mirror of
https://github.com/python/cpython
synced 2024-11-02 11:04:01 +00:00
bf95ff91f2
This reduces the size of the data segment by **300 KB** of the executable because if the modules are deep-frozen then the marshalled frozen data just wastes space. This was inspired by comment by @gvanrossum in https://github.com/python/cpython/pull/29118#issuecomment-958521863. Note: There is a new option `--deepfreeze-only` in `freeze_modules.py` to change this behavior, it is on be default to save disk space. ```console # du -s ./python before 27892 ./python # du -s ./python after 27524 ./python ``` Automerge-Triggered-By: GH:ericsnowcurrently |
||
---|---|---|
.. | ||
test | ||
bkfile.py | ||
checkextensions.py | ||
checkextensions_win32.py | ||
extensions_win32.ini | ||
flag.py | ||
freeze.py | ||
hello.py | ||
makeconfig.py | ||
makefreeze.py | ||
makemakefile.py | ||
parsesetup.py | ||
README | ||
regen_frozen.py | ||
win32.html | ||
winmakemakefile.py |
THE FREEZE SCRIPT ================= (Directions for Windows are at the end of this file.) What is Freeze? --------------- Freeze make it possible to ship arbitrary Python programs to people who don't have Python. The shipped file (called a "frozen" version of your Python program) is an executable, so this only works if your platform is compatible with that on the receiving end (this is usually a matter of having the same major operating system revision and CPU type). The shipped file contains a Python interpreter and large portions of the Python run-time. Some measures have been taken to avoid linking unneeded modules, but the resulting binary is usually not small. The Python source code of your program (and of the library modules written in Python that it uses) is not included in the binary -- instead, the compiled byte-code (the instruction stream used internally by the interpreter) is incorporated. This gives some protection of your Python source code, though not much -- a disassembler for Python byte-code is available in the standard Python library. At least someone running "strings" on your binary won't see the source. How does Freeze know which modules to include? ---------------------------------------------- Previous versions of Freeze used a pretty simple-minded algorithm to find the modules that your program uses, essentially searching for lines starting with the word "import". It was pretty easy to trick it into making mistakes, either missing valid import statements, or mistaking string literals (e.g. doc strings) for import statements. This has been remedied: Freeze now uses the regular Python parser to parse the program (and all its modules) and scans the generated byte code for IMPORT instructions. It may still be confused -- it will not know about calls to the __import__ built-in function, or about import statements constructed on the fly and executed using the 'exec' statement, and it will consider import statements even when they are unreachable (e.g. "if 0: import foobar"). This new version of Freeze also knows about Python's new package import mechanism, and uses exactly the same rules to find imported modules and packages. One exception: if you write 'from package import *', Python will look into the __all__ variable of the package to determine which modules are to be imported, while Freeze will do a directory listing. One tricky issue: Freeze assumes that the Python interpreter and environment you're using to run Freeze is the same one that would be used to run your program, which should also be the same whose sources and installed files you will learn about in the next section. In particular, your PYTHONPATH setting should be the same as for running your program locally. (Tip: if the program doesn't run when you type "python hello.py" there's little chance of getting the frozen version to run.) How do I use Freeze? -------------------- Normally, you should be able to use it as follows: python freeze.py hello.py where hello.py is your program and freeze.py is the main file of Freeze (in actuality, you'll probably specify an absolute pathname such as /usr/joe/python/Tools/freeze/freeze.py). What do I do next? ------------------ Freeze creates a number of files: frozen.c, config.c and Makefile, plus one file for each Python module that gets included named M_<module>.c. To produce the frozen version of your program, you can simply type "make". This should produce a binary file. If the filename argument to Freeze was "hello.py", the binary will be called "hello". Note: you can use the -o option to freeze to specify an alternative directory where these files are created. This makes it easier to clean up after you've shipped the frozen binary. You should invoke "make" in the given directory. Freezing Tkinter programs ------------------------- Unfortunately, it is currently not possible to freeze programs that use Tkinter without a Tcl/Tk installation. The best way to ship a frozen Tkinter program is to decide in advance where you are going to place the Tcl and Tk library files in the distributed setup, and then declare these directories in your frozen Python program using the TCL_LIBRARY, TK_LIBRARY and TIX_LIBRARY environment variables. For example, assume you will ship your frozen program in the directory <root>/bin/windows-x86 and will place your Tcl library files in <root>/lib/tcl8.2 and your Tk library files in <root>/lib/tk8.2. Then placing the following lines in your frozen Python script before importing Tkinter or Tix would set the environment correctly for Tcl/Tk/Tix: import os import os.path RootDir = os.path.dirname(os.path.dirname(os.getcwd())) import sys if sys.platform == "win32": sys.path = ['', '..\\..\\lib\\python-2.0'] os.environ['TCL_LIBRARY'] = RootDir + '\\lib\\tcl8.2' os.environ['TK_LIBRARY'] = RootDir + '\\lib\\tk8.2' os.environ['TIX_LIBRARY'] = RootDir + '\\lib\\tix8.1' elif sys.platform == "linux2": sys.path = ['', '../../lib/python-2.0'] os.environ['TCL_LIBRARY'] = RootDir + '/lib/tcl8.2' os.environ['TK_LIBRARY'] = RootDir + '/lib/tk8.2' os.environ['TIX_LIBRARY'] = RootDir + '/lib/tix8.1' elif sys.platform == "solaris": sys.path = ['', '../../lib/python-2.0'] os.environ['TCL_LIBRARY'] = RootDir + '/lib/tcl8.2' os.environ['TK_LIBRARY'] = RootDir + '/lib/tk8.2' os.environ['TIX_LIBRARY'] = RootDir + '/lib/tix8.1' This also adds <root>/lib/python-2.0 to your Python path for any Python files such as _tkinter.pyd you may need. Note that the dynamic libraries (such as tcl82.dll tk82.dll python20.dll under Windows, or libtcl8.2.so and libtcl8.2.so under Unix) are required at program load time, and are searched by the operating system loader before Python can be started. Under Windows, the environment variable PATH is consulted, and under Unix, it may be the environment variable LD_LIBRARY_PATH and/or the system shared library cache (ld.so). An additional preferred directory for finding the dynamic libraries is built into the .dll or .so files at compile time - see the LIB_RUNTIME_DIR variable in the Tcl makefile. The OS must find the dynamic libraries or your frozen program won't start. Usually I make sure that the .so or .dll files are in the same directory as the executable, but this may not be foolproof. A workaround to installing your Tcl library files with your frozen executable would be possible, in which the Tcl/Tk library files are incorporated in a frozen Python module as string literals and written to a temporary location when the program runs; this is currently left as an exercise for the reader. An easier approach is to freeze the Tcl/Tk/Tix code into the dynamic libraries using the Tcl ET code, or the Tix Stand-Alone-Module code. Of course, you can also simply require that Tcl/Tk is required on the target installation, but be careful that the version corresponds. There are some caveats using frozen Tkinter applications: Under Windows if you use the -s windows option, writing to stdout or stderr is an error. The Tcl [info nameofexecutable] will be set to where the program was frozen, not where it is run from. The global variables argc and argv do not exist. A warning about shared library modules -------------------------------------- When your Python installation uses shared library modules such as _tkinter.pyd, these will not be incorporated in the frozen program. Again, the frozen program will work when you test it, but it won't work when you ship it to a site without a Python installation. Freeze prints a warning when this is the case at the end of the freezing process: Warning: unknown modules remain: ... When this occurs, the best thing to do is usually to rebuild Python using static linking only. Or use the approach described in the previous section to declare a library path using sys.path, and place the modules such as _tkinter.pyd there. Troubleshooting --------------- If you have trouble using Freeze for a large program, it's probably best to start playing with a really simple program first (like the file hello.py). If you can't get that to work there's something fundamentally wrong -- perhaps you haven't installed Python. To do a proper install, you should do "make install" in the Python root directory. Usage under Windows 95 or NT ---------------------------- Under Windows 95 or NT, you *must* use the -p option and point it to the top of the Python source tree. WARNING: the resulting executable is not self-contained; it requires the Python DLL, currently PYTHON20.DLL (it does not require the standard library of .py files though). It may also require one or more extension modules loaded from .DLL or .PYD files; the module names are printed in the warning message about remaining unknown modules. The driver script generates a Makefile that works with the Microsoft command line C compiler (CL). To compile, run "nmake"; this will build a target "hello.exe" if the source was "hello.py". Only the files frozenmain.c and frozen.c are used; no config.c is generated or used, since the standard DLL is used. In order for this to work, you must have built Python using the VC++ (Developer Studio) 5.0 compiler. The provided project builds python20.lib in the subdirectory pcbuild\Release of thje Python source tree, and this is where the generated Makefile expects it to be. If this is not the case, you can edit the Makefile or (probably better) winmakemakefile.py (e.g., if you are using the 4.2 compiler, the python20.lib file is generated in the subdirectory vc40 of the Python source tree). It is possible to create frozen programs that don't have a console window, by specifying the option '-s windows'. See the Usage below. Usage ----- Here is a list of all of the options (taken from freeze.__doc__): usage: freeze [options...] script [module]... Options: -p prefix: This is the prefix used when you ran ``make install'' in the Python build directory. (If you never ran this, freeze won't work.) The default is whatever sys.prefix evaluates to. It can also be the top directory of the Python source tree; then -P must point to the build tree. -P exec_prefix: Like -p but this is the 'exec_prefix', used to install objects etc. The default is whatever sys.exec_prefix evaluates to, or the -p argument if given. If -p points to the Python source tree, -P must point to the build tree, if different. -e extension: A directory containing additional .o files that may be used to resolve modules. This directory should also have a Setup file describing the .o files. On Windows, the name of a .INI file describing one or more extensions is passed. More than one -e option may be given. -o dir: Directory where the output files are created; default '.'. -m: Additional arguments are module names instead of filenames. -a package=dir: Additional directories to be added to the package's __path__. Used to simulate directories added by the package at runtime (eg, by OpenGL and win32com). More than one -a option may be given for each package. -l file: Pass the file to the linker (windows only) -d: Debugging mode for the module finder. -q: Make the module finder totally quiet. -h: Print this help message. -x module Exclude the specified module. -i filename: Include a file with additional command line options. Used to prevent command lines growing beyond the capabilities of the shell/OS. All arguments specified in filename are read and the -i option replaced with the parsed params (note - quoting args in this file is NOT supported) -s subsystem: Specify the subsystem (For Windows only.); 'console' (default), 'windows', 'service' or 'com_dll' -w: Toggle Windows (NT or 95) behavior. (For debugging only -- on a win32 platform, win32 behavior is automatic.) Arguments: script: The Python script to be executed by the resulting binary. module ...: Additional Python modules (referenced by pathname) that will be included in the resulting binary. These may be .py or .pyc files. If -m is specified, these are module names that are search in the path instead. --Guido van Rossum (home page: https://www.python.org/~guido/)