%PDF- %PDF-
Direktori : /usr/share/hplip/prnt/ |
Current File : //usr/share/hplip/prnt/cups.py |
# -*- coding: utf-8 -*- # # (c) Copyright 2003-2015 HP Development Company, L.P. # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by # the Free Software Foundation; either version 2 of the License, or # (at your option) any later version. # # 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 General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA # # Author: Don Welch # # Std Lib import os import os.path import gzip import re import time import tempfile import glob # Local from base.g import * from base import utils, models, os_utils from base.sixext import PY3 INVALID_PRINTER_NAME_CHARS = """~`!@#$%^&*()=+[]{}()\\/,.<>?'\";:| """ # Handle case where cups.py (via device.py) is loaded # and cupsext doesn't exist yet. This happens in the # installer and in a fresh sandbox if the Python extensions # aren't installed yet. try: current_language = os.getenv("LANG") newlang = "C" # this is a workaround due CUPS rejecting all encoding except ASCII # and utf-8 # if the locale contains the encoding, switch to the same locale, # but with utf-8 encoding. Otherwise use C locale. if current_language is not None and current_language.count('.'): newlang, encoding = current_language.split('.') newlang += ".UTF-8" os.environ['LANG'] = newlang cupsext = utils.import_ext('cupsext') # restore the old env values if current_language is not None: os.environ['LANG'] = current_language except ImportError: if not os.getenv("HPLIP_BUILD"): log.warn("CUPSEXT could not be loaded. Please check HPLIP installation.") sys.exit(1) IPP_PRINTER_STATE_IDLE = 3 IPP_PRINTER_STATE_PROCESSING = 4 IPP_PRINTER_STATE_STOPPED = 5 # Std CUPS option types PPD_UI_BOOLEAN = 0 # True or False option PPD_UI_PICKONE = 1 # Pick one from a list PPD_UI_PICKMANY = 2 # Pick zero or more from a list # Non-std: General UI_SPINNER = 100 # Simple spinner with opt. suffix (ie, %) UI_UNITS_SPINNER = 101 # Spinner control w/pts, cm, in, etc. units (not impl.) UI_BANNER_JOB_SHEETS = 102 # dual combos for banner job-sheets UI_PAGE_RANGE = 103 # Radio + page range entry field # Non-std: Job storage UI_JOB_STORAGE_MODE = 104 # Combo w/linkage UI_JOB_STORAGE_PIN = 105 # Radios w/PIN entry UI_JOB_STORAGE_USERNAME = 106 # Radios w/text entry UI_JOB_STORAGE_ID = 107 # Radios w/text entry UI_JOB_STORAGE_ID_EXISTS = 108 # Combo UI_INFO = 109 # Information field, required Information name and Value # ipp_op_t IPP_PAUSE_PRINTER = 0x0010 IPP_RESUME_PRINTER = 0x011 IPP_PURGE_JOBS = 0x012 CUPS_GET_DEFAULT = 0x4001 CUPS_GET_PRINTERS = 0x4002 CUPS_ADD_MODIFY_PRINTER = 0x4003 CUPS_DELETE_PRINTER = 0x4004 CUPS_GET_CLASSES = 0x4005 CUPS_ADD_MODIFY_CLASS = 0x4006 CUPS_DELETE_CLASS = 0x4007 CUPS_ACCEPT_JOBS = 0x4008 CUPS_REJECT_JOBS = 0x4009 CUPS_SET_DEFAULT = 0x400a CUPS_GET_DEVICES = 0x400b CUPS_GET_PPDS = 0x400c CUPS_MOVE_JOB = 0x400d CUPS_AUTHENTICATE_JOB = 0x400e # ipp_jstate_t IPP_JOB_PENDING = 3 # Job is waiting to be printed IPP_JOB_HELD = 4 # Job is held for printing IPP_JOB_PROCESSING = 5 # Job is currently printing IPP_JOB_STOPPED = 6 # Job has been stopped IPP_JOB_CANCELLED = 7 # Job has been cancelled IPP_JOB_ABORTED = 8 # Job has aborted due to error IPP_JOB_COMPLETED = 8 # Job has completed successfully # ipp_status_e IPP_OK = 0x0000 # successful-ok IPP_OK_SUBST = 0x001 # successful-ok-ignored-or-substituted-attributes IPP_OK_CONFLICT = 0x002 # successful-ok-conflicting-attributes IPP_OK_IGNORED_SUBSCRIPTIONS = 0x003 # successful-ok-ignored-subscriptions IPP_OK_IGNORED_NOTIFICATIONS = 0x004 # successful-ok-ignored-notifications IPP_OK_TOO_MANY_EVENTS = 0x005 # successful-ok-too-many-events IPP_OK_BUT_CANCEL_SUBSCRIPTION = 0x006 # successful-ok-but-cancel-subscription IPP_OK_EVENTS_COMPLETE = 0x007 # successful-ok-events-complete IPP_REDIRECTION_OTHER_SITE = 0x300 IPP_BAD_REQUEST = 0x0400 # client-error-bad-request IPP_FORBIDDEN = 0x0401 # client-error-forbidden IPP_NOT_AUTHENTICATED = 0x0402 # client-error-not-authenticated IPP_NOT_AUTHORIZED = 0x0403 # client-error-not-authorized IPP_NOT_POSSIBLE = 0x0404 # client-error-not-possible IPP_TIMEOUT = 0x0405 # client-error-timeout IPP_NOT_FOUND = 0x0406 # client-error-not-found IPP_GONE = 0x0407 # client-error-gone IPP_REQUEST_ENTITY = 0x0408 # client-error-request-entity-too-large IPP_REQUEST_VALUE = 0x0409 # client-error-request-value-too-long IPP_DOCUMENT_FORMAT = 0x040a # client-error-document-format-not-supported IPP_ATTRIBUTES = 0x040b # client-error-attributes-or-values-not-supported IPP_URI_SCHEME = 0x040c # client-error-uri-scheme-not-supported IPP_CHARSET = 0x040d # client-error-charset-not-supported IPP_CONFLICT = 0x040e # client-error-conflicting-attributes IPP_COMPRESSION_NOT_SUPPORTED = 0x040f # client-error-compression-not-supported IPP_COMPRESSION_ERROR = 0x0410 # client-error-compression-error IPP_DOCUMENT_FORMAT_ERROR = 0x0411 # client-error-document-format-error IPP_DOCUMENT_ACCESS_ERROR = 0x0412 # client-error-document-access-error IPP_ATTRIBUTES_NOT_SETTABLE = 0x0413 # client-error-attributes-not-settable IPP_IGNORED_ALL_SUBSCRIPTIONS = 0x0414 # client-error-ignored-all-subscriptions IPP_TOO_MANY_SUBSCRIPTIONS = 0x0415 # client-error-too-many-subscriptions IPP_IGNORED_ALL_NOTIFICATIONS = 0x0416 # client-error-ignored-all-notifications IPP_PRINT_SUPPORT_FILE_NOT_FOUND = 0x0417 # client-error-print-support-file-not-found IPP_INTERNAL_ERROR = 0x0500 # server-error-internal-error IPP_OPERATION_NOT_SUPPORTED = 0x0501 # server-error-operation-not-supported IPP_SERVICE_UNAVAILABLE = 0x0502 # server-error-service-unavailable IPP_VERSION_NOT_SUPPORTED = 0x0503 # server-error-version-not-supported IPP_DEVICE_ERROR = 0x0504 # server-error-device-error IPP_TEMPORARY_ERROR = 0x0505 # server-error-temporary-error IPP_NOT_ACCEPTING = 0x0506 # server-error-not-accepting-jobs IPP_PRINTER_BUSY = 0x0507 # server-error-busy IPP_ERROR_JOB_CANCELLED = 0x0508 # server-error-job-canceled IPP_MULTIPLE_JOBS_NOT_SUPPORTED = 0x0509 # server-error-multiple-document-jobs-not-supported IPP_PRINTER_IS_DEACTIVATED = 0x050a # server-error-printer-is-deactivated CUPS_ERROR_BAD_NAME = 0x0f00 CUPS_ERROR_BAD_PARAMETERS = 0x0f01 nickname_pat = re.compile(r'''\*NickName:\s*\"(.*)"''', re.MULTILINE) pat_cups_error_log = re.compile(r"""^loglevel\s?(debug|debug2|warn|info|error|none)""", re.I) ppd_pat = re.compile(r'''.*hp-(.*?)(-.*)*\.ppd.*''', re.I) ppd_pat1 = re.compile(r'''.*hp-(.*?)(_.*)*\.ppd.*''', re.I) def getFamilyClassName(model): models_dir = getPPDPath() m = models.ModelData() dict=m.read_all_files(False) family_type = [] for m in dict: if model in m: family_type= dict[m]['family-class'] for f in models.FAMILY_CLASSES: if f in family_type: return f def getPpdName(model): """ get ppds from models.dat for postscript printers for pclm printers forming ppd name from model name """ m = models.ModelData() dict=m.read_all_files(False) for m in dict: if model in m: print('dict[m]',dict[m]) key = 'ppd-name' if key in dict[m].keys(): # postscript name from models.dat is model-ps.ppd so adding .gz string ppd_name= dict[m]['ppd-name'] + ".gz" else: # for pclm forming ppd with model_name.ppd.gz if model.startswith("apollo"): ppd_name=model+".ppd.gz" elif model.startswith("hp_"): model1 =model.replace("hp_","hp-") ppd_name=model1+".ppd.gz" else: ppd_name="hp-"+model+".ppd.gz" else: #print("Model not present ", model) pass return ppd_name def isfamilydrv(ppds): family_check=0 #for f in ppds: # for m in models.FAMILY_CLASSES: # if m in f: # family_check=1 filename_config = "/etc/hp/hplip.conf" file_conf = open(filename_config,'r') for line in file_conf: if 'class-driver' in line: count = line.find('=') family_check_str = line[count+1:len(line)-1] if family_check_str == 'yes': family_check = 1 return family_check def getPPDPath(addtional_paths=None): """ Returns the CUPS ppd path (not the foomatic one under /usr/share/ppd). Usually this is /usr/share/cups/model. """ if addtional_paths is None: addtional_paths = [] search_paths = prop.ppd_search_path.split(';') + addtional_paths for path in search_paths: ppd_path = os.path.join(path, 'cups/model') if os.path.exists(ppd_path): return ppd_path def getPPDPath1(addtional_paths=None): """ return path for hplip ppds. """ filename_config = "/etc/hp/hplip.conf" file_conf = open(filename_config,'r') for line in file_conf: if 'ppd=' in line: count = line.find('=') ppd_path = line[count+1:len(line)-1] return ppd_path def getAllowableMIMETypes(): """ Scan all /etc/cups/*.convs and /usr/share/cups/mime files for allowable file formats. """ paths = [] allowable_mime_types = [] files = [] if os.path.exists("/etc/cups"): paths.append("/etc/cups/*.convs") if os.path.exists("/usr/share/cups/mime"): paths.append("/usr/share/cups/mime/*.convs") for path in paths: files.extend(glob.glob(path)) for f in files: #log.debug( "Capturing allowable MIME types from: %s" % f ) conv_file = open(f, 'r') for line in conv_file: if not line.startswith("#") and len(line) > 1: try: source, dest, cost, prog = line.split() except ValueError: continue if source not in ('application/octet-stream', 'application/vnd.cups-postscript'): allowable_mime_types.append(source) # Add some well-known MIME types that may not appear in the .convs files allowable_mime_types.append("image/x-bmp") allowable_mime_types.append("text/cpp") allowable_mime_types.append("application/x-python") allowable_mime_types.append("application/hplip-fax") return allowable_mime_types def getPPDDescription(f): if f.endswith('.gz'): nickname = gzip.GzipFile(f, 'r').read(4096) else: nickname = open(f, 'r').read(4096) try: desc = nickname_pat.search(nickname.decode('utf-8')).group(1) except AttributeError: desc = '' return desc def getSystemPPDs(): major, minor, patch = getVersionTuple() ppds = {} # {'ppd name' : 'desc', ...} if major == 1 and minor < 2: ppd_dir = sys_conf.get('dirs', 'ppd') log.debug("(CUPS 1.1.x) Searching for PPDs in: %s" % ppd_dir) for f in utils.walkFiles(ppd_dir, pattern="HP*ppd*;hp*ppd*", abs_paths=True): desc = getPPDDescription(f) if not ('foo2' in desc or 'gutenprint' in desc.lower() or 'gutenprint' in f): ppds[f] = desc log.debug("%s: %s" % (f, desc)) else: # 1.2.x log.debug("(CUPS 1.2.x) Getting list of PPDs using CUPS_GET_PPDS...") ppd_dict = cupsext.getPPDList() cups_ppd_path = getPPDPath() # usually /usr/share/cups/model foomatic_ppd_path = sys_conf.get('dirs', 'ppdbase', '/usr/share/ppd') if not foomatic_ppd_path or not os.path.exists(foomatic_ppd_path): foomatic_ppd_path = '/usr/share/ppd' log.debug("CUPS PPD base path = %s" % cups_ppd_path) log.debug("Foomatic PPD base path = %s" % foomatic_ppd_path) for ppd in ppd_dict: if not ppd: continue if 'hp-' in ppd.lower() or 'hp_' in ppd.lower() and \ ppd_dict[ppd]['ppd-make'] == 'HP': desc = ppd_dict[ppd]['ppd-make-and-model'] if not ('foo2' in desc.lower() or 'gutenprint' in desc.lower() or 'gutenprint' in ppd): # PPD files returned by CUPS_GET_PPDS (and by lpinfo -m) # can be relative to /usr/share/ppd/ or to # /usr/share/cups/model/. Not sure why this is. # Here we will try both and see which one it is... if os.path.exists(ppd): path = ppd else: try: path = os.path.join(foomatic_ppd_path, ppd) except AttributeError: # happens on some boxes with provider: style ppds (foomatic: etc) path = ppd else: if not os.path.exists(path): try: path = os.path.join(cups_ppd_path, ppd) except AttributeError: path = ppd else: if not os.path.exists(path): path = ppd # foomatic: or some other driver ppds[path] = desc #log.debug("%s: %s" % (path, desc)) return ppds ## TODO: Move this to CUPSEXT for better performance def levenshtein_distance(a,b): """ Calculates the Levenshtein distance between a and b. Written by Magnus Lie Hetland. """ n, m = len(a), len(b) if n > m: a,b = b,a n,m = m,n current = list(range(n+1)) for i in range(1,m+1): previous, current = current, [i]+[0]*m for j in range(1,n+1): add, delete = previous[j]+1, current[j-1]+1 change = previous[j-1] if a[j-1] != b[i-1]: change = change + 1 current[j] = min(add, delete, change) return current[n] number_pat = re.compile(r""".*?(\d+)""", re.IGNORECASE) STRIP_STRINGS2 = ['foomatic:', 'hp-', 'hp_', 'hp ', '.gz', '.ppd', 'drv:', '-pcl', '-pcl3', '-jetready', '-zxs', '-zjs', '-ps', '-postscript', '-pdf', '-jr', '-lidl', '-lidil', '-ldl', '-hpijs'] for p in list(models.TECH_CLASS_PDLS.values()): pp = '-%s' % p if pp not in STRIP_STRINGS2: STRIP_STRINGS2.append(pp) STRIP_STRINGS = STRIP_STRINGS2[:] STRIP_STRINGS.extend(['-series', ' series', '_series']) def stripModel2(model): # For new 2.8.10+ PPD find algorithm model = model.lower() for x in STRIP_STRINGS2: model = model.replace(x, '') return model def stripModel(model): # for old PPD find algorithm (removes "series" as well) model = model.lower() for x in STRIP_STRINGS: model = model.replace(x, '') return model def getPPDFile(stripped_model, ppds): # Old PPD find """ Match up a model name to a PPD from a list of system PPD files. """ log.debug("1st stage edit distance match") mins = {} eds = {} min_edit_distance = sys.maxsize log.debug("Determining edit distance from %s (only showing edit distances < 4)..." % stripped_model) for f in ppds: t = stripModel(os.path.basename(f)) eds[f] = levenshtein_distance(stripped_model, t) if eds[f] < 4: log.debug("dist('%s') = %d" % (t, eds[f])) min_edit_distance = min(min_edit_distance, eds[f]) log.debug("Min. dist = %d" % min_edit_distance) for f in ppds: if eds[f] == min_edit_distance: for m in mins: if os.path.basename(m) == os.path.basename(f): break # File already in list possibly with different path (Ubuntu, etc) else: mins[f] = ppds[f] log.debug(mins) if len(mins) > 1: # try pattern matching the model number log.debug("2nd stage matching with model number") try: model_number = number_pat.match(stripped_model).group(1) model_number = int(model_number) except AttributeError: pass except ValueError: pass else: log.debug("model_number=%d" % model_number) matches = {} #[] for x in range(3): # 1, 10, 100 factor = 10**x log.debug("Factor = %d" % factor) adj_model_number = int(model_number/factor)*factor number_matching, match = 0, '' for m in mins: try: mins_model_number = number_pat.match(os.path.basename(m)).group(1) mins_model_number = int(mins_model_number) log.debug("mins_model_number= %d" % mins_model_number) except AttributeError: continue except ValueError: continue mins_adj_model_number = int(mins_model_number/factor)*factor log.debug("mins_adj_model_number=%d" % mins_adj_model_number) log.debug("adj_model_number=%d" % adj_model_number) if mins_adj_model_number == adj_model_number: log.debug("match") number_matching += 1 matches[m] = ppds[m] log.debug(matches) log.debug("***") if len(matches): mins = matches break return mins def getPPDFile2(mq,model, ppds): # New PPD find # This routine is for the new PPD naming scheme begun in 2.8.10 # and beginning with implementation in 2.8.12 (Qt4 hp-setup) # hp-<model name from models.dat w/o beginning hp_>[-<pdl>][-<pdl>][...].ppd[.gz] # 3.9.6: Added handling for hpijs vs. hpcups PPDs/DRVs #Check if common ppd name is already given in models.dat(This is needed because in case of devices having more than one derivatives #will have diffrent model name strings in device ID, because of which we don't get the common ppd name for search) family_check=isfamilydrv(ppds) family_class=getFamilyClassName(model) model = models.normalizeModelName(model) if family_check==0: ppd_name = mq.get('ppd-name',0) else: ppd_name = mq.get('family-ppd',0) if ppd_name == 0: stripped_model = stripModel2(model) else: stripped_model = stripModel2(ppd_name) log.debug("Matching PPD list to model %s..." % stripped_model) matches = [] if family_check ==0 : for f in ppds: match = ppd_pat.match(f) if match is not None: if match.group(1) == stripped_model: log.debug("Found match: %s" % f) try: pdls = match.group(2).split('-') except AttributeError: pdls = [] if (prop.hpcups_build and 'hpijs' not in f) or \ ((prop.hpijs_build and 'hpijs' in pdls) or (prop.hpcups_build and 'hpijs' not in pdls)) or \ ('ps' in pdls) or ('pdf' in pdls): matches.append((f, [p for p in pdls if p and p != 'hpijs'])) else: for f in ppds: match = ppd_pat1.match(f) if match is not None: if match.group(1) == family_class: log.debug("Found match: %s" % f) try: pdls = match.group(2).split('-') except AttributeError: pdls = [] if (prop.hpcups_build and 'hpijs' not in f) or \ ((prop.hpijs_build and 'hpijs' in pdls) or (prop.hpcups_build and 'hpijs' not in pdls)) or \ ('ps' in pdls) or ('pdf' in pdls): matches.append((f, [p for p in pdls if p and p != 'hpijs'])) log.debug(matches) num_matches = len(matches) if num_matches == 0: log.debug("No PPD found for model %s using new algorithm. Trying old algorithm..." % stripped_model) #Using Old algo, ignores the series keyword in ppd searching. matches2 = list(getPPDFile(stripModel(stripped_model), ppds).items()) log.debug(matches2) num_matches2 = len(matches2) if num_matches2: for f, d in matches2: match = ppd_pat.match(f) if match is not None: log.debug("Found match: %s" % f) try: pdls = match.group(2).split('-') except AttributeError: pdls = [] if (prop.hpcups_build and 'hpijs' not in f) or \ ((prop.hpijs_build and 'hpijs' in pdls) or (prop.hpcups_build and 'hpijs' not in pdls)) or \ ('ps' in pdls) or ('pdf' in pdls): matches.append((f, [p for p in pdls if p and p != 'hpijs'])) log.debug(matches) num_matches = len(matches) if num_matches == 0: log.error("No PPD found for model %s using old algorithm." % stripModel(stripped_model)) return None elif num_matches == 1: log.debug("One match found.") return (matches[0][0], '') # > 1 log.debug("%d matches found. Searching based on PDL: Host > PS,PDF > PCL/Other" % num_matches) for p in [models.PDL_TYPE_HOST, models.PDL_TYPE_PS,models.PDL_TYPE_PDF, models.PDL_TYPE_PCL]: for f, pdl_list in matches: for x in pdl_list: # default to HOST-based PDLs, as newly supported PDLs will most likely be of this type if models.PDL_TYPES.get(x, models.PDL_TYPE_HOST) == p: log.debug("Selecting '-%s' PPD: %s" % (x, f)) return (f, '') log.debug("%d matches found. Searching based on Filters: HPCUPS > HPIJS" % num_matches) for p in ["hpcups","hpijs"]: for f, pdl_list in matches: if p in f: log.debug("Selecting PPD: %s" % (f)) return (f, '') # No specific PDL or Filter found, so just return 1st found PPD file log.debug("No specific PDL located. Defaulting to first found PPD file.") return (matches[0][0], '') ## # Function :- getFaxPPDFile() # Arguments:- # 1) mq --> Device model query object # 2) model --> Fax model name # Return arguments:- # 1) fax_ppd --> Found Fax ppd file. (Returns None if not found) # 2) expt_fax_ppd_name -> Expected Fax PPD name # 3) nick --> Expected Fax PPD description # def getFaxPPDFile(mq, model): try: fax_ppd = None nick = "HP Fax hpcups" expected_fax_ppd_name = "HP-Fax-hpcups" log.debug("Searching for fax PPD for model %s hpcups_build =%d" % (model,prop.hpcups_build)) if prop.hpcups_build: faxType = mq.get('fax-type', FAX_TYPE_NONE) if faxType == FAX_TYPE_MARVELL: expected_fax_ppd_name = "HP-Fax3-hpcups" # Fixed width (2528 pixels) and 300dpi rendering nick = "HP Fax3 hpcups" elif faxType == FAX_TYPE_SOAP or faxType == FAX_TYPE_LEDMSOAP: expected_fax_ppd_name = "HP-Fax2-hpcups" # Fixed width (2528 pixels) and 300dpi rendering nick = "HP Fax2 hpcups" elif faxType == FAX_TYPE_LEDM: expected_fax_ppd_name = "HP-Fax4-hpcups"# Fixed width (2528 pixels) and 300dpi rendering nick = "HP Fax4 hpcups" elif faxType == FAX_TYPE_CDM: expected_fax_ppd_name = "HP-Fax-CDM-hpcups"# Fixed width (2528 pixels) and 300dpi rendering nick = "HP Fax CDM hpcups" else: expected_fax_ppd_name = "HP-Fax-hpcups" # Standard nick = "HP Fax hpcups" else: # hpijs if mq.get('fax-type', FAX_TYPE_NONE) == FAX_TYPE_MARVELL: expected_fax_ppd_name = "HP-Fax3-hpijs" # Fixed width (2528 pixels) and 300dpi rendering nick = "HP Fax3 hpijs" if mq.get('fax-type', FAX_TYPE_NONE) == FAX_TYPE_SOAP or mq.get('fax-type', FAX_TYPE_NONE) == FAX_TYPE_LEDMSOAP: expected_fax_ppd_name = "HP-Fax2-hpijs" # Fixed width (2528 pixels) and 300dpi rendering nick = "HP Fax2 hpijs" if mq.get('fax-type', FAX_TYPE_NONE) == FAX_TYPE_LEDM: expected_fax_ppd_name = "HP-Fax4-hpijs" # Fixed width (2528 pixels) and 300dpi rendering nick = "HP Fax4 hpijs" else: expected_fax_ppd_name = "HP-Fax-hpijs" # Standard nick = "HP Fax hpijs" ppds = [] for f in utils.walkFiles(sys_conf.get('dirs', 'ppd'), pattern="HP-Fax*.ppd*", abs_paths=True): ppds.append(f) log.debug("ppds=%s"%ppds) for f in ppds: if f.find(expected_fax_ppd_name) >= 0 and getPPDDescription(f) == nick: fax_ppd = f log.debug("Found fax PPD: %s" % f) break else: log.error("Unable to locate the HPLIP Fax PPD file: %s.ppd.gz file."%expected_fax_ppd_name) finally: return fax_ppd,expected_fax_ppd_name, nick def getErrorLogLevel(): cups_conf = '/etc/cups/cupsd.conf' try: f = open(cups_conf, 'r') except OSError: log.error("%s not found." % cups_conf) except IOError: log.error("%s: I/O error." % cups_conf) else: for l in f: m = pat_cups_error_log.match(l) if m is not None: level = m.group(1).lower() log.debug("CUPS error_log LogLevel: %s" % level) return level log.debug("CUPS error_log LogLevel: unknown") return 'unknown' def getPrintJobErrorLog(job_id, max_lines=1000, cont_interval=5): ret = [] s = '[Job %d]' % job_id #level = getErrorLogLevel() cups_conf = '/var/log/cups/error_log' #if level in ('debug', 'debug2'): if 1: try: f = open(cups_conf, 'r') except (IOError, OSError): log.error("Could not open the CUPS error_log file: %s" % cups_conf) return '' else: if s in open(cups_conf, 'r').read(): queue = utils.Queue() job_found = False while True: line = f.readline() if s in line: job_found = True while len(queue): ret.append(queue.get()) ret.append(line.strip()) if len(ret) > max_lines: break else: if job_found: queue.put(line.strip()) if len(queue) > cont_interval: break return '\n'.join(ret) # # cupsext wrappers # def getDefaultPrinter(): r = cupsext.getDefaultPrinter() if r is None: log.debug("The CUPS default printer is not set.") return r def setDefaultPrinter(printer_name): if PY3: printer_name = str(printer_name, "utf-8") setPasswordPrompt("You do not have permission to set the default printer. You need authentication.") return cupsext.setDefaultPrinter(printer_name) def accept(printer_name): setPasswordPrompt("You do not have permission to accept jobs on a printer queue. You need authentication.") return controlPrinter(printer_name, CUPS_ACCEPT_JOBS) def reject(printer_name): setPasswordPrompt("You do not have permission to reject jobs on a printer queue. You need authentication.") return controlPrinter(printer_name, CUPS_REJECT_JOBS) def start(printer_name): setPasswordPrompt("You do not have permission to start a printer queue. You need authentication.") return controlPrinter(printer_name, IPP_RESUME_PRINTER) def stop(printer_name): setPasswordPrompt("You do not have permission to stop a printer queue. You need authentication.") return controlPrinter(printer_name, IPP_PAUSE_PRINTER) def purge(printer_name): setPasswordPrompt("You do not have permission to purge jobs. You need authentication.") return controlPrinter(printer_name, IPP_PURGE_JOBS) def controlPrinter(printer_name, cups_op): if cups_op in (CUPS_ACCEPT_JOBS, CUPS_REJECT_JOBS, IPP_PAUSE_PRINTER, IPP_RESUME_PRINTER, IPP_PURGE_JOBS): return cupsext.controlPrinter(printer_name, cups_op) return 0; def openPPD(printer): if not printer: return return cupsext.openPPD(printer) def closePPD(): return cupsext.closePPD() def getPPD(printer): if not printer: return return cupsext.getPPD(printer) def getPPDOption(option): return cupsext.getPPDOption(option) def getPPDPageSize(): return cupsext.getPPDPageSize() def getPrinters(): ## p2 = [] ## p = cupsext.getPrinters() ## for pp in p: ## print pp ## try: ## pn = pp.name.decode('utf-8') ## except UnicodeError: ## pass ## ## p2.append(pp) ## ## return p2 return cupsext.getPrinters() def getJobs(my_job=0, completed=0): return cupsext.getJobs(my_job, completed) def getAllJobs(my_job=0): return cupsext.getJobs(my_job, 0) + cupsext.getJobs(my_job, 1) def getVersion(): return cupsext.getVersion() def getVersionTuple(): return cupsext.getVersionTuple() def getServer(): return cupsext.getServer() def cancelJob(jobid, dest=None): setPasswordPrompt("You do not have permission to cancel a job. You need authentication.") if dest is not None: return cupsext.cancelJob(dest, jobid) else: jobs = cupsext.getJobs(0, 0) for j in jobs: if j.id == jobid: return cupsext.cancelJob(j.dest, jobid) return False def resetOptions(): return cupsext.resetOptions() def addOption(option): return cupsext.addOption(option) def getOptions(): return cupsext.getOptions() def duplicateSection(section): return cupsext.duplicateSection(section) def printFile(printer, filename, title): if os.path.exists(filename): if not PY3: printer = printer.encode('utf-8') filename = filename.encode('utf-8') title = title.encode('utf-8') return cupsext.printFileWithOptions(printer, filename, title) else: return -1 def addPrinter(printer_name, device_uri, location, ppd_file, model, info): setPasswordPrompt("You do not have permission to add a printer. You need authentication.") log.debug("addPrinter('%s', '%s', '%s', '%s', '%s', '%s')" % ( printer_name, device_uri, location, ppd_file, model, info)) if ppd_file and not os.path.exists(ppd_file): log.error("PPD file '%s' not found." % ppd_file) return (-1, "PPD file not found") return cupsext.addPrinter(printer_name, device_uri, location, ppd_file, model, info) def delPrinter(printer_name): setPasswordPrompt("You do not have permission to delete a printer. You need authentication.") return cupsext.delPrinter(printer_name) def enablePrinter(printer_name): setPasswordPrompt("You do not have permission to enable a printer. You need authentication.") cmd_full_path = utils.which('cupsenable', True) cmd= "%s %s" % (cmd_full_path, printer_name) return os_utils.execute(cmd) def getGroupList(): return cupsext.getGroupList() def getGroup(group): return cupsext.getGroup(group) def getOptionList(group): return cupsext.getOptionList(group) def getOption(group, option): return cupsext.getOption(group, option) def getChoiceList(group, option): return cupsext.getChoiceList(group, option) def getChoice(group, option, choice): return cupsext.getChoice(group, option, choice) def setOptions(): return cupsext.setOptions() def removeOption(option): return cupsext.removeOption(option) def setPasswordCallback(func): return cupsext.setPasswordCallback(func) def setPasswordPrompt(prompt): return cupsext.setPasswordPrompt(prompt) def findPPDAttribute(name, spec): return cupsext.findPPDAttribute(name, spec) def releaseCupsInstance(): return cupsext.releaseCupsInstance() def cups_operation(operation_func, mode, ui_toolkit, ui_obj, *cups_op_args): cnt = 0 while cnt < 3: cnt += 1 result, status_str = operation_func(*cups_op_args) if result != IPP_FORBIDDEN: break else: releaseCupsInstance() if cnt < 3: if mode == INTERACTIVE_MODE: log.error("Could not connect to CUPS Server due to insufficient privileges.Try with valid user") elif ui_toolkit == 'qt3': ui_obj.FailureUI("<b>Could not connect to CUPS Server due to insufficient privileges.</b><p>Try with valid user") else: from ui4 import ui_utils ui_utils.FailureUI(ui_obj, "<b>Could not connect to CUPS Server due to insufficient privileges.</b><p>Try with valid user") return result, status_str