homepage

When I try to build msi installer for my module that contains files with long names it crashes with the following traceback
C:\home\python\dist_msi_bug>python setup.py bdist_msi
running bdist_msi
running build
running build_scripts
installing to build\bdist.win32\msi
running install_scripts
running install_data
running install_egg_info
Removing build\bdist.win32\msi\Lib\site-packages\hello-0.1-py2.6.egg-info
Writing build\bdist.win32\msi\Lib\site-packages\hello-0.1-py2.6.egg-info
Traceback (most recent call last):
 File "setup.py", line 8, in <module>
 data_files = [(".", ["data.seq.one.dat", "data.seq.two.dat"])],
 File "c:\python26\lib\distutils\core.py", line 152, in setup
 dist.run_commands()
 File "c:\python26\lib\distutils\dist.py", line 975, in run_commands
 self.run_command(cmd)
 File "c:\python26\lib\distutils\dist.py", line 995, in run_command
 cmd_obj.run()
 File "c:\python26\lib\distutils\command\bdist_msi.py", line 242, in run
 self.add_files()
 File "c:\python26\lib\distutils\command\bdist_msi.py", line 270, in add_files
 key = dir.add_file(file)
 File "c:\python26\lib\msilib\__init__.py", line 340, in add_file
 short = self.make_short(file)
 File "c:\python26\lib\msilib\__init__.py", line 299, in make_short
 assert file not in self.short_names
AssertionError
It seems that Directory.make_short fails to create unique short names in some cases. The test files are attached.

I can confirm this issue. It prevents building a IPython msi installer on Python 2.6 for Windows. 
A patch to the Directory.make_short function in msilib\__init__.py is attached. It falls back to generating prefix~pos filenames when a short name is already taken and also generates short names more similar to Windows/NTFS.

This looks a lot like bug 1128, too. I think the patch there would also fix this one.

Creating bdist_msi installers with files such as 'aixc++' (containing '+'), '.buildinfo' (starting with '.'), and 'py.~1.5.~' (containing '~') currently also fails, even with the proposed patch. 
The revised patch should fix these cases and further improves the make_short function to generate short names compatible with NTFS.

I can confirm that this issue persists in Python 3.1.2 on win32. The patch by cgohlke from 2010年10月22日 fixes the problem here as well.

New changeset e51366a05b84 by Martin v. Löwis in branch '2.7':
Fix short file name generation in bdist_msi.
http://hg.python.org/cpython/rev/e51366a05b84
New changeset fd8336947f90 by Martin v. Löwis in branch '3.1':
Fix short file name generation in bdist_msi.
http://hg.python.org/cpython/rev/fd8336947f90
New changeset c7d0fc181376 by Martin v. Löwis in branch '3.2':
merge #7639
http://hg.python.org/cpython/rev/c7d0fc181376
New changeset a7e0a1dbfbb6 by Martin v. Löwis in branch 'default':
merge #7639
http://hg.python.org/cpython/rev/a7e0a1dbfbb6 

Following up from my comment in issue1128, please find the patch msilib.make_id_fix_and_tests.patch
This improves the make_id() function in the following way:
 - ensures that NO invalid identifier characters make it through (it does this by only including good characters rather than filtering out some bad characters). The list of bad characters is immense (most unicode characters!) while the list of acceptable characters is limited.
This also adds some tests.

For Directory.make_short() the only things which are left could be considered splitting hairs.
1. Do we need to remove '"\/:' from the file name - if these characters are there then the file name is not valid in the first place.
2. These characters '+,;=[]' should really be changed to '_' if we want to match more closely how Windows makes short names.
I can write a patch for the above if they should be fixed and either way I can add tests for this function. (the test I had for testing my own patch - almost completely pass - but the only failure was depending on point 2.)