Skip to content

Latest commit

 

History

History
157 lines (114 loc) · 4.89 KB

README.md

File metadata and controls

157 lines (114 loc) · 4.89 KB

Package: Repackage

Laurent Franceschetti March/June 2013 - 2018 MIT License.

Purpose

This module allows any Python program to call a non-registered package in a reliable way. With this module, you may call "non-official" repositories, including with relative paths.

CAUTION: This form is an alternative to system of relative paths for python imports (PEP 328). It allows additional flexibility, at the cost of breaking some assumptions.

Install

If you are using pip: :

pip install repackage

Otherwise, download the package and type:

python setup.py install

The problem

In Python, registered packages are called by name in import instructions, and lower directories may be treated for all purposes as packages.

Two practical problems arise: a) How to easily call unregistered packages which have been dumped in an adjacent directory? b) How to easily move python files into a sub-directory without messing up the import statements?

There are complicated issues with relative imports (see PEP366). The basic idea here is to add the source directory of the package to the lib path (thanks to a call to sys.path.append).

But the probem, is how to programmatically find the source directory, from a relative path?

Two often advocated methods to determine the path are: a. from current directory or b. from __FILE__ .

Both those methods have a flaw:

  • The first does not take into account the file where the import is made, hence will fail if the project is using more than one directory.
  • The second does not allow to delegate those operations to a module that would handle those issues (as __FILE__ is going to point now to point to the module and not the caller).

The solution

This package uses a simple strategy that is likely to work in a good range of cases: it inspects the stack to determine which file is the caller and works out the relative path from there. The only delicate part consisted in working out how many steps down the stack this is, but the answer should be invariant and can be computed both by reasoning and by trial and error (in this case: 3).

Usage

Situation 1) Moving the files into a lower directory. From the module you want to make the call, just use the following statement before the imports: :

import repackage
repackage.up()

It should work without changing the imports that were previously pointing to the upper directory.

If it's two directories up, write: :

repackage.up(2)

Situation 2) Calling a non-registered directory somewhere else (absolute or relative path): :

repackage.add("../../otherdir")

Clearly, repackage.up() would be equivalent to repackage.add("..") . I prefer the first because it is more terse and syntactically more robust.

Importing a module using the system PATH

In some cases, it is convenient to install a Python app as a command in the PATH, usually with a symlink (e.g. foo => /my/path/foo.py). This allows us to execute it, regardless of where it is installed.

The idea behind Python packages is more or less the same.

But what if the module is not part of a package? Wouldn't it be intuitive to exploit the PATH of the OS to import it?

This is possible. Supposing foois in the path, as a symlink to my/path/foo.py, then you can write:

repackage.add_path("foo")
import foo

Repackage will follow the symlinks all the way to the destination and add the proper directory to your libpath. Needless to see, it would work also if there is no symlink (note that in this case, the file in your path might be called 'foo.py')

Limitations

If at some points in the execution, you attempt to add several times the same directory to the lib path, this should remain without effect (this is a feature of sys.path.append).

This module has worked reliably for a while, so it is a beta version. The method seems robust so far, but not all ins and outs have been explored. One precaution might be to ensure that the repackaging always points to the same source directory of a package (not to subdirectories of the same package), so as to avoid possible ambiguities in the lib path. (If this really turned out to be a problem, this could be checked on the fly and a warning issued?).

If you find bugs, or even find this approach useless, essentially flawed or against the Zen of Python, I will be glad to hear about it. Similarly, if you liked it or have ideas on how to improve it, let me know.