import sys, os, types, logging, stat #sys.path.append('C:\\Expo\\expoweb') #from troggle import * #os.environ['DJANGO_SETTINGS_MODULE']='troggle.settings' import settings from troggle.core.models import * from PIL import Image #import settings #import core.models as models import csv import re import datetime from utils import save_carefully def get_or_create_placeholder(year): """ All surveys must be related to a logbookentry. We don't have a way to automatically figure out which survey went with which logbookentry, so we create a survey placeholder logbook entry for each year. This function always returns such a placeholder, and creates it if it doesn't exist yet. """ lookupAttribs={'date__year':int(year), 'title':"placeholder for surveys",} nonLookupAttribs={'text':"surveys temporarily attached to this should be re-attached to their actual trips", 'date':datetime.date(int(year),1,1)} placeholder_logbook_entry, newly_created = save_carefully(LogbookEntry, lookupAttribs, nonLookupAttribs) return placeholder_logbook_entry # dead def readSurveysFromCSV(): try: # could probably combine these two surveytab = open(os.path.join(settings.SURVEY_SCANS, "Surveys.csv")) except IOError: import cStringIO, urllib surveytab = cStringIO.StringIO(urllib.urlopen(settings.SURVEY_SCANS + "/Surveys.csv").read()) dialect=csv.Sniffer().sniff(surveytab.read()) surveytab.seek(0,0) surveyreader = csv.reader(surveytab,dialect=dialect) headers = surveyreader.next() header = dict(zip(headers, range(len(headers)))) #set up a dictionary where the indexes are header names and the values are column numbers # test if the expeditions have been added yet if Expedition.objects.count()==0: print("There are no expeditions in the database. Please run the logbook parser.") sys.exit() logging.info("Deleting all scanned images") ScannedImage.objects.all().delete() logging.info("Deleting all survey objects") Survey.objects.all().delete() logging.info("Beginning to import surveys from "+str(os.path.join(settings.SURVEYS, "Surveys.csv"))+"\n"+"-"*60+"\n") for survey in surveyreader: #I hate this, but some surveys have a letter eg 2000#34a. The next line deals with that. walletNumberLetter = re.match(r'(?P\d*)(?P[a-zA-Z]*)',survey[header['Survey Number']]) # print(walletNumberLetter.groups()) year=survey[header['Year']] surveyobj = Survey( expedition = Expedition.objects.filter(year=year)[0], wallet_number = walletNumberLetter.group('number'), logbook_entry = get_or_create_placeholder(year), comments = survey[header['Comments']], location = survey[header['Location']] ) surveyobj.wallet_letter = walletNumberLetter.group('letter') if survey[header['Finished']]=='Yes': #try and find the sketch_scan pass surveyobj.save() logging.info("added survey " + survey[header['Year']] + "#" + surveyobj.wallet_number + "\r") # dead def listdir(*directories): try: return os.listdir(os.path.join(settings.SURVEYS, *directories)) except: import urllib url = settings.SURVEYS + reduce(lambda x, y: x + "/" + y, ["listdir"] + list(directories)) folders = urllib.urlopen(url.replace("#", "%23")).readlines() return [folder.rstrip(r"/") for folder in folders] # add survey scans def parseSurveyScans(expedition, logfile=None): # yearFileList = listdir(expedition.year) try: yearPath=os.path.join(settings.SURVEY_SCANS, "surveyscans", expedition.year) yearFileList=os.listdir(yearPath) print(yearFileList) for surveyFolder in yearFileList: try: surveyNumber=re.match(r'\d\d\d\d#(X?)0*(\d+)',surveyFolder).groups() #scanList = listdir(expedition.year, surveyFolder) scanList=os.listdir(os.path.join(yearPath,surveyFolder)) except AttributeError: print(surveyFolder + " ignored\r",) continue for scan in scanList: try: scanChopped=re.match(r'(?i).*(notes|elev|plan|elevation|extend)(\d*)\.(png|jpg|jpeg)',scan).groups() scanType,scanNumber,scanFormat=scanChopped except AttributeError: print(scan + " ignored\r",) continue if scanType == 'elev' or scanType == 'extend': scanType = 'elevation' if scanNumber=='': scanNumber=1 if type(surveyNumber)==types.TupleType: surveyLetter=surveyNumber[0] surveyNumber=surveyNumber[1] try: placeholder=get_or_create_placeholder(year=int(expedition.year)) survey=Survey.objects.get_or_create(wallet_number=surveyNumber, wallet_letter=surveyLetter, expedition=expedition, defaults={'logbook_entry':placeholder})[0] except Survey.MultipleObjectsReturned: survey=Survey.objects.filter(wallet_number=surveyNumber, wallet_letter=surveyLetter, expedition=expedition)[0] file_=os.path.join(yearPath, surveyFolder, scan) scanObj = ScannedImage( file=file_, contents=scanType, number_in_wallet=scanNumber, survey=survey, new_since_parsing=False, ) print("Added scanned image at " + str(scanObj)) #if scanFormat=="png": #if isInterlacedPNG(os.path.join(settings.SURVEY_SCANS, "surveyscans", file_)): # print file_+ " is an interlaced PNG. No can do." #continue scanObj.save() except (IOError, OSError): yearPath=os.path.join(settings.SURVEY_SCANS, "surveyscans", expedition.year) print("No folder found for " + expedition.year + " at:- " + yearPath) # dead def parseSurveys(logfile=None): try: readSurveysFromCSV() except (IOError, OSError): print("Survey CSV not found..") pass for expedition in Expedition.objects.filter(year__gte=2000): #expos since 2000, because paths and filenames were nonstandard before then parseSurveyScans(expedition) # dead def isInterlacedPNG(filePath): #We need to check for interlaced PNGs because the thumbnail engine can't handle them (uses PIL) file=Image.open(filePath) print(filePath) if 'interlace' in file.info: return file.info['interlace'] else: return False # handles url or file, so we can refer to a set of scans on another server def GetListDir(sdir): res = [ ] if sdir[:7] == "http://": assert False, "Not written" s = urllib.urlopen(sdir) else: for f in os.listdir(sdir): if f[0] != ".": ff = os.path.join(sdir, f) res.append((f, ff, os.path.isdir(ff))) return res def LoadListScansFile(survexscansfolder): gld = [ ] # flatten out any directories in these book files for (fyf, ffyf, fisdiryf) in GetListDir(survexscansfolder.fpath): if fisdiryf: gld.extend(GetListDir(ffyf)) else: gld.append((fyf, ffyf, fisdiryf)) for (fyf, ffyf, fisdiryf) in gld: #assert not fisdiryf, ffyf if re.search(r"\.(?:png|jpg|jpeg)(?i)$", fyf): survexscansingle = SurvexScanSingle(ffile=ffyf, name=fyf, survexscansfolder=survexscansfolder) survexscansingle.save() # this iterates through the scans directories (either here or on the remote server) # and builds up the models we can access later def LoadListScans(): print('Loading Survey Scans...') SurvexScanSingle.objects.all().delete() SurvexScansFolder.objects.all().delete() # first do the smkhs (large kh survey scans) directory survexscansfoldersmkhs = SurvexScansFolder(fpath=os.path.join(settings.SURVEY_SCANS, "smkhs"), walletname="smkhs") if os.path.isdir(survexscansfoldersmkhs.fpath): survexscansfoldersmkhs.save() LoadListScansFile(survexscansfoldersmkhs) # iterate into the surveyscans directory for f, ff, fisdir in GetListDir(os.path.join(settings.SURVEY_SCANS, "surveyscans")): if not fisdir: continue # do the year folders if re.match(r"\d\d\d\d$", f): for fy, ffy, fisdiry in GetListDir(ff): if fisdiry: assert fisdiry, ffy survexscansfolder = SurvexScansFolder(fpath=ffy, walletname=fy) survexscansfolder.save() LoadListScansFile(survexscansfolder) # do the elif f != "thumbs": survexscansfolder = SurvexScansFolder(fpath=ff, walletname=f) survexscansfolder.save() LoadListScansFile(survexscansfolder) def FindTunnelScan(tunnelfile, path): scansfolder, scansfile = None, None mscansdir = re.search(r"(\d\d\d\d#\d+\w?|1995-96kh|92-94Surveybookkh|1991surveybook|smkhs)/(.*?(?:png|jpg))$", path) if mscansdir: scansfolderl = SurvexScansFolder.objects.filter(walletname=mscansdir.group(1)) if len(scansfolderl): assert len(scansfolderl) == 1 scansfolder = scansfolderl[0] if scansfolder: scansfilel = scansfolder.survexscansingle_set.filter(name=mscansdir.group(2)) if len(scansfilel): assert len(scansfilel) == 1 scansfile = scansfilel[0] if scansfolder: tunnelfile.survexscansfolders.add(scansfolder) if scansfile: tunnelfile.survexscans.add(scansfile) elif path and not re.search(r"\.(?:png|jpg)$(?i)", path): name = os.path.split(path)[1] print("ttt", tunnelfile.tunnelpath, path, name) rtunnelfilel = TunnelFile.objects.filter(tunnelname=name) if len(rtunnelfilel): assert len(rtunnelfilel) == 1, ("two paths with name of", path, "need more discrimination coded") rtunnelfile = rtunnelfilel[0] #print "ttt", tunnelfile.tunnelpath, path, name, rtunnelfile.tunnelpath tunnelfile.tunnelcontains.add(rtunnelfile) tunnelfile.save() def SetTunnelfileInfo(tunnelfile): ff = os.path.join(settings.TUNNEL_DATA, tunnelfile.tunnelpath) tunnelfile.filesize = os.stat(ff)[stat.ST_SIZE] fin = open(ff) ttext = fin.read() fin.close() mtype = re.search("<(fontcolours|sketch)", ttext) assert mtype, ff tunnelfile.bfontcolours = (mtype.group(1)=="fontcolours") tunnelfile.npaths = len(re.findall(" # for path, style in re.findall('