/usr/athena/bin/python breaks with /usr/athena symlink

[andersk]

debathena / python / jdreed  2011-09-21 16:42:37  (This zephyr does not necessarily reflect the views of IS&T, MIT, its)
    WTF?
    
    $ python -V
    Python 2.7.1+
    
    $ python -c 'import platform;print platform.python_version()'
    2.7.1+
    
    $ /usr/athena/bin/python -V
    Python 2.7.1+
    
    $ /usr/athena/bin/python -c 'import platform;print platform.python_version()'
    Traceback (most recent call last):
    File "/usr/athena/lib/python2.7/site.py", line 562, in <module>
      main()
    File "/usr/athena/lib/python2.7/site.py", line 544, in main
      known_paths = addusersitepackages(known_paths)
    File "/usr/athena/lib/python2.7/site.py", line 271, in addusersitepackages
      user_site = getusersitepackages()
    File "/usr/athena/lib/python2.7/site.py", line 246, in getusersitepackages
      user_base = getuserbase() # this will also set USER_BASE
    File "/usr/athena/lib/python2.7/site.py", line 236, in getuserbase
      USER_BASE = get_config_var('userbase')
    File "/usr/athena/lib/python2.7/sysconfig.py", line 558, in get_config_var
      return get_config_vars().get(name)
    File "/usr/athena/lib/python2.7/sysconfig.py", line 457, in get_config_vars
      _init_posix(_CONFIG_VARS)
    File "/usr/athena/lib/python2.7/sysconfig.py", line 310, in _init_posix
      raise IOError(msg)
    IOError: invalid Python installation: unable to open /usr/athena/local/lib/python2.7/config/Makefile (No such file or directory)
debathena / python / jdreed  2011-09-21 16:43:24  (This zephyr does not necessarily reflect the views of IS&T, MIT, its)
    (this is on, like, natty)
debathena / python / andersk  2011-09-21 16:45:04  (Anders Kaseorg)
    Haha, /usr/athena.
debathena / python / jdreed  2011-09-21 16:44:21  (This zephyr does not necessarily reflect the views of IS&T, MIT, its)
    so not the point
debathena / python / andersk  2011-09-21 16:49:02  (Anders Kaseorg)
    A workaround is python -S.
debathena / python / jdreed  2011-09-21 16:52:59  (This zephyr does not necessarily reflect the views of IS&T, MIT, its)
    Wait, what about /usr/athena is causing it to lose?  If I symlink
    python into /usr/local/bin or even /opt/bin it's fine
debathena / python / andersk  2011-09-21 16:56:18  (Anders Kaseorg)
    python being symlinked is different from its containing directory
    being symlinked.
debathena / python / andersk  2011-09-21 17:01:31  (Anders Kaseorg)
    From my sysconfig.py (on oneiric, which has the same problem):
    
    def _get_makefile_filename():
        if _PYTHON_BUILD:
            return os.path.join(_PROJECT_BASE, "Makefile")
        return os.path.join(get_path('platstdlib').replace("/usr/local","/usr",1), "config" + (sys.pydebug and "_d" or ""), "Makefile")
    
    and /usr/athena/local doesn’t match that.
debathena / python / jdreed  2011-09-21 17:02:17  (This zephyr does not necessarily reflect the views of IS&T, MIT, its)
    I don't, then, understand why this still works:
    jdreed@infinite-loop:~$ sudo ln -nsf /usr /opt/usr
    [sudo] password for jdreed: 
    jdreed@infinite-loop:~$ /opt/usr/bin/python -c 'import platform;print platform.python_version()'
    2.7.1+
debathena / python / andersk  2011-09-21 17:03:51  (Anders Kaseorg)
    Because /opt/usr/local does match that.
debathena / python / jdreed  2011-09-21 17:03:54  (This zephyr does not necessarily reflect the views of IS&T, MIT, its)
    Hrm.  We should probably still care somehow, since we just broke our
    compatibility layer without telling anyone
debathena / python / andersk  2011-09-21 17:05:11  (Anders Kaseorg)
    $ sudo ln -nsf /usr /opt/usr
    $ /opt/usr/bin/python -c ''
    $ sudo ln -nsf /usr /opt/rsu
    $ /opt/rsu/bin/python -c ''
    …
    IOError: invalid Python installation: unable to open /opt/rsu/local/lib/python2.7/config/Makefile (No such file or directory)
debathena / python / andersk  2011-09-21 17:08:55  (Anders Kaseorg)
    Well, uh, you can read debian/patches/distutils-install-layout.diff,
    cry, and file a hilarious Debian bug?
debathena / python / andersk  2011-09-21 17:13:06  (Anders Kaseorg)
    It looks like it would work if /usr/athena/bin was the symlink
    instead of /usr/athena?

[jdreed]

It looks like it would work if /usr/athena/bin was the symlink instead of /usr/athena?

Is there a good reason not to do this? What would break? This is certainly easier than debathena-python-config, and if you have /usr mounted somewhere other than /usr _AND_ still rely on /usr/athena/bin working, then we don't have to support you.

mkdir -p /usr/athena
for d in bin include lib sbin share; do
   sudo ln -nsf /usr/$d /usr/athena/$d
done

[kaduk]

Replying to jdreed:

It looks like it would work if /usr/athena/bin was the symlink instead of /usr/athena?

Is there a good reason not to do this? What would break? This is certainly easier than debathena-python-config, and if you have /usr mounted somewhere other than /usr _AND_ still rely on /usr/athena/bin working, then we don't have to support you.

mkdir -p /usr/athena
for d in bin include lib sbin share; do
   sudo ln -nsf /usr/$d /usr/athena/$d
done

Karl notes on zephyr that this produces surprising behavior if the filesystem normally mounted at /usr is mounted someplace else. Though perhaps a relative symlink would work? I haven't followed closely enough to tell.

[jdreed]

Greg notes that changing the symlink target to, e.g. ../bin will work if the filesystem is mounted somewhere else, and also fixes the python issue. Should we just do that?

jdreed@infinite-loop:~$ sudo ln -nsf ../bin /usr/athena/bin
jdreed@infinite-loop:~$ ls -l /usr/athena/bin
lrwxrwxrwx 1 root root 6 Aug  1 12:30 /usr/athena/bin -> ../bin
jdreed@infinite-loop:~$ /usr/athena/bin/python -c ''

[jdreed]

This doesn't get us closer to deprecation, of course. The right way to do that is probably to provide symlinks for everything that was there under athena 9 (795 in /usr/athena/bin, yay), which whine loudly and then exec their thing anyway, and then eventually pull /usr/athena. But that is not this ticket.

[kaduk]

Replying to jdreed:

Greg notes that changing the symlink target to, e.g. ../bin will work if the filesystem is mounted somewhere else, and also fixes the python issue. Should we just do that?

jdreed@infinite-loop:~$ sudo ln -nsf ../bin /usr/athena/bin
jdreed@infinite-loop:~$ ls -l /usr/athena/bin
lrwxrwxrwx 1 root root 6 Aug  1 12:30 /usr/athena/bin -> ../bin
jdreed@infinite-loop:~$ /usr/athena/bin/python -c ''

I am happy with doing that and seeing how things work/break.

[jdreed]

I committed this change in r25755. However, as noted on zephyr (but not e-mail -- *glare*), this likely won't work. And in fact, it doesn't, because of how dpkg unpacks packages.

For maximum lulz, the fix which worked on in comment 5 does not in fact work on Precise. "Yay."

That having been said, removing /usr/athena and replacing it with a directory seems to have no effect on a running perl script launched from /usr/athena/bin/perl. So we can either do something stupid with debathena-python-config, or we can finally deprecate /usr/athena/bin, which I think I'm also fine with doing.

[jdreed]

I committed r25777, though Anders notes this might be "fixed" in Quantal.

[jdreed]

 http://kb.mit.edu/confluence/x/0IQBCQ