| msg260138 - (view) |
Author: Wolfgang Richter (Wolfgang Richter) * |
Date: 2016-02-11 22:39 |
| My understanding of `sys.meta_path` is that it is supposed to allow customized loading of Python modules and packages. In fact the `importlib` machinery appears to have support for identifying packages with `__init__` files with non-standard suffixes: https://github.com/python/cpython/blob/master/Lib/importlib/_bootstrap_external.py#L645 https://github.com/python/cpython/blob/master/Lib/importlib/_bootstrap_external.py#L1233 However, I find that when I `import wolftest` inside a folder with structure: ./ /wolftest /__init__.wolf /something.wolf None of my sys.meta_path finders are called at all, and instead a namespace is returned. I was wondering why the `import` statement appears to short-circuit and not check with `sys.meta_path` handlers in this case? |
|
|
| msg260142 - (view) |
Author: Brett Cannon (brett.cannon) *  |
Date: 2016-02-11 23:01 |
| I'm at work and so I don't have access to tar on this machine ATM, so I can't look at your code example. But sys.meta_path might be one level above what you want; you might be looking for sys.path_hooks since https://docs.python.org/3/library/importlib.html#importlib.machinery.PathFinder is what searches entries on sys.path. This also means that if you don't put your finder on sys.meta_path before importlib.machinery.PathFinder then it will never pick up a directory since any directory will always be viewed as a namespace package by that meta path finder. |
|
|
| msg260145 - (view) |
Author: Wolfgang Richter (Wolfgang Richter) * |
Date: 2016-02-11 23:17 |
| Brett thanks for the very quick response. I've inserted my finder as the first element in sys.meta_path. I also overrode the @classmethod find_loader, but it doesn't appear that it's being called. |
|
|
| msg260146 - (view) |
Author: Brett Cannon (brett.cannon) * |
Date: 2016-02-11 23:26 |
| As I said, I can't look at your code unless you upload the file separately so I don't know how much help I can be. Did you check sys.modules to make sure the directory had not already been imported or sys.path_importer_cache to make sure the directory didn't already have a finder associated with it? I know sys.meta_path isn't bypassed as that's how the builtin and frozen importer work. And have you tried this in a newer version of Python? 3.4 is no longer receiving bugfixes so even if there is a problem there it won't be changed. And you should try and use https://docs.python.org/3/library/importlib.html#importlib.util.find_spec instead of find_loader(). |
|
|
| msg260147 - (view) |
Author: Wolfgang Richter (Wolfgang Richter) * |
Date: 2016-02-11 23:43 |
| Sure Brett I'll upload the code separately in an hour or two. I've overridden all the publicly exposed functions (including find_spec). I wanted to initially see if anything was passing through my finder in the first position of the meta_path. |
|
|
| msg260152 - (view) |
Author: Wolfgang Richter (Wolfgang Richter) * |
Date: 2016-02-12 00:51 |
| Here's the file, before I experimented with find_loader. If you take out all my print's, it's quite short. |
|
|
| msg260156 - (view) |
Author: Brett Cannon (brett.cannon) * |
Date: 2016-02-12 03:23 |
| I found the problem: you have a bug in your code at line 45 (the first line of WolfPathFinder.find_spec()). When you try and import a top-level package the path will always be None since __path__ doesn't exist for a top-level package. You only end up with a `path` value when there is a __path__ in the parent package. If you put your print() call before the None check it will print out that you actually did have the method called. |
|
|
| msg260182 - (view) |
Author: Wolfgang Richter (Wolfgang Richter) * |
Date: 2016-02-12 14:09 |
| Ah, sorry Brett! Thanks for spotting this. The import machinery is *really* awesome. |
|
|