homepage

In a directory with 2 files, setup.cfg and a single C file containing source for an extension module. The same happens with a pure-python module. This is on Windows.
PS D:\Data\python-sample> D:\Data\cpython\PCbuild\python.exe -m packaging.run run --list-commands
List of available commands:
 bdist: create a built (binary) distribution
 bdist_dumb: create a "dumb" built distribution
Traceback (most recent call last):
 File "D:\Data\cpython\lib\packaging\util.py", line 652, in resolve_name
 ret = getattr(ret, part)
AttributeError: 'module' object has no attribute 'bdist_msi'
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
 File "D:\Data\cpython\lib\runpy.py", line 160, in _run_module_as_main
 "__main__", fname, loader, pkg_name)
 File "D:\Data\cpython\lib\runpy.py", line 73, in _run_code
 exec(code, run_globals)
 File "D:\Data\cpython\lib\packaging\run.py", line 666, in <module>
 sys.exit(main())
 File "D:\Data\cpython\lib\packaging\run.py", line 653, in main
 return dispatcher()
 File "D:\Data\cpython\lib\packaging\run.py", line 642, in __call__
 return func(self, self.args)
 File "D:\Data\cpython\lib\packaging\run.py", line 91, in wrapper
 return f(*args, **kwargs)
 File "D:\Data\cpython\lib\packaging\run.py", line 264, in _run
 cls = dispatcher.cmdclass.get(cmd) or get_command_class(cmd)
 File "D:\Data\cpython\lib\packaging\command\__init__.py", line 61, in get_command_class
 cls = resolve_name(cls)
 File "D:\Data\cpython\lib\packaging\util.py", line 654, in resolve_name
 raise ImportError(exc)
ImportError: 'module' object has no attribute 'bdist_msi'

Can you tell me if this works:
>>> import _msi
>>> from packaging.command.bdist_msi import bdist_msi

No it didn't - I had not built the _msi module when I built Python for some reason. I have built _msi now, and everything works. Sorry for the false alarm.
Arguably, the command shouldn't fail, it should simply omit the bdist_msi command from the listing. But as _msi is part of Python, the installation is broken if it isn't present so I guess that handling the issue gracefully isn't really important. (I assume the missing _msi extension isn't an issue on Unix).

> so I guess that handling the issue gracefully isn't really important.
Then should this issue be closed?

I suppose so, yes. But it feels symptomatic of a general lack of clean
error handling, which I think should be fixed :-(

> Arguably, the command shouldn't fail, it should simply omit the bdist_msi command from the listing.
> But as _msi is part of Python, the installation is broken if it isn't present so I guess that
> handling the issue gracefully isn't really important.
My first feeling was agreement with your request, but then you added that _msi is part of Python, so I’m not sure we should increase the code complexity for a non-issue.
> (I assume the missing _msi extension isn't an issue on Unix).
Yep, see Lib/packaging/command/__init__.py:
 # XXX this is crappy
 if os.name == 'nt':
 _COMMANDS['bdist_msi'] = 'packaging.command.bdist_msi.bdist_msi'
> it feels symptomatic of a general lack of clean error handling, which I think should be fixed :-(
I suppose the traceback would have been much more useful with #12703 fixed. (There are a number of resolve_name-like functions out there, it’s sad to see it reimplemented with various degree in the quality of error reporting over and over again.)
Besides the problem with resolve_name, I think there’s a specific problem with bdist_msi. As you can see with the comment I added in the code and quoted above, I’m not satisfied with the way we add bdist_msi to the set of available commands. It’s the only command to be special-cased; I guess it can’t be helped, so maybe I should remove the comment. I could change the conditional to register the command if _msi can be successfully imported; however, I fear this would hide bugs, as _msi is a part of a Python installation on Windows, so I’d rather continue to check os.name. Would you be satisfied with a more helpful traceback that would point you immediately to missing msi? Do you prefer that bdist_msi catch an ImportError for _msi and print a short error message instead of a traceback in all its glory?

> Would you be satisfied with a more helpful traceback that would point you immediately to missing msi? Do you prefer that bdist_msi
> catch an ImportError for _msi and print a short error message instead of a traceback in all its glory?
I'd be happy with a more explicit error. As you say, it's a broken
build so not worth worrying too much about, but it did take me quite a
while to work out what the problem was, including having to raise this
bug report. (The existing traceback doesn't even mention _msi).
So yes, a simple error would be ideal.

> The existing traceback doesn't even mention _msi.
Looks like #12703, which I’m going to fix shortly.