http://www.python.org/dev/buildbot/3.0/ New sprint idea: getting all (inc. trunk) the buildbots green by Thursday. Anyone interested? Trent.
New sprint idea: getting all (inc. trunk) the buildbots green by Thursday. Anyone interested?
I think the chance to achieve that is close to zero.
Sounds like a challenge if ever I've heard one -- care to wager a beer on it? (Only applies to buildbots that are connected/online.) (FWIW, I've got the x64 Windows build green on my dev server, tcl/tk and bsddb required patching, so did some tests, and so did some C code -- I'm in the process of filtering the efforts back into the tracker.) Trent.
New sprint idea: getting all (inc. trunk) the buildbots green by Thursday. Anyone interested?
I think the chance to achieve that is close to zero.
Sounds like a challenge if ever I've heard one -- care to wager a beer on it? (Only applies to buildbots that are connected/online.)
Even with that restriction: I'll happily buy you a beer if you manage to get all the online ones pass all tests consistently. Regards, Martin
On Sun, Mar 16, 2008 at 12:07 PM, "Martin v. Löwis"
New sprint idea: getting all (inc. trunk) the buildbots green by Thursday. Anyone interested?
I think the chance to achieve that is close to zero.
Sounds like a challenge if ever I've heard one -- care to wager a beer on it? (Only applies to buildbots that are connected/online.)
Even with that restriction: I'll happily buy you a beer if you manage to get all the online ones pass all tests consistently.
I think this is possible, though considerable work. Probably the biggest win will be creating a mock for socket and using mock sockets in the tests for asyn{core,chat}, smtplib, xmlrpc, etc. That will fix about 75% of the problems on 2.6. The remaining problems are: * test_asyn{chat,core} might not be meaningful with mock sockets and are flaky * the alpha fails test_signal/socket for weird alarm conditions. this might be hard to debug/fix (I have access to this box though) * test_sqlite is broken on x86 with an old sqlite (I have access to this box) * test_bsddb may be flaky, I'm not sure * probably a few platform specific problems 3.0 will get most of the improvements as the fixes are ported. 3.0 has a different problem on the x86 box dealing with signals. There are probably some other 3.0 specific problems. Although, using a mock socket could address this too. I can help on fixing these issues during the sprints. It will be happy to work with Trent to try to fix the problems. I'm sure we can greatly improve the situation. n
On Sun, Mar 16, 2008 at 1:32 PM, Neal Norwitz
I think this is possible, though considerable work. Probably the biggest win will be creating a mock for socket and using mock sockets in the tests for asyn{core,chat}, smtplib, xmlrpc, etc. That will fix about 75% of the problems on 2.6. The remaining problems are:
* test_asyn{chat,core} might not be meaningful with mock sockets and are flaky * the alpha fails test_signal/socket for weird alarm conditions. this might be hard to debug/fix (I have access to this box though) * test_sqlite is broken on x86 with an old sqlite (I have access to this box) * test_bsddb may be flaky, I'm not sure * probably a few platform specific problems
test_tokenize is also currently (sometimes) failing on many of the bots. I've been looking into it, but I'm struggling to find the problem. The traceback e.g. for the amd64 gentoo buildbot ends with File "/home/buildbot/slave/py-build/3.0.norwitz-amd64/build/Lib/io.py", line 1081, in decode output = self.decoder.decode(input, final=final) File "/home/buildbot/slave/py-build/3.0.norwitz-amd64/build/Lib/codecs.py", line 291, in decode (result, consumed) = self._buffer_decode(data, self.errors, final) UnicodeDecodeError: 'utf8' codec can't decode bytes in position 12-15: invalid data On my own machine (SuSE 9.3/i686) I'm seeing this test pass about 80% of the time and fail the other 20% with something like the above, the position of the reported invalid data changing from run to run. It looks like data are getting corrupted somewhere along the line. Anyone have any ideas? Mark
On Sun, Mar 16, 2008 at 5:13 PM, Mark Dickinson
On Sun, Mar 16, 2008 at 1:32 PM, Neal Norwitz
wrote: I think this is possible, though considerable work. Probably the biggest win will be creating a mock for socket and using mock sockets in the tests for asyn{core,chat}, smtplib, xmlrpc, etc. That will fix about 75% of the problems on 2.6. The remaining problems are:
* test_asyn{chat,core} might not be meaningful with mock sockets and are flaky * the alpha fails test_signal/socket for weird alarm conditions. this might be hard to debug/fix (I have access to this box though) * test_sqlite is broken on x86 with an old sqlite (I have access to this box) * test_bsddb may be flaky, I'm not sure * probably a few platform specific problems
test_tokenize is also currently (sometimes) failing on many of the bots. I've been looking into it, but I'm struggling to find the problem. The traceback e.g. for the amd64 gentoo buildbot ends with
File "/home/buildbot/slave/py-build/3.0.norwitz-amd64/build/Lib/io.py", line 1081, in decode output = self.decoder.decode(input, final=final) File "/home/buildbot/slave/py-build/3.0.norwitz-amd64/build/Lib/codecs.py", line 291, in decode (result, consumed) = self._buffer_decode(data, self.errors, final) UnicodeDecodeError: 'utf8' codec can't decode bytes in position 12-15: invalid data
On my own machine (SuSE 9.3/i686) I'm seeing this test pass about 80% of the time and fail the other 20% with something like the above, the position of the reported invalid data changing from run to run. It looks like data are getting corrupted somewhere along the line.
Anyone have any ideas?
Yeah, sounds like a memory issue. Did you try running with valgrind or purify? I haven't done so for a long time, perhaps never on 3k branch. It would be a good thing to run a tool soon. n
Yeah, sounds like a memory issue. Did you try running with valgrind or purify? I haven't done so for a long time, perhaps never on 3k branch. It would be a good thing to run a tool soon.
Maybe is this related? [Potential overflows due to incorrect usage of PyUnicode_AsString] http://bugs.python.org/issue1950 Thank you.
As it turns out, it's not memory related, but has to do with tokenize not supporting coding cookies in files. Mark picked up on this and linked it to an issue already in roundup that was raised way back in 2003: http://bugs.python.org/issue71988. I've just finished patching test_tokenizer.py to better represent this test case -- the current implementation doesn't lend itself very well to being debugged when things go wrong (I think Mark and I both felt like we were on a bit of a wild goose chase). I've fixed that and have a bunch of text files with various utf-8/bom sig permutations that are now used to test tokenizer's compliance with PEP 0263. I'll upload that now then move on to actually patching tokenizer.py. Trent "wishes-there-was-somewhere-to-get-some-food-after-11pm-at-pycon" Nelson. ________________________________________ From: python-dev-bounces+tnelson=onresolve.com@python.org [python-dev-bounces+tnelson=onresolve.com@python.org] On Behalf Of ocean [ocean@m2.ccsnet.ne.jp] Sent: 17 March 2008 01:34 To: Neal Norwitz; Mark Dickinson Cc: Python Dev Subject: Re: [Python-Dev] 3.0 buildbots all red
Yeah, sounds like a memory issue. Did you try running with valgrind or purify? I haven't done so for a long time, perhaps never on 3k branch. It would be a good thing to run a tool soon.
Maybe is this related? [Potential overflows due to incorrect usage of PyUnicode_AsString] http://bugs.python.org/issue1950 Thank you. _______________________________________________ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/tnelson%40onresolve.com
As it turns out, it's not memory related, but has to do with tokenize not supporting coding cookies in files. Mark picked up on this and linked it to an issue already in roundup that was raised way back in 2003: http://bugs.python.org/issue71988.
Oops, left off an 8. That's meant to read http://bugs.python.org/issue719888.
from test import test_tokenize [50808 refs] test_tokenize.test_tokenize_all()
Yeah test_tokenize is weird, I've been looking into it as well. Here's a sample failure from a Windows buildbot:
File "S:\buildbots\python\3.0.nelson-windows\build\lib\test\test_tokenize.py", line ?, in test.test_tokenize.__test__.doctests
Failed example:
for testfile in testfiles:
if not roundtrip(open(testfile)): break
else: True
Exception raised:
Traceback (most recent call last):
File "S:\buildbots\python\3.0.nelson-windows\build\lib\doctest.py", line 1227, in __run
compileflags, 1), test.globs)
File "
I think this is possible, though considerable work. Probably the biggest win will be creating a mock for socket and using mock sockets in the tests for asyn{core,chat}, smtplib, xmlrpc, etc. That will fix about 75% of the problems on 2.6. The remaining problems are:
* test_asyn{chat,core} might not be meaningful with mock sockets and are flaky * the alpha fails test_signal/socket for weird alarm conditions. this might be hard to debug/fix (I have access to this box though) * test_sqlite is broken on x86 with an old sqlite (I have access to this box) * test_bsddb may be flaky, I'm not sure * probably a few platform specific problems
test_tokenize is also currently (sometimes) failing on many of the bots. I've been looking into it, but I'm struggling to find the problem. The traceback e.g. for the amd64 gentoo buildbot ends with File "/home/buildbot/slave/py-build/3.0.norwitz-amd64/build/Lib/io.py", line 1081, in decode output = self.decoder.decode(input, final=final) File "/home/buildbot/slave/py-build/3.0.norwitz-amd64/build/Lib/codecs.py", line 291, in decode (result, consumed) = self._buffer_decode(data, self.errors, final) UnicodeDecodeError: 'utf8' codec can't decode bytes in position 12-15: invalid data On my own machine (SuSE 9.3/i686) I'm seeing this test pass about 80% of the time and fail the other 20% with something like the above, the position of the reported invalid data changing from run to run. It looks like data are getting corrupted somewhere along the line. Anyone have any ideas? Mark _______________________________________________ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/tnelson%40onresolve.com
On Sun, Mar 16, 2008 at 7:19 PM, Trent Nelson
Yeah test_tokenize is weird, I've been looking into it as well. Here's a sample failure from a Windows buildbot: [failure log snipped...] On that first line, 'f' is lib/test/tokenize_tests.txt, so basically, it's grabbing ten random test*.py files in lib/test and running untokenize(generate_tokens(f.readline)) on each one. In order to figure out which file it's dying on, I added the following to test_tokenize.py:
Aha. This is very helpful! It explains the randomness of the failures, at least. So it's probably not a C-level data corruption bug after all. test_shlex seems to be one of the problem files. Investigations are continuing. Trent, thanks for your help! I'll take further discussions off line and spare python-dev the gory details... Mark
Trent Nelson wrote:
New sprint idea: getting all (inc. trunk) the buildbots green by Thursday. Anyone interested?
I think the chance to achieve that is close to zero.
Sounds like a challenge if ever I've heard one -- care to wager a beer on it? (Only applies to buildbots that are connected/online.)
(FWIW, I've got the x64 Windows build green on my dev server, tcl/tk and bsddb required patching, so did some tests, and so did some C code -- I'm in the process of filtering the efforts back into the tracker.)
Make sure you get a screen shot for OnYourDesktop if/when they *do* go green! regards Steve
Sounds like a challenge if ever I've heard one -- care to wager a beer on it? (Only applies to buildbots that are connected/online.)
Make sure you get a screen shot for OnYourDesktop if/when they *do* go green!
Screenshot? I'm going to buy a pack of iron-on transfers and sell t-shirts of it online. "All the buildbots were green momentarily after PyCon 2008... ....and all I got was this lousy t-shirt." Trent.
Trent Nelson wrote:
Sounds like a challenge if ever I've heard one -- care to wager a beer on it? (Only applies to buildbots that are connected/online.)
Make sure you get a screen shot for OnYourDesktop if/when they *do* go green!
Screenshot? I'm going to buy a pack of iron-on transfers and sell t-shirts of it online.
"All the buildbots were green momentarily after PyCon 2008... ....and all I got was this lousy t-shirt."
<British humour> Momentarily? You mean they were only up for a few seconds?
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Mar 16, 2008, at 10:48 AM, Martin v. Lwis wrote:
New sprint idea: getting all (inc. trunk) the buildbots green by Thursday. Anyone interested?
I think the chance to achieve that is close to zero.
How broken do you want the next monthly alphas to be? :) - -Barry -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.8 (Darwin) iQCVAwUBR91OWHEjvBPtnXfVAQIL/gP/T5TqGLt6Z7LD3/vEKt/0qjZErkOQmmFH BciuGOs6CCud//N4fsSev2CBmeAQ/RspHhjLSSKBd6H4rimZWv5ePo0gMC2N7nGY 1wpUvixaZpYol4zywjBQRO+bjlnZtdt4WG09DdMrn0MsYHNpVFaAPTWx4X3BFHm+ k0QLzsJ7T2o= =7czp -----END PGP SIGNATURE-----
participants (9)
-
"Martin v. Lwis"
-
"Martin v. Löwis"
-
Barry Warsaw
-
Mark Dickinson
-
Neal Norwitz
-
ocean
-
Steve Holden
-
Tim Golden
-
Trent Nelson