The 'test_posix' of Python's 2.5.4.3 testsuite fails on HPUX. This has
happened on parisc and ia64 systems. On the parisc system failure occurs
for both parisc1.1 and parisc2.0w builds. The HPUX versions are 11.00
(parisc) and 11.22 (ia64).

The attached file contains the output of executing

./apy/bin/python2.5 ./apy/lib/python2.5/test/regrtest.py -v test_posix

If you try "chown root foobar", with <foobar> being a valid file you
own, does it succeed?

Yes.

Session log:

andreask@bertha:~/tmp/x> ll
total 0

andreask@bertha:/tmp/x> touch foobar
andreask@bertha:/tmp/x> ll
total 0
-rw-rw-r-- 1 andreask DomainUsers 0 Feb 2 10:17 foobar

andreask@bertha:/tmp/x> chown root foobar
andreask@bertha:/tmp/x> ll
total 0
-rw-rw-r-- 1 root DomainUsers 0 Feb 2 10:17 foobar

That explains what the test fails then. The test expects that you can't
chown a file to root as a normal user, and it verifies that an exception
is raised in that case.
Do you know if this behaviour is standard under HP-UX? If so, what do
sys.platform and platform.system() return?

(this won't be fixed in 2.5 by the way: 2.5 only receives security
fixes. 2.6 is the current stable branch and 2.7 is the in-development
version)

I do not know if this is standard for hpux or not.

I can say that right now we have six different HPUX machines which
behave in this manner, i.e. do not throw an error for the 'chown root
foobar' command for an ordinary (non-root) user.

Regarding the platform codes, here are the results.

andreask@bertha:~/tmp/autotest> ./apy/bin/python2.5
ActivePython 2.5.4.3 (ActiveState Software Inc.) based on
Python 2.5.4 (r254:67916, Jan 20 2009, 14:48:58) [C] on hp-ux11
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys, platform
>>> sys.platform
'hp-ux11'
>>> platform.system()
'HP-UX'
>>> quit()

The six machines I mentioned are

bertha - HP-UX 11.00 (PA-RISC)
bigsur03 - HP-UX 11.22 (Itanium)
hpa180 - HP-UX 11.00 (PA-RISC)
bigsur03 - HP-UX 11.22 (Itanium)
hpl3000 - HP-UX 11.23 (PA-RISC)
bigsur05 - HP-UX 11.32 (Itanium)

Ok, and when you try to chown back to yourself the root-owned foobar,
does it fail?

Yes, that does fail with error 'not owner'.

andreask@bertha:/tmp> mkdir x
andreask@bertha:/tmp> cd x
andreask@bertha:/tmp/x> touch foobar
andreask@bertha:/tmp/x> ll
total 0
-rw-rw-r-- 1 andreask DomainUsers 0 Feb 2 15:06 foobar
andreask@bertha:/tmp/x> chown root foobar
andreask@bertha:/tmp/x> ll
total 0
-rw-rw-r-- 1 root DomainUsers 0 Feb 2 15:06 foobar
andreask@bertha:/tmp/x> chown andreask foobar
foobar: Not owner
andreask@bertha:/tmp/x>

Actually, looking at the man page, this seems to be normal:

---
In order to change the owner or group, you must own the file and have the CHOWN privilege (see setprivgrp(1M)). [...] Note that a given user's or group's ability to use this command can be restricted by setprivgrp (see setprivgrp(1M)).
---

So depending on the setting of setprivgrp, chown'ing a file to root may be allowed or not.
It might be a good idea to skip this test on HP-UX. Solaris apparently behaves in the same way.

Patch attached.

It's surprising to raise a SkipTest *at the end* of the test. Here is another patch.

Using chown_hpux-4.diff, we don't skip anything on HP-UX.

Using chown_hpux-4.diff, we don't skip anything on HP-UX.

Victor, I don't understand your patch: if we try chown()ing a file to root, then this will fail on HP-UX (because chown won't fail with EPERM). That's the whole point of the patch.

if we try chown()ing a file to root, then this will fail on HP-UX

Oooh, I'm confused. I removed this stupid patch.

# HP-UX and Solaris allow a non-root user to chown to root

I didn't see failure on our Solaris / OpenIndiana buildbot. Should we really skip the test on Solaris?

Well, IIRC, I read in the Solaris man pages that it behaves like HP-UX
in this regard. But this it can be changed with setprivgrp, so it
might work on some machines, and fail on others.
According to http://unix.derkeiler.com/Newsgroups/comp.unix.solaris/2004-03/2579.html,
the test should work with the default setting (maybe the OP was
testing on a HP-UX machine with the non-default setting).
Anyway, since it depends on the system configuration, I wanted to stay
on the safe side, and just skip on both OS.
But I honestly don't have a strong opinion about this.

Oooh, I'm confused. I removed this stupid patch.

chown_hpux-3.diff has a problem too: it skips the successful chown call (that's why I put the SkipTest at the end).

Adi, since you have access to an HP-UX box, could you test the attached patch (chown_hpux.diff)?
Also, if you're interested, you could search for other isues HP-UX-specific to see if you can help.

Hi,

Not sure what codebase was used for the patch.

I have manually patched the test on 2.5.6 and the test_posix tests passed.

Thanks!
Adi

I am not sure how the search works, but for example I was not able to reach http://bugs.python.org/issue5895 while searching for hpux or hp-ux using the internal search.

Also I was not able to find any aix or solaris bug.

In case you bump over an hpux, aix, solaris bug, please add me to the noisy list and I will take a look.

Thanks!
Adi

Search is currently not returning all matching issues, unfortunately. You might get a few more hits by searching for hpux in the title field via advanced search.

New changeset 8a4c9a168d09 by Charles-François Natali in branch '2.7':
Issue bpo-5113: Fix a test_posix failure on HP-UX, where non-root users can
http://hg.python.org/cpython/rev/8a4c9a168d09

New changeset 428bece48029 by Charles-François Natali in branch '3.2':
Issue bpo-5113: Fix a test_posix failure on HP-UX, where non-root users can
http://hg.python.org/cpython/rev/428bece48029

New changeset 3d4922cf0d65 by Charles-François Natali in branch 'default':
Issue bpo-5113: Fix a test_posix failure on HP-UX, where non-root users can
http://hg.python.org/cpython/rev/3d4922cf0d65

Committed, thanks.