound)
                NotFoundUser = [
                    k for k, v in self.param.UserFound.items()
                    if v.get("id") == None
                ]
                if len(NotFoundUser) > 0:
                    logging.info("Some Users Not Found = %s", NotFoundUser)
                self.UpdateParam()  # On met à jour la Taxo
                if len(NotFoundUser) > 0 or len(NotFoundTaxo) > 0:
                    txt += "<a class='btn btn-primary' href='?src={0}&dest={1}&prjok=1'>Continue</a>".format(
                        gvg("src"), Prj.projid)
                else:
                    txt += "<a class='btn btn-primary' href='?src={0}&dest={1}&prjok=2'>Continue</a>".format(
                        gvg("src"), Prj.projid)

                return PrintInCharte(txt)

            if gvg("prjok") == "1" or gvg(
                    "prjok"
            ) == "2":  # 2 pas de mapping requis, simule un start task
                NotFoundTaxo = [
                    t["name"] for t in self.param.TaxoFound.values()
                    if t["newid"] is None
                ]  # liste des Taxon sans mapping restant
                NotFoundUsers = [
                    k for k, v in self.param.UserFound.items()
                    if v.get("id") == None
                ]
                app.logger.info("TaxoFound=%s", self.param.TaxoFound)
                app.logger.info("NotFoundTaxo=%s", NotFoundTaxo)
                app.logger.info("NotFoundUser=%s", NotFoundUsers)

                if gvp('starttask') == "Y" or gvg("prjok") == "2":
                    app.logger.info("Form Data = %s", request.form)
                    # Traitement des reponses sur la taxonomy
                    for i in range(1, 1 + len(NotFoundTaxo)):
                        orig = gvp(
                            "orig%d" % (i)
                        )  #Le nom original est dans origXX et la nouvelle valeur dans taxolbXX
                        origname = gvp(
                            "origname%d" % (i)
                        )  #Le nom original est dans origXX et la nouvelle valeur dans taxolbXX
                        action = gvp("action%d" % (i))
                        newvalue = gvp("taxolb%d" % (i))
                        if origname in NotFoundTaxo and action != "":
                            if action == "M":
                                if newvalue == "":
                                    errors.append(
                                        "Taxonomy Manual Mapping : No Value Selected  for '%s'"
                                        % (orig, ))
                                else:
                                    t = database.Taxonomy.query.filter(
                                        database.Taxonomy.id == int(
                                            newvalue)).first()
                                    app.logger.info(orig + " associated to " +
                                                    t.name)
                                    self.param.TaxoFound[orig]['newid'] = t.id
                            elif action == "U":  #create Under
                                if newvalue == "":
                                    errors.append(
                                        "Taxonomy Manual Mapping : No Parent Value Selected  for '%s'"
                                        % (orig, ))
                                else:
                                    t = database.Taxonomy()
                                    t.name = origname
                                    t.parent_id = int(newvalue)
                                    db.session.add(t)
                                    db.session.commit()
                                    self.param.TaxoFound[orig]['newid'] = t.id
                                    app.logger.info(orig + " created under " +
                                                    t.name)
                            else:
                                errors.append(
                                    "Taxonomy Manual Mapping : No Action Selected  for '%s'"
                                    % (orig, ))
                        else:
                            errors.append(
                                "Taxonomy Manual Mapping : Invalid value '%s' for '%s'"
                                % (newvalue, orig))
                    # Traitement des reponses sur les utilisateurs
                    for i in range(1, 1 + len(NotFoundUsers)):
                        orig = gvp(
                            "origuser%d" % (i)
                        )  #Le nom original est dans origuserXX et la nouvelle valeur dans userlbXX
                        action = gvp("useraction%d" % (i))
                        newvalue = gvp("userlb%d" % (i))
                        if orig in NotFoundUsers and (newvalue != ""
                                                      or action == "C"):
                            if action == "C":
                                u = database.users()
                                u.email = self.param.UserFound[orig]["email"]
                                u.name = self.param.UserFound[orig]["name"]
                                u.password = self.param.UserFound[orig][
                                    "password"]
                                u.organisation = ntcv(
                                    self.param.UserFound[orig]["organisation"]
                                )  #+" Imported on "+datetime.datetime.now().strftime("%Y-%m-%d %H:%M")
                                u.active = True
                                db.session.add(u)
                                db.session.commit()
                            else:
                                u = database.users.query.filter(
                                    database.users.id == int(
                                        newvalue)).first()
                            app.logger.info("User " + orig +
                                            " associated to " + u.name)
                            self.param.UserFound[orig]['id'] = u.id
                        else:
                            errors.append(
                                "User Manual Mapping : Invalid value '%s' for '%s'"
                                % (newvalue, orig))
                    app.logger.info("Final Taxofound = %s",
                                    self.param.TaxoFound)
                    self.param.ProjectId = int(gvg("dest"))
                    self.param.ProjectSrcId = int(gvg("src"))
                    self.param.IntraStep = 1
                    self.UpdateParam()  # On met à jour ce qui à été accepté
                    # Verifier la coherence des données
                    if len(errors) == 0:
                        return self.StartTask(self.param, step=2)
                    for e in errors:
                        flash(e, "error")
                    NotFoundTaxo = [
                        v["name"] for k, v in self.param.TaxoFound.items()
                        if v["newid"] == None
                    ]
                    NotFoundUsers = [
                        k for k, v in self.param.UserFound.items()
                        if v.get('id') == None
                    ]
                    app.logger.info("Final NotFoundTaxo = %s", NotFoundTaxo)
                NotFoundTaxoForTemplate = [{
                    'id': v["id"],
                    'name': v["name"],
                    'namefull': v["namefull"]
                } for k, v in self.param.TaxoFound.items()
                                           if v["newid"] == None]
                return render_template('task/importdb_question1.html',
                                       header=txt,
                                       taxo=NotFoundTaxoForTemplate,
                                       users=NotFoundUsers)
            return PrintInCharte(txt)
        if self.task.taskstep == 2:  # ################## Question Post Import Effectif d'un projet
            # Propose de voir le projet, de cleanner, ou d'importer un autre projet
            return PrintInCharte(self.GetDoneExtraAction())
コード例 #11
0
ファイル: taskimport.py プロジェクト: obs6/ecotaxa_dev
 def SPStep2(self):
     # raise Exception("TEST")
     logging.info("Start Step 2 : Effective data import")
     logging.info("Taxo Mapping = %s", self.param.TaxoFound)
     logging.info("Users Mapping = %s", self.param.UserFound)
     AstralCache = {
         'date': None,
         'time': None,
         'long': None,
         'lat': None,
         'r': ''
     }
     # Mise à jour du mapping en base
     Prj = database.Projects.query.filter_by(
         projid=self.param.ProjectId).first()
     LoadedFiles = ntcv(Prj.fileloaded).splitlines()
     logging.info("LoadedFiles = %s", LoadedFiles)
     Prj.mappingobj = EncodeEqualList({
         v['field']: v.get('title')
         for k, v in self.param.Mapping.items()
         if v['table'] == 'obj_field' and v['field'][0] in (
             't', 'n') and v.get('title') != None
     })
     Prj.mappingsample = EncodeEqualList({
         v['field']: v.get('title')
         for k, v in self.param.Mapping.items()
         if v['table'] == 'sample' and v['field'][0] in (
             't', 'n') and v.get('title') != None
     })
     Prj.mappingacq = EncodeEqualList({
         v['field']: v.get('title')
         for k, v in self.param.Mapping.items()
         if v['table'] == 'acq' and v['field'][0] in (
             't', 'n') and v.get('title') != None
     })
     Prj.mappingprocess = EncodeEqualList({
         v['field']: v.get('title')
         for k, v in self.param.Mapping.items()
         if v['table'] == 'process' and v['field'][0] in (
             't', 'n') and v.get('title') != None
     })
     db.session.commit()
     Ids = {
         "acq": {
             "tbl": "acquisitions",
             "pk": "acquisid"
         },
         "sample": {
             "tbl": "samples",
             "pk": "sampleid"
         },
         "process": {
             "tbl": "process",
             "pk": "processid"
         }
     }
     #recupération des orig_id des acq,sample,process
     for i in Ids:
         sql = "select orig_id," + Ids[i]['pk'] + " from " + Ids[i][
             'tbl'] + " where projid=" + str(self.param.ProjectId)
         Ids[i]["ID"] = {}
         for r in GetAll(sql):
             Ids[i]["ID"][r[0]] = int(r[1])
     # recuperation de les ID objet du projet
     self.ExistingObject = {}
     self.pgcur.execute(
         "SELECT o.orig_id,o.objid from objects o where o.projid=" +
         str(self.param.ProjectId))
     for rec in self.pgcur:
         self.ExistingObject[rec[0]] = rec[1]
     # recuperation de toutes les paire objet/Images du projet
     self.ExistingObjectAndImage = set()
     if self.param.SkipObjectDuplicate == 'Y':  # cette liste n'est necessaire qui si on ignore les doublons
         self.pgcur.execute(  # et doit être recahrgée depuis la base, car la phase 1 y a ajouté tous les objets pour le contrôle des doublons
             "SELECT concat(o.orig_id,'*',i.orig_file_name) from images i join objects o on i.objid=o.objid where o.projid="
             + str(self.param.ProjectId))
         for rec in self.pgcur:
             self.ExistingObjectAndImage.add(rec[0])
     #logging.info("Ids = %s",Ids)
     random.seed()
     sd = Path(self.param.SourceDir)
     TotalRowCount = 0
     for filter in ("**/ecotaxa*.txt", "**/ecotaxa*.tsv"):
         for CsvFile in sd.glob(filter):
             relname = CsvFile.relative_to(
                 sd)  # Nom relatif à des fins d'affichage uniquement
             if relname.as_posix(
             ) in LoadedFiles and self.param.SkipAlreadyLoadedFile == 'Y':
                 logging.info("File %s skipped, already loaded" %
                              (relname.as_posix()))
                 continue
             logging.info("Analyzing file %s" % (relname.as_posix()))
             with open(CsvFile.as_posix(), encoding='latin_1') as csvfile:
                 # lecture en mode dictionnaire basé sur la premiere ligne
                 rdr = csv.DictReader(csvfile,
                                      delimiter='\t',
                                      quotechar='"')
                 #lecture la la ligne des types (2nd ligne du fichier
                 LType = {
                     champ.strip(" \t").lower(): v
                     for champ, v in rdr.__next__().items()
                 }
                 ListeChamps = [
                     champ.strip(" \t").lower() for champ in rdr.fieldnames
                 ]
                 # Chargement du contenu du fichier
                 RowCount = 0
                 for rawlig in rdr:
                     lig = {
                         champ.strip(" \t").lower(): v
                         for champ, v in rawlig.items()
                     }
                     Objs = {
                         "acq": database.Acquisitions(),
                         "sample": database.Samples(),
                         "process": database.Process(),
                         "obj_head": database.Objects(),
                         "obj_field": database.ObjectsFields(),
                         "image": database.Images()
                     }
                     RowCount += 1
                     TotalRowCount += 1
                     if 'object_annotation_category_id' in ListeChamps and 'object_annotation_category' in ListeChamps:
                         if CleanValue(
                                 lig.get('object_annotation_category_id',
                                         '')) != '':
                             del lig[
                                 'object_annotation_category']  # s'il y a un ID on ignore le texte
                     for champ in ListeChamps:
                         m = self.param.Mapping.get(champ, None)
                         if m is None:
                             continue  # Le champ n'est pas considéré
                         FieldName = m.get("field", None)
                         FieldTable = m.get("table", None)
                         FieldValue = None
                         v = CleanValue(lig.get(champ))
                         if v != "":  # si pas de valeurs, on laisse le champ null
                             if m['type'] == 'n':
                                 FieldValue = ToFloat(v)
                             elif champ == 'object_date':
                                 FieldValue = datetime.date(
                                     int(v[0:4]), int(v[4:6]), int(v[6:8]))
                             elif champ == 'object_time':
                                 v = v.zfill(6)
                                 FieldValue = datetime.time(
                                     int(v[0:2]), int(v[2:4]), int(v[4:6]))
                             elif FieldName == 'classif_when':
                                 v2 = CleanValue(
                                     lig.get('object_annotation_time',
                                             '000000')).zfill(6)
                                 FieldValue = datetime.datetime(
                                     int(v[0:4]), int(v[4:6]), int(v[6:8]),
                                     int(v2[0:2]), int(v2[2:4]),
                                     int(v2[4:6]))
                             elif FieldName == 'classif_id':  # pour la version numerique, c'est traité par if type=n
                                 v = self.param.TaxoMap.get(
                                     v.lower(), v
                                 )  # Applique le mapping initial d'entrée
                                 FieldValue = self.param.TaxoFound[ntcv(
                                     v).lower()]
                             elif FieldName == 'classif_who':
                                 FieldValue = self.param.UserFound[ntcv(
                                     v).lower()].get('id', None)
                             elif FieldName == 'classif_qual':
                                 FieldValue = database.ClassifQualRevert.get(
                                     v.lower())
                             else:  # c'est un champ texte sans rien de special
                                 FieldValue = v
                             if FieldTable in Objs:
                                 # if FieldName in Objs[FieldTable].__dict__:
                                 if hasattr(Objs[FieldTable], FieldName):
                                     setattr(Objs[FieldTable], FieldName,
                                             FieldValue)
                                     # logging.info("setattr %s %s %s",FieldTable,FieldName,FieldValue)
                                 # else:
                                 # logging.info("skip F %s %s %s",FieldTable,FieldName,FieldValue)
                             else:
                                 logging.info("skip T %s %s %s", FieldTable,
                                              FieldName, FieldValue)
                     # Calcul de la position du soleil
                     if not (AstralCache['date']==Objs["obj_head"].objdate and AstralCache['time']==Objs["obj_head"].objtime \
                         and AstralCache['long']==Objs["obj_head"].longitude and AstralCache['lat']==Objs["obj_head"].latitude) :
                         AstralCache = {
                             'date': Objs["obj_head"].objdate,
                             'time': Objs["obj_head"].objtime,
                             'long': Objs["obj_head"].longitude,
                             'lat': Objs["obj_head"].latitude,
                             'r': ''
                         }
                         from astral import AstralError
                         try:
                             AstralCache['r'] = appli.CalcAstralDayTime(
                                 AstralCache['date'], AstralCache['time'],
                                 AstralCache['lat'], AstralCache['long'])
                         except AstralError as e:  # dans certains endoit du globe il n'y a jamais de changement nuit/jour certains jours, ca provoque une erreur
                             app.logger.error("Astral error : %s for %s", e,
                                              AstralCache)
                         except Exception as e:  # autre erreurs par exemple si l'heure n'est pas valide;
                             app.logger.error("Astral error : %s for %s", e,
                                              AstralCache)
                     Objs["obj_head"].sunpos = AstralCache['r']
                     # Affectation des ID Sample, Acq & Process et creation de ces dernier si necessaire
                     for t in Ids:
                         if Objs[t].orig_id is not None:
                             if Objs[t].orig_id in Ids[t]["ID"]:
                                 setattr(Objs["obj_head"], Ids[t]["pk"],
                                         Ids[t]["ID"][Objs[t].orig_id])
                             else:
                                 Objs[t].projid = self.param.ProjectId
                                 db.session.add(Objs[t])
                                 db.session.commit()
                                 Ids[t]["ID"][Objs[t].orig_id] = getattr(
                                     Objs[t], Ids[t]["pk"])
                                 setattr(Objs["obj_head"], Ids[t]["pk"],
                                         Ids[t]["ID"][Objs[t].orig_id])
                                 logging.info("IDS %s %s", t, Ids[t])
                     self.pgcur.execute("select nextval('seq_images')")
                     Objs["image"].imgid = self.pgcur.fetchone()[0]
                     CleExistObj = Objs["obj_field"].orig_id + '*' + Objs[
                         "image"].orig_file_name
                     if self.param.SkipObjectDuplicate == 'Y' and CleExistObj in self.ExistingObjectAndImage:
                         continue
                     # Recherche de l'objet si c'est une images complementaire
                     if Objs["obj_field"].orig_id in self.ExistingObject:
                         Objs["obj_head"].objid = self.ExistingObject[
                             Objs["obj_field"].orig_id]
                     else:  # ou Creation de l'objet
                         Objs["obj_head"].projid = self.param.ProjectId
                         Objs["obj_head"].random_value = random.randint(
                             1, 99999999)
                         Objs["obj_head"].img0id = Objs["image"].imgid
                         db.session.add(Objs["obj_head"])
                         db.session.commit()
                         Objs["obj_field"].objfid = Objs["obj_head"].objid
                         db.session.add(Objs["obj_field"])
                         db.session.commit()
                         self.ExistingObject[
                             Objs["obj_field"].orig_id] = Objs[
                                 "obj_head"].objid  # Provoque un select object sauf si 'expire_on_commit':False
                     #Gestion de l'image, creation DB et fichier dans Vault
                     Objs["image"].objid = Objs["obj_head"].objid
                     ImgFilePath = CsvFile.with_name(
                         Objs["image"].orig_file_name)
                     VaultFolder = "%04d" % (Objs["image"].imgid // 10000)
                     vaultroot = Path("../../vault")
                     #creation du repertoire contenant les images si necessaire
                     CreateDirConcurrentlyIfNeeded(
                         vaultroot.joinpath(VaultFolder))
                     vaultfilename = "%s/%04d%s" % (
                         VaultFolder, Objs["image"].imgid % 10000,
                         ImgFilePath.suffix)
                     vaultfilenameThumb = "%s/%04d_mini%s" % (
                         VaultFolder, Objs["image"].imgid % 10000, '.jpg'
                     )  #on Impose le format de la miniature
                     Objs["image"].file_name = vaultfilename
                     #copie du fichier image
                     shutil.copyfile(
                         ImgFilePath.as_posix(),
                         vaultroot.joinpath(vaultfilename).as_posix())
                     im = Image.open(
                         vaultroot.joinpath(vaultfilename).as_posix())
                     Objs["image"].width = im.size[0]
                     Objs["image"].height = im.size[1]
                     SizeLimit = app.config['THUMBSIZELIMIT']
                     # génération d'une miniature si une image est trop grande.
                     if (im.size[0] > SizeLimit) or (im.size[1] >
                                                     SizeLimit):
                         im.thumbnail((SizeLimit, SizeLimit))
                         if im.mode == 'P':
                             im = im.convert("RGB")
                         im.save(
                             vaultroot.joinpath(
                                 vaultfilenameThumb).as_posix())
                         Objs["image"].thumb_file_name = vaultfilenameThumb
                         Objs["image"].thumb_width = im.size[0]
                         Objs["image"].thumb_height = im.size[1]
                     #ajoute de l'image en DB
                     if Objs["image"].imgrank is None:
                         Objs["image"].imgrank = 0  # valeur par defaut
                     db.session.add(Objs["image"])
                     db.session.commit()
                     if (TotalRowCount % 100) == 0:
                         self.UpdateProgress(
                             100 * TotalRowCount / self.param.TotalRowCount,
                             "Processing files %d/%d" %
                             (TotalRowCount, self.param.TotalRowCount))
                 logging.info("File %s : %d row Loaded", relname.as_posix(),
                              RowCount)
                 LoadedFiles.append(relname.as_posix())
                 Prj.fileloaded = "\n".join(LoadedFiles)
                 db.session.commit()
     self.pgcur.execute("""update obj_head o
                         set imgcount=(select count(*) from images where objid=o.objid)
                         ,img0id=(select imgid from images where objid=o.objid order by imgrank asc limit 1 )
                         where projid=""" + str(self.param.ProjectId))
     self.pgcur.connection.commit()
     self.pgcur.execute(
         """update samples s set latitude=sll.latitude,longitude=sll.longitude
           from (select o.sampleid,min(o.latitude) latitude,min(o.longitude) longitude
           from obj_head o
           where projid=%(projid)s and o.latitude is not null and o.longitude is not null
           group by o.sampleid) sll where s.sampleid=sll.sampleid and projid=%(projid)s """,
         {'projid': self.param.ProjectId})
     self.pgcur.connection.commit()
     appli.project.main.RecalcProjectTaxoStat(Prj.projid)
     appli.project.main.UpdateProjectStat(Prj.projid)
     self.task.taskstate = "Done"
     self.UpdateProgress(100, "Processing done")
コード例 #12
0
ファイル: objectdetails.py プロジェクト: obs6/ecotaxa_dev
def objectdetails(objid):
    #récuperation et ajustement des dimensions de la zone d'affichage
    try:
        PageWidth = int(gvg(
            "w")) - 40  # on laisse un peu de marge à droite et la scroolbar
        if PageWidth < 200: PageWidth = 20000
        WindowHeight = int(gvg("h")) - 40  # on laisse un peu de marge en haut
        if WindowHeight < 200: WindowHeight = 20000
    except:
        PageWidth = 20000
        WindowHeight = 20000

    obj = database.Objects.query.filter_by(objid=objid).first()
    t = list()
    # Dans cet écran on utilise ElevateZoom car sinon en mode popup il y a conflit avec les images sous la popup
    t.append("<script src='/static/jquery.elevatezoom.js'></script>")
    Prj = obj.project
    if Prj.visible == False and not Prj.CheckRight(
            0):  # Level 0 = Read, 1 = Annotate, 2 = Admin
        flash('You cannot view this project', 'error')
        return PrintInCharte("<a href=/>Back to home</a>")
    g.Projid = Prj.projid
    PrjManager = [(m.memberrel.email, m.memberrel.name)
                  for m in Prj.projmembers if m.privilege == 'Manage']
    t.append("<p>Project: <b><a href='/prj/%d'>%s</a></b> (managed by : %s)" %
             (Prj.projid, XSSEscape(Prj.title), ",".join(
                 ("<a href ='mailto:%s'>%s</a>" % m for m in PrjManager))))
    if len(PrjManager) > 0:
        t.append(
            "<br>To report a mistake, contact <a href ='mailto:{0}?subject=Ecotaxa%20mistake%20notification&body={2}'>{1}</a>"
            .format(
                PrjManager[0][0], PrjManager[0][1],
                urllib.parse.quote(
                    "Hello,\n\nI have discovered a mistake on this page " +
                    request.base_url + "\n")))


# //window.location="mailto:?subject=Ecotaxa%20page%20share&body="+encodeURIComponent("Hello,\n\nAn Ecotaxa user want share this page with you \n"+url);
    t.append("</p><p>Classification :")
    if obj.classif:
        t.append("<br>&emsp;<b>%s</b>" % XSSEscape(obj.classif.display_name))
        TaxoHierarchie = (r[0] for r in GetAll(
            """WITH RECURSIVE rq(id,name,parent_id) as ( select id,name,parent_id FROM taxonomy where id =%(taxoid)s
                        union
                        SELECT t.id,t.name,t.parent_id FROM rq JOIN taxonomy t ON t.id = rq.parent_id )
                        select name from rq""", {"taxoid": obj.classif.id}))
        t.append("<br>&emsp;" + (" &lt; ".join(TaxoHierarchie)) +
                 " (id=%s)" % obj.classif_id)
    else:
        t.append("<br>&emsp;<b>Unknown</b>")
    if obj.classiffier is not None:
        t.append(
            "<br>&emsp;%s " %
            (database.ClassifQual.get(obj.classif_qual, "To be classified")))
        t.append(" by %s (%s) " %
                 (obj.classiffier.name, obj.classiffier.email))
        if obj.classif_when is not None:
            t.append(" on %s " % (obj.classif_when.strftime("%Y-%m-%d %H:%M")))
    t.append("</p>")
    if obj.objfrel.object_link is not None:
        t.append("<p>External link :<a href='{0}' target=_blank> {0}</a></p>".
                 format(obj.objfrel.object_link))
    t.append(
        "<table><tr><td valign=top>Complementaty information <a href='javascript:gotocommenttab();' > ( edit )</a>: </td><td> <span id=spancomplinfo> {0}</span></td></tr></table>"
        .format(ntcv(obj.complement_info).replace('\n', '<br>\n')))
    # On affiche la liste des images, en selectionnant une image on changera le contenu de l'image Img1 + Redim
    # l'approche avec des onglets de marchait pas car les images sont superposées
    obj.images.sort(key=lambda x: x.imgrank)
    t.append("""<p>Image list : """)
    for img in obj.images:
        (width, height) = ComputeLimitForImage(img.width, img.height,
                                               PageWidth, WindowHeight)
        if img.thumb_file_name:
            minifile = img.thumb_file_name
            (miniwidth,
             miniheight) = ComputeLimitForImage(img.thumb_width,
                                                img.thumb_height, 30, 30)
        else:
            minifile = img.file_name
            (miniwidth,
             miniheight) = ComputeLimitForImage(img.width, img.height, 30, 30)
        t.append(
            """<a href="javascript:SwapImg1('{1}',{2},{3});" >{0} <img src=/vault/{4} width={5} height={6}></a> """
            .format(img.imgrank + 1, img.file_name, width, height, minifile,
                    miniwidth, miniheight))
    # Ajout de la 1ère image
    (width, height) = ComputeLimitForImage(obj.images[0].width,
                                           obj.images[0].height, PageWidth,
                                           WindowHeight)
    t.append(
        "</p><p><img id=img1 src=/vault/{1} data-zoom-image=/vault/{1} width={2} height={0}></p>"
        .format(height, obj.images[0].file_name, width))
    # Affichage de l'onglet de classification
    if Prj.CheckRight(1):
        t.append("""
<table><tr><td>Set a new classification :</td>
 <td style="width: 230px;">
     <div class="input-group">
       <select id="taxolbpop" name="taxolbpop" style="width: 200px" class='taxolb' > </select>"""
                 )
        if gvg("ajax", "0") == "0":
            t.append("""<span class="input-group-btn">
                    <button class="btn btn-default btn-sm" type="button"  data-toggle="modal" data-target="#TaxoModal" data-mytargetid="taxolbpop" title="Search on Taxonomy Tree">
                        <span id=OpenTaxoLB class="glyphicon glyphicon-th-list" aria-hidden="true"/></button>
                    </span>""")
        else:
            t.append("<br>")
        t.append("""</div><!-- /input-group -->
 <span id=PendingChangesPop></span></td><td width=30px></td><td valign=top>
    <button type="button" class="btn btn-success btn-xs" onclick="Save1Object('V');">Save as Validated</button>
    <button type="button" class="btn btn-warning btn-xs" onclick="Save1Object('D');">Save as dubious</button>
    <button id=btenableedit type="button" class="btn btn-gris btn-xs" onclick="EnableEdit();">Enable Editing</button>
    <button type="button" class="btn btn-default btn-xs"  onclick="$('#PopupDetails').modal('hide');">Close</button>
    </td></tr></table>
    """)
    # Ajout des Onglets sous l'image
    t.append("""<br><div><ul class="nav nav-tabs" role="tablist">
    <li role="presentation" class="active"><a href="#tabdobj" aria-controls="tabdobj" role="tab" data-toggle="tab"> Object details</a></li>
    <li role="presentation" ><a href="#tabdsample" aria-controls="tabdsample" role="tab" data-toggle="tab"> Sample details</a></li>
    <li role="presentation" ><a href="#tabdacquis" aria-controls="tabdacquis" role="tab" data-toggle="tab"> Acquisition details</a></li>
    <li role="presentation" ><a href="#tabdprocessrel" aria-controls="tabdprocess" role="tab" data-toggle="tab"> Processing details</a></li>
    <li role="presentation" ><a href="#tabdclassiflog" aria-controls="tabdclassiflog" role="tab" data-toggle="tab">Classification change log</a></li>
    <li role="presentation" ><a href="#tabdmap" aria-controls="tabdmap" role="tab" data-toggle="tab" id=atabdmap style="background: #5CB85C;color:white;">Map</a></li>
    """)
    if Prj.CheckRight(1):
        t.append(
            """<li role="presentation" ><a id=linktabdaddcomments href="#tabdaddcomments" aria-controls="tabdaddcomments" role="tab" data-toggle="tab">Edit complementary informations</a></li>"""
        )
    if obj.classif_auto:
        classif_auto_name = obj.classif_auto.name
        if obj.classif_auto_score:
            classif_auto_name += " (%0.3f)" % (obj.classif_auto_score, )
    else:
        classif_auto_name = ''
    t.append("""</ul>
    <div class="tab-content">
    <div role="tabpanel" class="tab-pane active" id="tabdobj">
    <table class='table table-bordered table-condensed' data-table='object'><tr>
    <td style=' background-color: #f2f2f2;' data-edit='longitude'><b>longitude</td><td>{0}</td>
    <td style=' background-color: #f2f2f2;' data-edit='latitude'><b>latitude</td><td>{1}</td>
    <td style=' background-color: #f2f2f2;' data-edit='objdate'><b>Date</td><td>{2}</td>
    <td style=' background-color: #f2f2f2;'><b>Time (daytime)</td><td>{3} ({10})</td>
    </tr><tr><td style=' background-color: #f2f2f2;' data-edit='depth_min'><b>Depth min</td><td>{4}</td>
    <td style=' background-color: #f2f2f2;' data-edit='depth_max'><b>Depth max</td><td>{5}</td>
    <td><b>Classif auto</td><td>{6}</td><td><b>Classif auto when</td><td>{7}</td>
    </tr><tr><td><b>Object #</td><td>{8}</td>
    <td data-edit='orig_id'><b>Original Object ID</td><td colspan=5>{9}</td></tr><tr>"""
             .format(nonetoformat(obj.longitude, '.5f'),
                     nonetoformat(obj.latitude,
                                  '.5f'), obj.objdate, obj.objtime,
                     obj.depth_min, obj.depth_max, classif_auto_name,
                     obj.classif_auto_when, objid, obj.objfrel.orig_id,
                     database.DayTimeList.get(obj.sunpos, '?')))
    cpt = 0
    # Insertion des champs object
    for k, v in collections.OrderedDict(
            sorted(DecodeEqualList(Prj.mappingobj).items())).items():
        if cpt > 0 and cpt % 4 == 0:
            t.append("</tr><tr>")
        cpt += 1
        t.append("<td data-edit='{2}'><b>{0}</td><td>{1}</td>".format(
            v, ScaleForDisplay(getattr(obj.objfrel, k, "???")), k))
    t.append("</tr></table></div>")
    # insertion des champs Sample, Acquisition & Processing dans leurs onglets respectifs
    for r in (("Sample", "mappingsample", "sample"), ("Acquisition",
                                                      "mappingacq", "acquis"),
              ("Processing", "mappingprocess", "processrel")):
        t.append(
            '<div role="tabpanel" class="tab-pane" id="tabd' + r[2] + '">' +
            r[0] +
            " details :<table class='table table-bordered table-condensed'  data-table='"
            + r[2] + "'><tr>")
        cpt = 0
        if getattr(obj, r[2]):
            if r[2] == "sample":
                t.append(
                    "<td data-edit='orig_id'><b>{0}</td><td colspan=3>{1}</td><td data-edit='longitude'><b>{2}</td><td>{3}</td><td data-edit='latitude'><b>{4}</td><td>{5}</td></tr><tr>"
                    .format(
                        "Original ID",
                        ScaleForDisplay(obj.sample.orig_id),
                        "longitude",
                        ScaleForDisplay(obj.sample.longitude),
                        "latitude",
                        ScaleForDisplay(obj.sample.latitude),
                    ))
            elif r[2] == "acquis":
                t.append(
                    "<td data-edit='orig_id'><b>{0}</td><td colspan=3>{1}</td><td data-edit='instrument'><b>{2}</td><td>{3}</td></tr><tr>"
                    .format(
                        "Original ID",
                        ScaleForDisplay(obj.acquis.orig_id),
                        "Instrument",
                        ScaleForDisplay(obj.acquis.instrument),
                    ))
            else:
                t.append(
                    "<td data-edit='orig_id'><b>{0}</td><td>{1}</td></tr><tr>".
                    format(
                        "Original ID.",
                        ScaleForDisplay(
                            getattr(getattr(obj, r[2]), "orig_id", "???"))))
            for k, v in collections.OrderedDict(
                    sorted(DecodeEqualList(getattr(Prj,
                                                   r[1])).items())).items():
                if cpt > 0 and cpt % 4 == 0:
                    t.append("</tr><tr>")
                cpt += 1
                t.append("<td data-edit='{2}'><b>{0}</td><td>{1}</td>".format(
                    v, ScaleForDisplay(getattr(getattr(obj, r[2]), k, "???")),
                    k))
            if r[2] == "sample":
                t.append(
                    "</tr><tr><td><b>{0}</td><td colspan=7>{1}</td></tr><tr>".
                    format(
                        "Dataportal Desc.",
                        ScaleForDisplay(
                            html.escape(ntcv(
                                obj.sample.dataportal_descriptor)))))
        else:
            t.append("<td>No {0}</td>".format(r[0]))
        t.append("</tr></table></div>")

    # Affichage de l'historique des classification
    t.append("""<div role="tabpanel" class="tab-pane" id="tabdclassiflog">
Current Classification : Quality={} , date={}    
    <table class='table table-bordered table-condensed'><tr>
    <td>Date</td><td>Type</td><td>Taxo</td><td>Author</td><td>Quality</td></tr>"""
             .format(obj.classif_qual, obj.classif_when))
    Histo = GetAll(
        """SELECT to_char(classif_date,'YYYY-MM-DD HH24:MI:SS') datetxt,classif_type ,t.display_name as name,u.name username,classif_qual
  from objectsclassifhisto h
  left join taxonomy t on h.classif_id=t.id
  LEFT JOIN users u on u.id = h.classif_who
WHERE objid=%(objid)s
order by classif_date desc""", {"objid": objid},
        doXSSEscape=True)
    for r in Histo:
        t.append("<tr><td>" + ("</td><td>".join([
            str(r[x]) if r[x] else "-"
            for x in ("datetxt", "classif_type", "name", "username",
                      "classif_qual")
        ])) + "</td></tr>")
    t.append("</table></div>")
    if Prj.CheckRight(1):
        t.append("""<div role="tabpanel" class="tab-pane" id="tabdaddcomments">
        <textarea id=compinfo rows=5 cols=120 autocomplete=off>%s</textarea><br>
        <button type="button" class='btn btn-primary' onclick="UpdateComment();">Save additional comment</button>
        <span id=ajaxresultcomment></span>
        """ % (ntcv(obj.complement_info), ))
        t.append("</div>")
    # Affichage de la carte
    t.append("""
    <div role="tabpanel" class="tab-pane" id="tabdmap">
<div id="map2" class="map2" style="width: 100%; height: 450px;">
  Displaying Map requires Internet Access to load map from https://server.arcgisonline.com
</div>""")

    t.append("</table></div>")
    t.append(
        render_template("common/objectdetailsscripts.html",
                        Prj=Prj,
                        objid=objid,
                        obj=obj))

    # En mode popup ajout en haut de l'écran d'un hyperlien pour ouvrir en fenete isolée
    # Sinon affichage sans lien dans la charte.
    if gvg("ajax", "0") == "1":
        return """<table width=100% style='margin: 3px'><tr><td><a href='/objectdetails/{0}?w={1}&h={2}' target=_blank><b>Open in a separate window</b> (right click to copy link)</a>
        </td><td align='right'><button type="button" class="btn btn-default"  onclick="$('#PopupDetails').modal('hide');">Close</button>&nbsp;&nbsp;
        </td></tr></table><div style='margin: 0 5px;'>""".format(
            objid, gvg("w"), gvg("h")) + "\n".join(t)
    return PrintInCharte("<div style='margin-left:10px;'>" + "\n".join(t) +
                         render_template('common/taxopopup.html'))