DeprecationWarning in gzip.py causes make check-dist-tarball to fail on Python 2.6

[Jonathan Lange]

make check-dist-tarball fails with this error when run on a system with python2.6 as the default:

======================================================================
ERROR: test_filenames (bzrlib.tests.test_multiparent.TestMultiVersionedFile)

vvvv[log from bzrlib.tests.test_multiparent.TestMultiVersionedFile.test_filenames]
1068.606 opening working tree '/tmp/testbzr-9V8eOp.tmp'

^^^^[log from bzrlib.tests.test_multiparent.TestMultiVersionedFile.test_filenames]
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/tmp/tmp_bzr_check_dist.MRBDNZBclG/bzr-1.16rc1/bzrlib/tests/test_multiparent.py", line 268, in test_filenames
    vf.add_version('a\nb\nc\nd'.splitlines(True), 'a', [])
  File "/tmp/tmp_bzr_check_dist.MRBDNZBclG/bzr-1.16rc1/bzrlib/multiparent.py", line 365, in add_version
    self.add_diff(diff, version_id, parent_ids)
  File "/tmp/tmp_bzr_check_dist.MRBDNZBclG/bzr-1.16rc1/bzrlib/multiparent.py", line 573, in add_diff
    zipfile.close()
  File "/usr/lib/python2.6/gzip.py", line 313, in close
    write32u(self.fileobj, self.crc)
  File "/usr/lib/python2.6/gzip.py", line 21, in write32u
    output.write(struct.pack("<L", value))
DeprecationWarning: struct integer overflow masking is deprecated

----------------------------------------------------------------------
Ran 14646 tests in 1063.068s

[Martin Pool (mbp)]

It seems like we should (probably in the bzr script) be masking out warnings from libraries known to not be up to date with the Python interpreter.

[Robert Collins (lifeless)]

Ugh, thats just nasty (python 2.6's own code spitting out deprecations like that). Agreed on us masking this out.

[Martin Pool (mbp)]

Biting me too trying to test one of my branches

[John A Meinel (jameinel)]

I'm pretty sure this is python2.6 + 64-bit system.

I'm not positive, but I would guess that the "<L" is defined as a 32-bit op, but "value" is a PyInt which is a 64-bit object.

That, and if it is a crc32, then it is an 'unsigned' 32-bit value, stored in a 64-bit int. Which on 32-bit is exposed as a 32-bit *signed* integer (and thus wouldn't fail the check), but on 64-bit it shows up as a signed *64-bit* integer which can hold a 32-bit unsigned integer just fine.

I'm concerned that gzip.py might be broken on 64-bit. My first guess is that it isn't or it would have shown up for people a while ago. (It is warning because functionality may change in the future, but for now it is still encoding all 32-bits into the output, *without* truncating oddly because of signed/unsigned issues.)

Anyway, as long as it is correct, suppressing the error seems reasonable.

[Martin Pool (mbp)]

I think the warning means "this works, but it won't work in 2.7", so we just need to hope they fix it before then.

[Martin Pool (mbp)]

With that particular warning masked off, we also get this, which seems almost equivalent. I don't know if struct.py is raising both one after the other...

^^^^[log from bzrlib.tests.test_multiparent.TestMultiVersionedFile.test_save_load]
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/mbp/bzr/387139-warnings/bzrlib/tests/test_multiparent.py", line 258, in test_save_load
    vf.add_version('a\nb\nc\nd'.splitlines(True), 'a', [])
  File "/home/mbp/bzr/387139-warnings/bzrlib/multiparent.py", line 365, in add_version
    self.add_diff(diff, version_id, parent_ids)
  File "/home/mbp/bzr/387139-warnings/bzrlib/multiparent.py", line 573, in add_diff
    zipfile.close()
  File "/usr/lib/python2.6/gzip.py", line 313, in close
    write32u(self.fileobj, self.crc)
  File "/usr/lib/python2.6/gzip.py", line 21, in write32u
    output.write(struct.pack("<L", value))
DeprecationWarning: 'L' format requires 0 <= number <= 4294967295

[Martin Pool (mbp)]

Fix for at least this one warning in the attached branch

[Martin Pool (mbp)]

Sent to pqm