This is a port of Python 2.3 to OS/2 using the EMX development tools ========================================================================= What's new since the previous release ------------------------------------- This release of the port incorporates the following changes from the December 24, 2001 release of the Python 2.2 port: - based on the Python v2.3 final release source. Licenses and info about Python and EMX -------------------------------------- Please read the file README.Python-2.3 included in this package for information about Python 2.3. This file is the README file from the Python 2.3 source distribution available via http://www.python.org/ and its mirrors. The file LICENCE.Python-2.3 is the text of the Licence from the Python 2.3 source distribution. Note that the EMX package that this package depends on is released under the GNU General Public Licence. Please refer to the documentation accompanying the EMX Runtime libraries for more information about the implications of this. A copy of version 2 of the GPL is included as the file COPYING.gpl2. Readline and GDBM are covered by the GNU General Public Licence. I think Eberhard Mattes' porting changes to BSD DB v1.85 are also GPL'ed (BSD DB itself is BSD Licenced). ncurses and expat appear to be covered by MIT style licences - please refer to the source distributions for more detail. zlib is distributable under a very free license. GNU MP and GNU UFC are under the GNU LGPL (see file COPYING.lib). My patches to the Python-2.x source distributions, and any other packages used in this port, are placed in the public domain. This software is provided 'as-is', without any express or implied warranty. In no event will the author be held liable for any damages arising from the use of the software. I do hope however that it proves useful to someone. Other ports ----------- There have been ports of previous versions of Python to OS/2. The best known would be that by Jeff Rush, most recently of version 1.5.2. Jeff used IBM's Visual Age C++ (v3) for his ports, and his patches have been included in the Python 2.3 source distribution. Andrew Zabolotny implemented a port of Python v1.5.2 using the EMX development tools. His patches against the Python v1.5.2 source distribution have become the core of this port, and without his efforts this port wouldn't exist. Andrew's port also appears to have been compiled with his port of gcc 2.95.2 to EMX, which I have but have chosen not to use for the binary distribution of this port (see item 21 of the "YOU HAVE BEEN WARNED" section below). Previous Python port releases by me:- - v2.0 on March 31, 2001; - v2.0 on April 25, 2001 (cleanup release + Stackless variant); - v2.1 on June 17, 2001; - v2.0 (Stackless re-release) on June 18, 2001. - v2.1.1 on August 5, 2001; - v2.1.1 on August 12, 2001 (cleanup release); - v2.1.1 (updated DLL) on August 14, 2001. - v2.2b2 on December 8, 2001 (not uploaded to archive sites) - v2.2c1 on December 16, 2001 (not uploaded to archive sites) - v2.2 on December 24, 2001 It is possible to have these earlier ports still usable after installing this port - see the README.os2emx.multiple_versions file, contributed by Dr David Mertz, for a suggested approach to achieving this. Software requirements --------------------- This package requires the EMX Runtime package, available from the Hobbes (http://hobbes.nmsu.edu/) and LEO (http://archiv.leo.org/) archives of OS/2 software. I have used EMX version 0.9d fix04 in developing this port. My development system is running OS/2 v4 with fixpack 12. 3rd party software which has been linked into dynamically loaded modules: - ncurses (see http://dickey.his.com/ for more info, v5.2) - GNU Readline (Kai Uwe Rommel's port available from Hobbes or LEO, v2.1) - GNU GDBM (Kai Uwe Rommel's port available from Hobbes or LEO, v1.7.3) - zlib (Hung-Chi Chu's port available from Hobbes or LEO, v1.1.3) - expat (from ftp://ftp.jclark.com/pub/xml/, v1.2) - GNU MP (Peter Meerwald's port available from LEO, v2.0.2) - GNU UFC (Kai Uwe Rommel's port available from LEO, v2.0.4) The zlib module requires the Z.DLL to be installed - see the Installation section and item 12 of the "YOU HAVE BEEN WARNED" section for more information. About this port --------------- I have attempted to make this port as complete and functional as I can, notwithstanding the issues in the "YOU HAVE BEEN WARNED" section below. Core components: Python.exe is linked as an a.out executable, ie using EMX method E1 to compile & link the executable. This is so that fork() works (see "YOU HAVE BEEN WARNED" item 2). Python23.dll is created as a normal OMF DLL, with an OMF import library and module definition file. There is also an a.out (.a) import library to support linking the DLL to a.out executables. This port has been built with complete support for multithreading. Modules: As far as possible, extension modules have been made dynamically loadable when the module is intended to be built this way. I haven't yet changed the building of Python's standard modules over to using the DistUtils. See "YOU HAVE BEEN WARNED" item 5 for notes about the fcntl module, and "YOU HAVE BEEN WARNED" item 14 for notes about the pwd and grp modules. Support for case sensitive module import semantics has been added to match the Windows release. This can be deactivated by setting the PYTHONCASEOK environment variable (the value doesn't matter) - see "YOU HAVE BEEN WARNED" item 16. Optional modules: Where I've been able to locate the required 3rd party packages already ported to OS/2, I've built and included them. These include ncurses (_curses, _curses_panel), BSD DB (bsddb), GNU GDBM (gdbm, dbm), zlib (zlib), GNU Readline (readline), expat (pyexpat), GNU MP (mpz) and GNU UFC (crypt). I have built these modules statically linked against the 3rd party libraries, with the exception of zlib. Unfortunately my attempts to use the dll version of GNU readline have been a dismal failure, in that when the dynamically linked readline module is active other modules immediately provoke a core dump when imported. Only the BSD DB package (part of the BSD package distributed with EMX) needed source modifications to be used for this port, pertaining to use of errno with multithreading. The other packages, except for ncurses and zlib, needed Makefile changes for multithreading support but no source changes. The _curses_panel module is a potential problem - see "YOU HAVE BEEN WARNED" item 17. Upstream source patches: No updates to the Python 2.3 release have become available. Eberhard Mattes' EMXFIX04 update to his EMX 0.9d tools suite includes bug fixes for the BSD DB library. The bsddb module included in this port incorporates these fixes. Library and other distributed Python code: The Python standard library lives in the Lib directory. All the standard library code included with the Python 2.3 source distribution is included in the binary archive, with the exception of the dos-8x3 and tkinter subdirectories which have been omitted to reduce the size of the binary archive - the dos-8x3 components are unnecessary duplicates and Tkinter is not supported by this port (yet). All the plat-* subdirectories in the source distribution have also been omitted, and a plat-os2emx directory included. The Tools and Demo directories contain a collection of Python scripts. To reduce the size of the binary archive, the Demo/sgi, Demo/Tix, Demo/tkinter, Tools/audiopy and Tools/IDLE subdirectories have been omitted as not being supported by this port. The Misc directory has also been omitted. All subdirectories omitted from the binary archive can be reconstituted from the Python 2.3 source distribution, if desired. Support for building Python extensions: The Config subdirectory contains the files describing the configuration of the interpreter and the Makefile, import libraries for the Python DLL, and the module definition file used to create the Python DLL. The Include subdirectory contains all the standard Python header files needed for building extensions. As I don't have the Visual Age C++ compiler, I've made no attempt to have this port support extensions built with that compiler. Packaging --------- This port is packaged into several archives: - python-2.3-os2emx-bin-02????.zip (binaries, library modules) - python-2.3-os2emx-src-03????.zip (source patches and makefiles) Documentation for the Python language, as well as the Python 2.3 source distibution, can be obtained from the Python website (http://www.python.org/) or the Python project pages at Sourceforge (http://sf.net/projects/python/). Installation ------------ Obtain and install, as per the included instructions, the EMX runtime package. If you wish to use the zlib module, you will need to obtain and install the Z.DLL from Hung-Chi Chu's port of zlib v1.1.3 (zlib113.zip). See also "YOU HAVE BEEN WARNED" item 12 below. Unpack this archive, preserving the subdirectories, in the root directory of the drive where you want Python to live. Add the Python directory (eg C:\Python23) to the PATH and LIBPATH variables in CONFIG.SYS. You should then set the PYTHONHOME and PYTHONPATH environment variables in CONFIG.SYS. PYTHONHOME should be set to Python's top level directory. PYTHONPATH should be set to the semicolon separated list of principal Python library directories. I use: SET PYTHONHOME=F:/Python23 SET PYTHONPATH=F:/Python23/Lib;F:/Python23/Lib/plat-os2emx; F:/Python23/Lib/lib-dynload;F:/Python23/Lib/site-packages NOTE!: the PYTHONPATH setting above is linewrapped for this document - it should all be on one line in CONFIG.SYS! If you wish to use the curses module, you should set the TERM and TERMINFO environment variables appropriately. If you don't already have ncurses installed, I have included a copy of the EMX subset of the Terminfo database included with the ncurses-5.2 source distribution. This can be used by setting the TERMINFO environment variable to the path of the Terminfo subdirectory below the Python home directory. On my system this looks like: SET TERMINFO=F:/Python23/Terminfo For the TERM environment variable, I would try one of the following: SET TERM=ansi SET TERM=os2 SET TERM=window You will have to reboot your system for these changes to CONFIG.SYS to take effect. If you wish to compile all the included Python library modules to bytecode, you can change into the Python home directory and run the COMPILEALL.CMD batch file. You can execute the regression tests included with the Python 2.3 source distribution by changing to the Python 2.3 home directory and executing the REGRTEST.CMD batch file. The following tests are known to fail at this time: - test_longexp (see "YOU HAVE BEEN WARNED" item 1); - test_mhlib (I don't know of any port of MH to OS/2); - test_pwd (see "YOU HAVE BEEN WARNED" item 14, probably a bug in my code); - test_grp (as per test_pwd); - test_strftime (see "YOU HAVE BEEN WARNED" item 20); - test_socketserver (fork() related, see "YOU HAVE BEEN WARNED" item 2). YOU HAVE BEEN WARNED!! ---------------------- I know about a number of nasties in this port. 1. EMX's malloc() and/or the underlying OS/2 VM system aren't particularly comfortable with Python's use of heap memory. The test_longexp regression test exhausts the available swap space on a machine with 64MB of RAM with 150MB of available swap space. Using a crudely instrumented wrapper around malloc()/realloc()/free(), the heap memory usage of the expression at the core of the test (eval('[' + '2,' * NUMREPS + ']')) is as follows (approximately): NUMREPS = 1 => 300k NUMREPS = 10000 => 22MB NUMREPS = 20500 => 59MB I don't even have enough memory to try for NUMREPS = 25000 :-(, let alone the NUMREPS = 65580 in test_longexp! I do have a report that the test succeeds in the presence of sufficient memory (~200MB RAM). During the course of running the test routine, the Python parser allocates lots of 21 byte memory chunks, each of which is actually a 64 byte allocation. There are a smaller number of 3 byte allocations which consume 12 bytes each. Consequently, more than 3 times as much memory is allocated than is actually used. The Python Object Allocator code (PyMalloc) was introduced in Python 2.1 for Python's core to be able to wrap the malloc() system to deal with problems with "unfriendly" malloc() behaviour, such as this. Unfortunately for the OS/2 port, it is only supported for the allocation of memory for objects, whereas my research into this problem indicates it is the parser which is source of this particular malloc() frenzy. I have attempted using PyMalloc to manage all of Python's memory allocation. While this works fine (modulo the socket regression test failing in the absence of a socket.pyc), it is a significant performance hit - the time to run the regression test blows out from ~3.5 minutes to ~5.75 minutes on my system. I therefore don't plan to pursue this any further for the time being. Be aware that certain types of expressions could well bring your system to its knees as a result of this issue. I have modified the longexp test to report failure to highlight this. 2. Eberhard Mattes, author of EMX, writes in his documentation that fork() is very inefficient in the OS/2 environment. It also requires that the executable be linked in a.out format rather than OMF. Use the os.exec and/or the os.spawn family of functions where possible. {3. Issue resolved...} 4. In the absence of GNU Readline, terminating the interpreter requires a control-Z (^Z) followed by a carriage return. Jeff Rush documented this problem in his Python 1.5.2 port. With Readline, a control-D (^D) works as per the standard Unix environment. 5. EMX only has a partial implementation of fcntl(). The fcntl module in this port supports what EMX supports. If fcntl is important to you, please review the EMX C Library Reference (included in .INF format in the EMXVIEW.ZIP archive as part of the complete EMX development tools suite). Because of other side-effects I have modified the test_fcntl.py test script to deactivate the exercising of the missing functionality. 6. The BSD DB module is linked against DB v1.85. This version is widely known to have bugs, although some patches have become available (and are incorporated into the included bsddb module). Unless you have problems with software licenses which would rule out GDBM (and the dbm module because it is linked against the GDBM library) or need it for file format compatibility, you may be better off deleting it and relying on GDBM. I haven't looked at porting the version of the module supporting the later SleepyCat releases of BSD DB, which would also require a port of the SleepyCat DB package. 7. The readline module has been linked against ncurses rather than the termcap library supplied with EMX. {8. Workaround implemented} 9. I have configured this port to use "/" as the preferred path separator character, rather than "\" ('\\'), in line with the convention supported by EMX. Backslashes are still supported of course, and still appear in unexpected places due to outside sources that don't get normalised. 10. While the DistUtils components are now functional, other packaging/binary handling tools and utilities such as those included in the Demo and Tools directories - freeze in particular - are unlikely to work. If you do get them going, I'd like to know about your success. 11. I haven't set out to support the [BEGIN|END]LIBPATH functionality supported by one of the earlier ports (Rush's??). If it works let me know. 12. There appear to be several versions of Z.DLL floating around - the one I have is 45061 bytes and dated January 22, 1999. I have a report that another version causes SYS3175s when the zlib module is imported. 14. As a result of the limitations imposed by EMX's library routines, the standard extension module pwd only synthesises a simple passwd database, and the grp module cannot be supported at all. I have written substitutes, in Python naturally, which can process real passwd and group files for those applications (such as MailMan) that require more than EMX emulates. I have placed pwd.py and grp.py in Lib/plat-os2emx, which is usually before Lib/lib-dynload (which contains pwd.pyd) in the PYTHONPATH. If you have become attached to what pwd.pyd supports, you can put Lib/lib-dynload before Lib/plat-os2emx in PYTHONPATH or delete/rename pwd.py & grp.py. pwd.py & grp.py support locating their data files by looking in the environment for them in the following sequence: pwd.py: $ETC_PASSWD (%ETC_PASSWD%) $ETC/passwd (%ETC%/passwd) $PYTHONHOME/Etc/passwd (%PYTHONHOME%/Etc/passwd) grp.py: $ETC_GROUP (%ETC_GROUP%) $ETC/group (%ETC%/group) $PYTHONHOME/Etc/group (%PYTHONHOME%/Etc/group) Both modules support using either the ":" character (Unix standard) or ";" (OS/2, DOS, Windows standard) field separator character, and pwd.py implements the following drive letter conversions for the home_directory and shell fields (for the ":" separator only): $x -> x: x; -> x: Example versions of passwd and group are in the Etc subdirectory. Note that as of this release, this code fails the regression test. I'm looking into why, and hope to have this fixed. 15. As of Python 2.1, termios support has mutated. There is no longer a platform specific TERMIOS.py containing the symbolic constants - these now live in the termios module. EMX's termios routines don't support all of the functionality now exposed by the termios module - refer to the EMX documentation to find out what is supported. 16. The case sensitive import semantics introduced in Python 2.1 for other case insensitive but case preserving file/operating systems (Windows etc), have been incorporated into this port, and are active by default. Setting the PYTHONCASEOK environment variable (to any value) reverts to the previous (case insensitive) semantics. 17. Because I am statically linking ncurses, the _curses_panel module has potential problems arising from separate library data areas. To avoid this, I have configured the _curses_.pyd (imported as "_curses_panel") to import the ncurses symbols it needs from _curses.pyd. As a result the _curses module must be imported before the _curses_panel module. As far as I can tell, the modules in the curses package do this. If you have problems attempting to use the _curses_panel support please let me know, and I'll look into an alternative solution. 18. I tried enabling the Python Object Allocator (PYMALLOC) code. While the port built this way passes the regression test, the Numpy extension (I tested v19.0.0) as built with with the port's DistUtils code doesn't work. Specifically, attempting to "import Numeric" provokes a core dump. Supposedly Numpy v20.1.0 contains a fix for this, but for reason outlined in item 1 above, PYMALLOC is not enabled in this release. 19. sys.platform now reports "os2emx" instead of "os2". os.name still reports "os2". This change was to make it easier to distinguish between the VAC++ build (being maintained by Michael Muller) and the EMX build (this port), principally for DistUtils. 20. it appears that the %W substitution in the EMX strftime() routine has an off-by-one bug. strftime was listed as passing the regression tests in previous releases, but this fact appears to have been an oversight in the regression test suite. To fix this really requires a portable strftime routine - I'm looking into using one from FreeBSD, but its not ready yet. 21. previous releases of my Python ports have used the GCC optimisations "-O2 -fomit-frame-pointer". After experimenting with various optimisation settings, including deactivating assert()ions, I have concluded that "-O2" appears the best compromise for GCC 2.8.1 on my hardware. Curiously, deactivating assert() (via defining NDEBUG) _negatively_ impacts performance, allbeit only slightly, so I've chosen to leave the assert()s active. I did try using Andrew Zabolotny's (p)gcc 2.95.2 compiler, and in general concluded that it produced larger objects that ran slower than Mattes' gcc 2.8.1 compiler. Pystone ratings varied from just over 2000/s (no optimisation at all) to just under 3300/s (gcc 2.8.1, -O2) on my K6/2-300 system, for 100,000 iterations per run (rather than the default 10000). As a result of the optimisation change, the Python DLL is about 10% smaller than in the 2.1 release, and many of the dynamically loadable modules are smaller too. [2001/08/12] 22. As of this release, os.spawnv() and os.spawnve() now expose EMX's library routines rather than use the emulation in os.py. In order to make use of some of the features this makes available in the OS/2 environment, you should peruse the relevant EMX documentation (EMXLIB.INF in the EMXVIEW.ZIP archive accompanying the EMX archives on Hobbes or LEO). Be aware that I have exposed all the "mode" options supported by EMX, but there are combinations that either cannot be practically used by/in Python or have the potential to compromise your system's stability. 23. pythonpm.exe in previous releases was just python.exe with the WINDOWAPI linker option set in the pythonpm.def file. In practice, this turns out to do nothing useful. I have written a replacement which wraps the Python DLL in a genuine Presentation Manager application. This version actually runs the Python interpreter in a separate thread from the PM shell, in order that PythonPM has a functioning message queue as good PM apps should. In its current state, PythonPM's window is hidden. It can be displayed, although it will have no content as nothing is ever written to the window. Only the "hide" button is available. Although the code has support for shutting PythonPM down when the Python interpreter is still busy (via the "control" menu), this is not well tested and given comments I've come across in EMX documentation suggesting that the thread killing operation has problems I would suggest caution in relying on this capability. PythonPM processes commandline parameters normally. The standard input, output and error streams are only useful if redirected, as PythonPM's window is not a console in any form and so cannot accept or display anything. This means that the -i option is ineffective. Because the Python thread doesn't create its own message queue, creating PM Windows and performing most PM operations is not possible from within this thread. How this will affect supporting PM extensions (such as Tkinter using a PM port of Tcl/Tk, or wxPython using the PM port of WxWindows) is still being researched. Note that os.fork() _DOES_NOT_WORK_ in PythonPM - SYS3175s are the result of trying. os.spawnv() _does_ work. PythonPM passes all regression tests that the standard Python interpreter (python.exe) passes, with the exception of test_fork1 and test_socket which both attempt to use os.fork(). I very much want feedback on the performance, behaviour and utility of PythonPM. I would like to add a PM console capability to it, but that will be a non-trivial effort. I may be able to leverage the code in Illya Vaes' Tcl/Tk port, which would make it easier. [2001/08/14] 24. os.chdir() now uses EMX's _chdir2(), which supports changing both drive and directory at once. Similarly, os.getcwd() now uses EMX's _getcwd() which returns drive as well as path. [2001/12/08] - 2.2 Beta 2 25. pyconfig.h (previously known as config.h) is now located in the Include subdirectory with all other include files. [2001/12/16] - 2.2 Release Candidate 1 [2001/12/08] - 2.2 Final ... probably other issues that I've not encountered, or don't remember :-( If you encounter other difficulties with this port, which can be characterised as peculiar to this port rather than to the Python release, I would like to hear about them. However I cannot promise to be able to do anything to resolve such problems. See the Contact section below... To do... -------- In no particular order of apparent importance or likelihood... - support Tkinter and/or alternative GUI (wxWindows??) Credits ------- In addition to people identified above, I'd like to thank: - the BDFL, Guido van Rossum, and crew for Python; - Dr David Mertz, for trying out a pre-release of this port; - the Python-list/comp.lang.python community; - John Poltorak, for input about pwd/grp. Contact ------- Constructive feedback, negative or positive, about this port is welcome and should be addressed to me at the e-mail addresses below. I intend creating a private mailing list for announcements of fixes & updates to this port. If you wish to receive such e-mail announcments, please send me an e-mail requesting that you be added to this list. Andrew MacIntyre E-mail: andymac@bullseye.apana.org.au, or andymac@pcug.org.au Web: http://www.andymac.org/ 24 December, 2001.