Fix problems with import and zipimport in particular

[jeff5]

Beginning work on the failure that happens on startup of Jython. First commit is just whitespace and comment. The actual problem is that the constructor has 3 args, but is called (correctly) with just the archive path.

I can fix this by moving some path-wrangling from __new__ into a one-arg constructor. However, shouldn't __new__ be called rather than the constructor by reflection? But that's what's on sys.path_hooks.

[stuaxo]

Not sure the exact bit of code that you are talking about, but sounds like a job for *args ?

[jeff5]

@stuaxo When I write __new__ here I mean this code: https://github.com/jython/jython3/blob/master/src/org/python/modules/zipimport/PyZipImporter.java#L63.

This stuff is an example of how Jython makes Python objects. It's clever, and not terribly well explained anywhere, but the best explanation I know is here: https://wiki.python.org/jython/PythonTypesInJava.

The problem referred to is that PyZipImporter is being treated as a conventional Java class, and so the Java constructor is called (https://github.com/jython/jython3/blob/master/src/org/python/modules/zipimport/PyZipImporter.java#L55), which expects 3 arguments. And this is corrected in my latest commit here: 8ec8720#diff-990d9685668c8a38f104c60c2dac735dL147

However, this doesn't suddenly make Jython 3 work, so there's more to find.

[jeff5]

Getting quite nice conformance now with CPython as far as construction of a zipimporter goes:

> dist\bin\jython -S
Jython 3.5.1a1+ (, Jun 12 2017, 22:23:54)
[Java HotSpot(TM) 64-Bit Server VM (Oracle Corporation)] on java1.8.0_131
>>> from zipimport import zipimporter as zi
>>> archive = r'dist\Lib\test\test_importlib\namespace_pkgs\missing_directory.zip'
>>> foo = archive + r'\foo'
>>> za = zi(archive)
>>> za.archive
'dist\\Lib\\test\\test_importlib\\namespace_pkgs\\missing_directory.zip'
>>> za.prefix
''
>>> za._files
{'bar\\': ('dist\\Lib\\test\\test_importlib\\namespace_pkgs\\missing_directory.zip\\bar\\', 0, 0, 0, -1, 1336121151000, 1336121151000, 0), 'foo\\one.py': ('dist\\Lib\\test\\test_importlib\\namespace_pkgs\\missing_directory.zip\\foo\\one.py', 0, 26, 26, -1, 1336121156000, 1336121156000, 680762358), 'bar\\two.py': ('dist\\Lib\\test\\test_importlib\\namespace_pkgs\\missing_directory.zip\\bar\\two.py', 0, 35, 35, -1, 1336121151000, 1336121151000, 3195395652)}
>>> zf = zi(foo)
>>> zf.archive
'dist\\Lib\\test\\test_importlib\\namespace_pkgs\\missing_directory.zip'
>>> zf.prefix
'foo\\'
>>> zf._files is za._files
True
>>>

We copy CPython, which does not have absolute file paths in the _files cache.

Use of relative paths has generally been problematic in Jython, although it's what CPython does extensively. We have our own idea of the current directory, and some complex path manipulation using it, because it seemed Java would not pay attention to the changed value of the current working directory in property user.dir. I remember struggling with this. However, see https://stackoverflow.com/questions/840190/changing-the-current-working-directory-in-java/13981910#13981910 .

[jeff5]

There are many more things I'd like to change in PyZipImporter, but given that it long-since ceased to be what prevents running Jython, it's not the most useful thing to address. Could probably merge this and tackle something else. Prefer not to squash-merge I think, in order to make the steps/reasoning apparent to subsequent contributors.