changing sys.path

jmfauth wxjmfauth at
Thu Feb 2 14:45:23 CET 2012

On 2 fév, 11:03, Andrea Crotti <andrea.crott... at> wrote:
> On 02/02/2012 12:51 AM, Steven D'Aprano wrote:
> > On Wed, 01 Feb 2012 17:47:22 +0000, Andrea Crotti wrote:
> >> Yes they are exactly the same, because in that file I just write exactly
> >> the same list,
> >> but when modifying it at run-time it doesn't work, while if at the
> >> application start
> >> there is this file everything works correctly...
> >> That's what really puzzles me.. What could that be then?
> > Are you using IDLE or WingIDE or some other IDE which may not be
> > honouring sys.path? If so, that's a BAD bug in the IDE.
> > Are you changing the working directory manually, by calling os.chdir? If
> > so, that could be interfering with the import somehow. It shouldn't, but
> > you never know...
> > Are you adding absolute paths or relative paths?
> No, no and absolute paths..
> > You say that you get an ImportError, but that covers a lot of things
> > going wrong. Here's a story. Could it be correct? I can't tell because
> > you haven't posted the traceback.
> > When you set site-packages/my_paths.pth you get a sys path that looks
> > like ['a', 'b', 'fe', 'fi', 'fo', 'fum']. You then call "import spam"
> > which locates b/ and everything works.
> > But when you call sys.path.extend(['a', 'b']) you get a path that looks
> > like ['fe', 'fi', 'fo', 'fum', 'a', 'b']. Calling "import spam" locates
> > some left over junk file, fi/ or fi/spam.pyc, which doesn't
> > import, and you get an ImportError.
> And no the problem is not that I already checked inspecting at run-time..
> This is the traceback and it might be related to the fact that it runs
> from the
> .exe wrapper generated by setuptools:
> Traceback (most recent call last):
>    File "c:\python25\scripts\", line 8, in <module>
>      load_entry_point('psi.devsonly==0.1', 'console_scripts', 'dev_main')()
>    File "h:\git_projs\psi\psi.devsonly\psi\devsonly\bin\",
> line 152, in main
>      Develer(ns).full_run()
>    File "h:\git_projs\psi\psi.devsonly\psi\devsonly\bin\",
> line 86, in full_run
>      run(project_name, test_only=self.ns.test_only)
>    File "h:\git_projs\psi\psi.devsonly\psi\devsonly\",
> line 277, in run
>      from import Runner
>    File "h:\git_projs\psi\psi.devsonly\psi\devsonly\", line 7, in
> <module>
>      from psi.workbench.api import Workbench, set_new_dev_main
> ImportError: No module named workbench.api
> Another thing which might matter is that I'm launching Envisage
> applications, which
> heavily rely on the use of entry points, so I guess that if something is
> not in the path
> the entry point is not loaded automatically (but it can be forced I
> guess somehow).
> I solved in another way now, since I also need to keep a dev_main.pth in
> site-packages
> to make Eclipse happy, just respawning the same process on ImportError works
> already perfectly..

There is something strange here. I can not figure
out how correct code will fail with the sys.path.
It seems to me, the lib you are using is somehow not
able to recognize its own structure ("his own sys.path").

Idea. Are you sure you are modifying the sys.path at
the right place, understand at the right time
when Python processes?

I'm using this sys.path tweaking at run time very often;
eg to test or to run different versions of the same lib
residing in different dirs, and this, in *any* dir and
independently of *any* .pth file.


More information about the Python-list mailing list