DistributePeakApplications |
UserPreferences |
The PEAK Developers' Center | FrontPage | RecentChanges | TitleIndex | WordIndex | SiteNavigation | HelpContents |
Nearly each Python application depends on third-party Python modules. When the application is distributed, maintaining all dependencies may become a real headache. Some systems have built-in tools facilitaing this task (ports collection on *BSD systems, RPM and Debian package system on Linuxes etc.), but on Windows there seems to be no other way than include all required modules into the application distribution package.
Python application packaging may be done with several tools (like McMillian Installer or py2exe), but none of them works with PEAK out-of-the-box, because of module import tweaks used in PEAK core.
This page describes the steps taken in our company to make a distribution for PEAK-based application.
All sources listed on this page may be used under the same terms as Python or PEAK.
Currently, the application has single executable script (module) run as Windows NT service. This module uses PEAK ZConfigInterpreter to load and run all components. There are no direct references to included components in the main script.
The distribution package contains a binary wrapper for the executable script, and single subdirectory lib containing all required Python modules (both source and binary) as well as the Python dll and executable. Python modules are distributed in source form. This helps to avoid problems with PEAK lazy modules and is unlike py2exe or Installer packages including all modules as binaries.
We use McMillian Installer to build the list of module dependencies and to perform initial collecting of the module files.
The installer may be downloaded from http://mcmillan-inc.com/dnld/installer_6a1.zip or http://mcmillan-inc.com/dnld/installer_6a1.tar.gz Installer versions prior to 6a1 have some problems and cannot be used.
We use a simple c program to wrap executable python modules. This program is compiled to Windows executable stub pyrun.exe. When run, it takes Python command line options attached to the stub by utility script build_exe.py and runs Py_Main() with these options. Additional options may be specified on the executable command line; these are added to the end of "command line" passed to Py_Main.
In addition to options processing, this executable wrapper sets PYTHONPATH to the lib subdirectory, so that all modules are first looked up in the package, and not in the installed Python library (if any). site processing is disabled, and Python optimization level is set to 2 (enable basic optimizations and discard docstrings). Current directory is set to the directory of the exe file.
pyrun.c source:
/*! \file pyrun.c * * $Id: pyrun.c,v 1.1 2003/12/02 20:46:09 alex Exp $ * * Python script wrapper. * * Get command line options attached to the end of the ELF executable * and run Py_Main with altered argv. * * Python optimization level is set to 2. Site script is disabled. * * Environment variable PYTHONPATH is set to 'lib' subdirectory. * * Additional options may be passed on the command line. * * Runnable exe files are made from this stub by build_exe.py * * History (most recent first): * 02-dec-2003 [als] created */ #include <stdio.h> #include <string.h> #include <windows.h> #define FATALERROR printf #define MAGIC "ANK\014\013\012\011\0" #define INT4 int typedef struct _cookie { char magic[8]; /* 'ANK\014\013\012\013\0' */ INT4 argv; /* pos (rel to start) of argument strings */ INT4 argc; /* number of arguments */ INT4 pyvers; /* python version (e.g. 23) for dll name */ } COOKIE; int main(int argc, char **argv) { char here[_MAX_PATH + 1]; char thisfile[_MAX_PATH + 1]; char pythonpath[_MAX_PATH + 1]; char *pp; INT4 len, ii; FILE *f_fp; COOKIE f_cookie; int myargc, f_argc; char **myargv; char **arg; int rc; HINSTANCE dll; FARPROC Py_Main; /* fill in thisfile */ if (!GetModuleFileNameA(NULL, thisfile, _MAX_PATH)) { FATALERROR("System error - unable to load!"); return -1; } pp = thisfile+strlen(thisfile) - 4; if (strnicmp(pp, ".exe", 4) != 0) strcat(thisfile, ".exe"); /* fill in here (directory of thisfile) */ /* Note: GetModuleFileName returns an absolute path */ strcpy(here, thisfile); for (pp=here+strlen(here); *pp != '\\' && pp >= here+2; --pp); *++pp = '\0'; len = pp - here; /* Physically open the file */ f_fp = fopen(thisfile, "rb"); if (f_fp == NULL) { FATALERROR("Cannot open %s: %s\n", thisfile, strerror(errno)); return -1; } /* Seek to the Cookie at the end of the file. */ fseek(f_fp, 0, SEEK_END); if (fseek(f_fp, -(int)sizeof(COOKIE), SEEK_END)) { FATALERROR("Invalid executable %s\n", thisfile); return -1; } /* Read the Cookie, and check its MAGIC bytes */ fread(&f_cookie, sizeof(COOKIE), 1, f_fp); if (memcmp(f_cookie.magic, MAGIC, sizeof(MAGIC))) { FATALERROR("Bad magic value in %s\n", thisfile); return -1; } /* start filling myargv */ f_argc = f_cookie.argc; myargc = argc + f_argc + 1; /* one extra for hard-coded options */ myargv = arg = calloc(myargc, sizeof(char*)); /* FIXME: check error */ *arg = strdup(argv[0]); *(++arg) = "-SOO"; /* don't 'import site', run optimized w/o docstrings */ /* add frozen args */ fseek(f_fp, f_cookie.argv, SEEK_SET); /* FIXME: check error */ for(ii=0; ii < f_argc; ii++) { fread(&len, sizeof(INT4), 1, f_fp); /* FIXME: check error */ pp = malloc(len +1); pp[len] = '\0'; /* FIXME: check error */ fread(pp, len, 1, f_fp); /* FIXME: check error */ pp[len] = '\0'; *(++arg) = pp; }; /* close the file */ fclose(f_fp); /* add command line arguments */ for(ii=1; ii < argc; ii++) *(++arg) = strdup(argv[ii]); /* DEBUG */ /* printf("ARGS:\n"); for(ii=0; ii < myargc; ii++) printf("\t%s\n", myargv[ii]); */ /* set PYTHONPATH to 'lib' subdir */ sprintf(pythonpath, "%slib", here, here); SetEnvironmentVariableA("PYTHONPATH", pythonpath); /* enable optimizations */ /* SetEnvironmentVariableA("PYTHONOPTIMIZE", "2"); */ /* change current directory (script names may be relative) */ SetCurrentDirectory(here); /* load python DLL and get PY_Main entry */ sprintf(pythonpath, "%slib\\python%02d.dll", here, f_cookie.pyvers); dll = LoadLibraryEx(pythonpath, NULL, LOAD_WITH_ALTERED_SEARCH_PATH); if(dll == NULL) { FATALERROR("Cannot load %s (error %i)\n", pythonpath, GetLastError()); return -1; }; Py_Main = GetProcAddress(dll, "Py_Main"); /* run */ rc = Py_Main(myargc, myargv); /* FIXME? free myargv */ return rc; } /* vim: set ts=4 sts=4 sw=4 cino=(4 et : */
build_exe.py source:
1 # Build python script executable from pyrun.exe stub 2 # 3 # Copyright 2003 SIA "ANK" 4 # 5 # History (most recent first): 6 # 02-dec-2003 [als] created 7 # 8 9 import os 10 import sys 11 from struct import pack 12 13 STUB = os.path.join(os.path.dirname(__file__), "pyrun.exe") 14 MAGIC = "ANK\014\013\012\011\0" 15 16 def run(out, *args): 17 _f = open(STUB, "rb") 18 _stub = _f.read() 19 _f.close() 20 21 _f = open(out, "wb") 22 _f.write(_stub) 23 _len = _f.tell() 24 for _arg in args: 25 _f.write(pack("i", len(_arg))) 26 _f.write(_arg) 27 _f.write(MAGIC) 28 # _dll_version = sys.version_info[0] * 10 + sys.version_info[1] 29 _dll_version = int("".join(map(str, sys.version_info[:2]))) 30 _f.write(pack("iii", _len, len(args), _dll_version)) 31 _f.close() 32 33 if __name__ == "__main__": 34 run(*sys.argv[1:]) 35 36 # vim: set sts=4 sw=4 et :
As i said above, there are no direct references to included components in the main script. We have to explicitely list all components for Installer to include and analyse them. Also some modules from PEAK and standard lib are not found automatically, and need to be named explicitely.
I did this by making a modulefinder hook for the package in which the main executable module resides. This hook has hiddenimports attribute which is a list of additional module import names. The list is composed by collecting all modules in the main components package (ank.BBS.Protocol) and adding some well-known names.
hooks/hook-ank.BBS.py source:
1 # McMillian Installer hooks for BBS Server application 2 # 3 # Copyright 2003 SIA "ANK" 4 # 5 # List hidden imports for Installer Analysis: BBS Protocols, 6 # some PEAK modules, utf-8 encoding etc. 7 # 8 # History (most recent first): 9 # 02-dec-2003 [als] created 10 # 11 try: 12 import os 13 except ImportError: 14 import sys 15 os = sys.modules["os"] 16 from pprint import pprint 17 18 hiddenimports = [ 19 "encodings.utf_8", 20 "select", 21 "peak.config.load_zconfig", 22 "peak.naming.factories.peak_imports", 23 "peak.running.logs", 24 "peak.running.process", 25 "persistence.interfaces", 26 "ank.BBS.AL.PythonShell", 27 "ank.BBS.AL.WLProcessor", 28 "ank.BBS.MAC.TCP.SocketServer", 29 ] 30 31 _proto_path = ["ank", "BBS", "Protocol"] 32 for _file in os.listdir(os.path.join(*_proto_path)): 33 if _file.endswith(".py") and (_file != "__init__.py"): 34 hiddenimports.append(".".join(_proto_path + [_file[:-3]])) 35 36 for _mod in hiddenimports: 37 try: 38 __import__(_mod, globals(), locals(), ["__name__"]) 39 except Exception, _err: 40 print "Cannot import %s: %s" % (_mod, _err) 41 hiddenimports.remove(_mod) 42 43 # vim: set sts=4 sw=4 et :
Below is the spec file for McMillian Installer:
1 # McMillian Installer script for BBS Server application 2 # 3 # Copyright 2003 SIA "ANK" 4 # 5 # This script does not use any of the Installer packaging features. 6 # The Installer is used to analyse script dependencies. 7 # After that, all TOCs are altered to build the distribution tree 8 # from module sources. 9 # 10 # The scripts are wrapped into pyrun.exe. 11 # 12 # PythonService.exe is added to run the server and python.exe 13 # is added to use as poor-man's python runtime. 14 # 15 # PEAK imports are tweaked to make Analysis successfull. 16 # 17 # Distutils build directory is used to analyze and collect BBS sources 18 # 19 # History (most recent first): 20 # 02-dec-2003 [als] created 21 # 22 23 SCRIPTS = ["ank\\BBS\\BBSService.py"] 24 DIST_DIR = os.path.abspath(os.path.join(os.path.dirname(BUILDPATH), "dist")) 25 print "paths", SPECPATH, BUILDPATH, DIST_DIR 26 27 # search for modules in distutils build dir first 28 sys.path.insert(0, os.path.abspath( 29 os.path.join(SPECPATH, "..", "build", "lib.win32-2.3"))) 30 31 # load PEAK to avoid problems with lazy modules 32 from peak.api import binding, config, naming, running 33 34 # remove old build directory and distribution tree 35 if os.path.isdir(BUILDPATH): 36 shutil.rmtree(BUILDPATH) 37 os.mkdir(BUILDPATH) 38 if os.path.isdir(DIST_DIR): 39 shutil.rmtree(DIST_DIR) 40 41 # analyse dependencies 42 a = Analysis( 43 SCRIPTS, 44 hookspath=["installer\\hooks"], 45 ) 46 47 # list required modules that are not found by Analisys 48 _addmodules = [] 49 for _name in ( 50 "site", 51 "peak.binding.api", 52 "peak.running.api", 53 ): 54 _addmodules.append((_name, sys.modules[_name].__file__, "PYMODULE")) 55 56 # convert collected pure module list to the list of python sources 57 _sources = TOC() 58 for (_name, _path, _typecode) in (a.pure + _addmodules): 59 #print (_name, _path, _typecode) 60 _name = ["lib"] + _name.split(".") 61 if _path[-1] in "co": 62 _path = _path[:-1] 63 if _path.endswith("__init__.py"): 64 _name.append("__init__") 65 _sources.append((os.path.join(*_name) + ".py", _path, _typecode)) 66 # also list scripts as sources 67 for _name in SCRIPTS: 68 _sources.append((os.path.join("lib", _name), _name, "PYMODULE")) 69 70 # change binaries directory to "lib"; 71 # for inner modules, convert module paths to fs paths 72 _binaries = TOC() 73 for (_name, _path, _typecode) in a.binaries: 74 if _typecode == "LINK": 75 continue 76 _basename, _ext = os.path.splitext(_name) 77 if _ext.lower() not in [".dll", ".pyd"]: 78 _basename = _name 79 _ext = "" 80 _basename = os.path.join(*_basename.split(".")) + _ext 81 _binaries.append((_basename, _path, _typecode)) 82 # add Windows NT service runner 83 _binaries.append(("PythonService.exe", 84 config.fileNearModule("win32service", "PythonService.exe"), "BINARY")) 85 # add Python executable 86 _binaries.append(("python.exe", sys.executable, "BINARY")) 87 88 # list additional data 89 _data = TOC() 90 for _name in ( 91 "ank/BBS/app_config.xml", 92 "ank/BBS/AL/component.xml", 93 "ank/BBS/MAC/TCP/component.xml", 94 "ank/BBS/Protocol/component.xml", 95 ): 96 _data.append((os.path.join("lib", _name), _name, "DATA")) 97 _data.append((os.path.join("lib", "peak", "peak.ini"), 98 config.fileNearModule("peak", "peak.ini"), "DATA")) 99 100 # build distribution tree 101 coll = COLLECT( 102 _sources, 103 _data, 104 _binaries, 105 strip=0, 106 upx=0, 107 name=DIST_DIR 108 ) 109 110 _support_path = os.path.join(DIST_DIR, "support") 111 _lib_path = os.path.join(DIST_DIR, "lib") 112 _support_path_len = len(_support_path) 113 # move 'support' directory contents to 'lib' 114 for (_path, _dirs, _files) in os.walk(_support_path, topdown=False): 115 _newpath = _lib_path + _path[_support_path_len:] 116 for _name in _files: 117 os.rename(os.path.join(_path, _name), os.path.join(_newpath, _name)) 118 for _name in _dirs: 119 os.rmdir(os.path.join(_path, _name)) 120 os.rmdir(_support_path) 121 122 sys.path.insert(0, SPECPATH) 123 import build_exe 124 for _name in SCRIPTS: 125 _basename = os.path.splitext(os.path.basename(_name))[0] 126 _basename = os.path.join(DIST_DIR, _basename) + ".exe" 127 build_exe.run(_basename, os.path.join("lib", _name)) 128 129 # vim: set filetype=python sts=4 sw=4 et :