fpdb/pyfpdb/fpdb_import.py
Worros 497354d4f4 Make BulkImport work only with HHCs
Added new section to config for converters for ease of creating the gui.

Removed the section in the site line for filter and supported games
2009-03-24 22:58:45 +09:00

412 lines
17 KiB
Python

#!/usr/bin/python
#Copyright 2008 Steffen Jobbagy-Felso
#This program is free software: you can redistribute it and/or modify
#it under the terms of the GNU Affero General Public License as published by
#the Free Software Foundation, version 3 of the License.
#
#This program is distributed in the hope that it will be useful,
#but WITHOUT ANY WARRANTY; without even the implied warranty of
#MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
#GNU General Public License for more details.
#
#You should have received a copy of the GNU Affero General Public License
#along with this program. If not, see <http://www.gnu.org/licenses/>.
#In the "official" distribution you can find the license in
#agpl-3.0.txt in the docs folder of the package.
#see status.txt for site/games support info
# Standard Library modules
import os # todo: remove this once import_dir is in fpdb_import
import sys
from time import time
import traceback
import math
import datetime
import re
# fpdb/FreePokerTools modules
import fpdb_simple
import fpdb_db
import fpdb_parse_logic
import Configuration
# database interface modules
try:
import MySQLdb
mysqlLibFound=True
except:
pass
try:
import psycopg2
pgsqlLibFound=True
except:
pass
class Importer:
def __init__(self, caller, settings, config):
"""Constructor"""
self.settings = settings
self.caller = caller
self.config = config
self.fdb = None
self.cursor = None
self.filelist = {}
self.dirlist = {}
self.siteIds = {}
self.addToDirList = {}
self.removeFromFileList = {} # to remove deleted files
self.monitor = False
self.updated = {} #Time last import was run {file:mtime}
self.lines = None
self.faobs = None # File as one big string
self.pos_in_file = {} # dict to remember how far we have read in the file
#Set defaults
self.callHud = self.config.get_import_parameters().get("callFpdbHud")
if 'minPrint' not in self.settings:
#TODO: Is this value in the xml file?
self.settings['minPrint'] = 30
if 'handCount' not in self.settings:
#TODO: Is this value in the xml file?
self.settings['handCount'] = 0
self.fdb = fpdb_db.fpdb_db() # sets self.fdb.db self.fdb.cursor and self.fdb.sql
self.fdb.do_connect(self.config)
#Set functions
def setCallHud(self, value):
self.callHud = value
def setMinPrint(self, value):
self.settings['minPrint'] = int(value)
def setHandCount(self, value):
self.settings['handCount'] = int(value)
def setQuiet(self, value):
self.settings['quiet'] = value
def setFailOnError(self, value):
self.settings['failOnError'] = value
def setHandsInDB(self, value):
self.settings['handsInDB'] = value
def setThreads(self, value):
self.settings['threads'] = value
def setDropIndexes(self, value):
self.settings['dropIndexes'] = value
# def setWatchTime(self):
# self.updated = time()
def clearFileList(self):
self.filelist = {}
#Add an individual file to filelist
def addImportFile(self, filename, site = "default", filter = "passthrough"):
#TODO: test it is a valid file -> put that in config!!
self.filelist[filename] = [site] + [filter]
if site not in self.siteIds:
# Get id from Sites table in DB
self.fdb.cursor.execute(self.fdb.sql.query['getSiteId'], (site,))
result = self.fdb.cursor.fetchall()
if len(result) == 1:
self.siteIds[site] = result[0][0]
else:
if len(result) == 0:
print "[ERROR] Database ID for %s not found" % site
else:
print "[ERROR] More than 1 Database ID found for %s - Multiple currencies not implemented yet" % site
# Called from GuiBulkImport to add a file or directory.
def addBulkImportImportFileOrDir(self, inputPath, site = "PokerStars"):
"""Add a file or directory for bulk import"""
filter = self.config.hhcs[site].converter
# Bulk import never monitors
# if directory, add all files in it. Otherwise add single file.
# TODO: only add sane files?
if os.path.isdir(inputPath):
for subdir in os.walk(inputPath):
for file in subdir[2]:
self.addImportFile(os.path.join(inputPath, subdir[0], file), site=site, filter=filter)
else:
self.addImportFile(inputPath, site=site, filter=filter)
#Add a directory of files to filelist
#Only one import directory per site supported.
#dirlist is a hash of lists:
#dirlist{ 'PokerStars' => ["/path/to/import/", "filtername"] }
def addImportDirectory(self,dir,monitor = False, site = "default", filter = "passthrough"):
#gets called by GuiAutoImport.
#This should really be using os.walk
#http://docs.python.org/library/os.html
if os.path.isdir(dir):
if monitor == True:
self.monitor = True
self.dirlist[site] = [dir] + [filter]
for file in os.listdir(dir):
self.addImportFile(os.path.join(dir, file), site, filter)
else:
print "Warning: Attempted to add non-directory: '" + str(dir) + "' as an import directory"
def runImport(self):
""""Run full import on self.filelist."""
start = datetime.datetime.now()
print "started at", start, "--", len(self.filelist), "files to import.", self.settings['dropIndexes']
if self.settings['dropIndexes'] == 'auto':
self.settings['dropIndexes'] = self.calculate_auto()
if self.settings['dropIndexes'] == 'drop':
fpdb_simple.prepareBulkImport(self.fdb)
totstored = 0
totdups = 0
totpartial = 0
toterrors = 0
tottime = 0
# if threads <= 1: do this bit
for file in self.filelist:
(stored, duplicates, partial, errors, ttime) = self.import_file_dict(file, self.filelist[file][0], self.filelist[file][1])
totstored += stored
totdups += duplicates
totpartial += partial
toterrors += errors
tottime += ttime
if self.settings['dropIndexes'] == 'drop':
fpdb_simple.afterBulkImport(self.fdb)
fpdb_simple.analyzeDB(self.fdb)
return (totstored, totdups, totpartial, toterrors, tottime)
# else: import threaded
def calculate_auto(self):
"""An heuristic to determine a reasonable value of drop/don't drop"""
if len(self.filelist) == 1: return "don't drop"
if 'handsInDB' not in self.settings:
try:
tmpcursor = self.fdb.db.cursor()
tmpcursor.execute("Select count(1) from Hands;")
self.settings['handsInDB'] = tmpcursor.fetchone()[0]
except:
pass # if this fails we're probably doomed anyway
if self.settings['handsInDB'] < 5000: return "drop"
if len(self.filelist) < 50: return "don't drop"
if self.settings['handsInDB'] > 50000: return "don't drop"
return "drop"
#Run import on updated files, then store latest update time.
def runUpdated(self):
#Check for new files in monitored directories
#todo: make efficient - always checks for new file, should be able to use mtime of directory
# ^^ May not work on windows
for site in self.dirlist:
self.addImportDirectory(self.dirlist[site][0], False, site, self.dirlist[site][1])
for file in self.filelist:
if os.path.exists(file):
stat_info = os.stat(file)
try:
lastupdate = self.updated[file]
if stat_info.st_mtime > lastupdate:
self.import_file_dict(file, self.filelist[file][0], self.filelist[file][1])
self.updated[file] = time()
except:
self.updated[file] = time()
# If modified in the last minute run an immediate import.
# This codepath only runs first time the file is found.
if os.path.isdir(file) or (time() - stat_info.st_mtime) < 60:
# TODO attach a HHC thread to the file
# TODO import the output of the HHC thread -- this needs to wait for the HHC to block?
self.import_file_dict(file, self.filelist[file][0], self.filelist[file][1])
# TODO we also test if directory, why?
#if os.path.isdir(file):
#self.import_file_dict(file, self.filelist[file][0], self.filelist[file][1])
else:
removeFromFileList[file] = True
self.addToDirList = filter(lambda x: self.addImportDirectory(x, True, self.addToDirList[x][0], self.addToDirList[x][1]), self.addToDirList)
for file in self.removeFromFileList:
if file in self.filelist:
del self.filelist[file]
self.addToDirList = {}
self.removeFromFileList = {}
# This is now an internal function that should not be called directly.
def import_file_dict(self, file, site, filter):
if os.path.isdir(file):
self.addToDirList[file] = [site] + [filter]
return
if filter == "passthrough" or filter == "":
(stored, duplicates, partial, errors, ttime) = self.import_fpdb_file(file, site)
else:
conv = None
# Load filter, process file, pass returned filename to import_fpdb_file
print "converting %s" % file
hhbase = self.config.get_import_parameters().get("hhArchiveBase")
hhbase = os.path.expanduser(hhbase)
hhdir = os.path.join(hhbase,site)
try:
out_path = os.path.join(hhdir, file.split(os.path.sep)[-2]+"-"+os.path.basename(file))
except:
out_path = os.path.join(hhdir, "x"+strftime("%d-%m-%y")+os.path.basename(file))
filter_name = filter.replace("ToFpdb", "")
mod = __import__(filter)
obj = getattr(mod, filter_name, None)
if callable(obj):
conv = obj(in_path = file, out_path = out_path)
if(conv.getStatus()):
(stored, duplicates, partial, errors, ttime) = self.import_fpdb_file(out_path, site)
else:
# conversion didn't work
# TODO: appropriate response?
return (0, 0, 0, 1, 0)
else:
print "Unknown filter filter_name:'%s' in filter:'%s'" %(filter_name, filter)
return
#This will barf if conv.getStatus != True
return (stored, duplicates, partial, errors, ttime)
def import_fpdb_file(self, file, site):
starttime = time()
last_read_hand=0
loc = 0
if (file=="stdin"):
inputFile=sys.stdin
else:
if os.path.exists(file):
inputFile = open(file, "rU")
else:
self.removeFromFileList[file] = True
return (0, 0, 0, 1, 0)
try:
loc = self.pos_in_file[file]
except:
pass
# Read input file into class and close file
inputFile.seek(loc)
self.lines=fpdb_simple.removeTrailingEOL(inputFile.readlines())
self.pos_in_file[file] = inputFile.tell()
inputFile.close()
try: # sometimes we seem to be getting an empty self.lines, in which case, we just want to return.
firstline = self.lines[0]
except:
print "DEBUG: import_fpdb_file: failed on self.lines[0]: '%s' '%s' '%s' '%s' " %( file, site, self.lines, loc)
return (0,0,0,1,0)
if firstline.find("Tournament Summary")!=-1:
print "TODO: implement importing tournament summaries"
#self.faobs = readfile(inputFile)
#self.parseTourneyHistory()
return (0,0,0,1,0)
category=fpdb_simple.recogniseCategory(firstline)
startpos=0
stored=0 #counter
duplicates=0 #counter
partial=0 #counter
errors=0 #counter
for i in xrange (len(self.lines)):
if (len(self.lines[i])<2): #Wierd way to detect for '\r\n' or '\n'
endpos=i
hand=self.lines[startpos:endpos]
if (len(hand[0])<2):
hand=hand[1:]
if (len(hand)<3):
pass
#TODO: This is ugly - we didn't actually find the start of the
# hand with the outer loop so we test again...
else:
isTourney=fpdb_simple.isTourney(hand[0])
if not isTourney:
hand = fpdb_simple.filterAnteBlindFold(hand)
self.hand=hand
try:
handsId=fpdb_parse_logic.mainParser(self.settings['db-backend'], self.fdb.db
,self.fdb.cursor, self.siteIds[site], category, hand, self.config)
self.fdb.db.commit()
stored+=1
if self.callHud:
#print "call to HUD here. handsId:",handsId
#pipe the Hands.id out to the HUD
self.caller.pipe_to_hud.stdin.write("%s" % (handsId) + os.linesep)
except fpdb_simple.DuplicateError:
duplicates+=1
except (ValueError), fe:
errors+=1
self.printEmailErrorMessage(errors, file, hand)
if (self.settings['failOnError']):
self.fdb.db.commit() #dont remove this, in case hand processing was cancelled.
raise
except (fpdb_simple.FpdbError), fe:
errors+=1
self.printEmailErrorMessage(errors, file, hand)
self.fdb.db.rollback()
if (self.settings['failOnError']):
self.fdb.db.commit() #dont remove this, in case hand processing was cancelled.
raise
if (self.settings['minPrint']!=0):
if ((stored+duplicates+errors)%self.settings['minPrint']==0):
print "stored:", stored, "duplicates:", duplicates, "errors:", errors
if (self.settings['handCount']!=0):
if ((stored+duplicates+errors)>=self.settings['handCount']):
if (not self.settings['quiet']):
print "quitting due to reaching the amount of hands to be imported"
print "Total stored:", stored, "duplicates:", duplicates, "errors:", errors, " time:", (time() - starttime)
sys.exit(0)
startpos=endpos
ttime = time() - starttime
print "\rTotal stored:", stored, "duplicates:", duplicates, "errors:", errors, " time:", ttime
if stored==0:
if duplicates>0:
for line_no in range(len(self.lines)):
if self.lines[line_no].find("Game #")!=-1:
final_game_line=self.lines[line_no]
handsId=fpdb_simple.parseSiteHandNo(final_game_line)
else:
print "failed to read a single hand from file:", inputFile
handsId=0
#todo: this will cause return of an unstored hand number if the last hand was error
self.fdb.db.commit()
self.handsId=handsId
return (stored, duplicates, partial, errors, ttime)
def printEmailErrorMessage(self, errors, filename, line):
traceback.print_exc(file=sys.stderr)
print "Error No.",errors,", please send the hand causing this to steffen@sycamoretest.info so I can fix it."
print "Filename:", filename
print "Here is the first line so you can identify it. Please mention that the error was a ValueError:"
print self.hand[0]
print "Hand logged to hand-errors.txt"
logfile = open('hand-errors.txt', 'a')
for s in self.hand:
logfile.write(str(s) + "\n")
logfile.write("\n")
logfile.close()
if __name__ == "__main__":
print "CLI for fpdb_import is now available as CliFpdb.py"