Add AssertImportHook.is_package() method

[dahlia]

Although implementing .is_package() is optional according to PEP 302, it makes several Python codes that assume every import hook implements this method (e.g. Flask) to work well with Attest.

[SimonSapin]

@dahlia , thanks for working on this but nobody really is maintaining Attest at the moment, see #131

[bradleyayers]

I'll have a look over this.

[bradleyayers]

After reading PEP 302 (thanks for the link) it appears that your patch doesn't properly implement the spec.

It says "All three methods should raise ImportError if the module wasn't found.". If you have a look at the existing code for get_source you'll find:

try:
    (fd, fn, info), path = self._cache[name]
except KeyError:
    raise ImportError(name)

I think this pattern should be repeated for is_package.

It would also seem that we should add support for get_code, as per "… each of which should be implemented, or none at all". While were at it it's probably worth adding get_filename too.

[dahlia]

Thanks for your review. AssertImportHook.get_source() method currently returns a triple of (code, filename, newpath) instead of a source string. Should it be changed too?

[bradleyayers]

From reading the PEP, that definitely seems to be the case. I wonder why it was originally written to return (code, filename, newpath), and ideas?

[dahlia]

As I guessed from 5109379, this method was originally written for internal use, not for implementing the loader protocol, and it chose the name get_source by accident.

We can simply rename this method. Any suggestions for its new name?

[bradleyayers]

I think we could probably get away with just implementing the other methods (get_filename, etc) and refactor load_module to use those. Then we can make get_source just return the file contents as it's suppose to.

[dahlia]

Implemented get_filename(), get_code() methods, and refactored load_module(), get_source() methods.

[bradleyayers]

Instead of storing to code and returning at the end, you could just return from within each if/elif, would save a couple of lines.

[dahlia]

Actually I firstly wrote to return from within each case, but when I saw the original code it was written in such way (returning at th end), so I followed the way. Okay, I’ll change this anyway. Thank you!

[bradleyayers]

Overall I really like this, my only criticism now is the lack of tests. I'm not sure the best strategy for testing it, any thoughts?

[dahlia]

Actually I have no idea how to test it correctly and well.

[bradleyayers]

This should be raise, not return.

[bradleyayers]

The loader.get_code(fullname) method should return the code object associated with the module, or None if it's a built-in or extension module.

How should we handle this? How do we determine if a fullname refers to a built-in or extension module?

[bradleyayers]

So where should we go from here? I haven't had a chance to test this with anything that's likely to expose problems (e.g. with Flask) and I'd like to do that before merging. Have you had any great ideas on how to automated test this?

[dahlia]

I wrote a small test for get_code() and made it to return None when the name is a builtin or an extension. Is this test too naive?

[leiserfg]

I resolve this problem by adding the method is_package to the AssertImportHook in the hook.py file, this is my snippet:

    def is_package(self, name):
        try:
            info = self._cache[name][0][2]
        except KeyError:
            raise ImportError(name)
        return info[2] == imp.PKG_DIRECTORY

It's work fine with flask, please add it to the next Attest version.