Tagging 3.4a1 with an unhappy AMD64 Win7 SP1 buildbot?
I'm considering for Python 3.4.0a1. The buildbots are all happy, except for AMD64 Win7 SP1:
http://buildbot.python.org/all/waterfall?category=3.x.stable
The two failures are zipimport and signal. zipimport is sporadic, and looks like some sort of heisenissue. signal is much more consistently broken, and has been broken for a while. (Not that Windows support for signal has ever been that great. But presumably the regression test is supposed to light up green, even on Windows.)
Can we, uh, live with that, for alpha 1?
//arry/
On 08/03/2013 02:08 AM, Larry Hastings wrote:
Can we, uh, live with that, for alpha 1?
Ned pointed out, signal and zipimport worked on the retry. So it's sporadic, whatever it is. I think we can live with that.
I've got some documentation warnings I can't quash, so I'm not tagging yet.
//arry/
On Sat, Aug 3, 2013 at 2:45 AM, Larry Hastings larry@hastings.org wrote:
On 08/03/2013 02:08 AM, Larry Hastings wrote:
Can we, uh, live with that, for alpha 1?
Ned pointed out, signal and zipimport worked on the retry. So it's sporadic, whatever it is. I think we can live with that.
This is confusing. Why do we need an alpha release in the first place? Is it more important to make it on the exact day than to have a more functional release? Assuming that 64-bit Windows 7 is the most popular (or close to it) Windows out there today, that sounds like an annoying flaw.
The above is genuine will to understand the release process rather than questioning your reasoning :-)
Eli
On 3 August 2013 19:08, Larry Hastings larry@hastings.org wrote:
I'm considering for Python 3.4.0a1. The buildbots are all happy, except for AMD64 Win7 SP1:
http://buildbot.python.org/all/waterfall?category=3.x.stable
The two failures are zipimport and signal. zipimport is sporadic, and looks like some sort of heisenissue. signal is much more consistently broken, and has been broken for a while. (Not that Windows support for signal has ever been that great. But presumably the regression test is supposed to light up green, even on Windows.)
The test_signal problem should be fixed now given my last push. On a second review, I actually understood the rationale behind Jeremy's proposed test patch in http://bugs.python.org/issue18396 and accepted it.
Cheers, Nick.
-- Nick Coghlan | ncoghlan@gmail.com | Brisbane, Australia
On 3 August 2013 22:43, Eli Bendersky eliben@gmail.com wrote:
On Sat, Aug 3, 2013 at 2:45 AM, Larry Hastings larry@hastings.org wrote:
On 08/03/2013 02:08 AM, Larry Hastings wrote:
Can we, uh, live with that, for alpha 1?
Ned pointed out, signal and zipimport worked on the retry. So it's sporadic, whatever it is. I think we can live with that.
This is confusing. Why do we need an alpha release in the first place? Is it more important to make it on the exact day than to have a more functional release? Assuming that 64-bit Windows 7 is the most popular (or close to it) Windows out there today, that sounds like an annoying flaw.
The test_signal problem was just a conflict between the details of the test and having faulthandler enabled. Nothing wrong with the signal handling itself.
The above is genuine will to understand the release process rather than questioning your reasoning :-)
Alpha deadlines are the ones with the *least* excuses for delays. It's better to expose the 99.9% that is working for broader feedback rather than holding that side of the process up for a couple of minor flaws.
The trade-offs change as you move closer to a final release, but this early in the cycle erring on the side of "just ship it" is a good way to go.
Cheers, Nick.
-- Nick Coghlan | ncoghlan@gmail.com | Brisbane, Australia
Quoting Eli Bendersky eliben@gmail.com:
On Sat, Aug 3, 2013 at 2:45 AM, Larry Hastings larry@hastings.org wrote:
On 08/03/2013 02:08 AM, Larry Hastings wrote:
Can we, uh, live with that, for alpha 1?
Ned pointed out, signal and zipimport worked on the retry. So it's sporadic, whatever it is. I think we can live with that.
This is confusing. Why do we need an alpha release in the first place? Is it more important to make it on the exact day than to have a more functional release? Assuming that 64-bit Windows 7 is the most popular (or close to it) Windows out there today, that sounds like an annoying flaw.
The above is genuine will to understand the release process rather than questioning your reasoning :-)
In case it hasn't been answered: the point of alpha1 really is to make a release, and run through the release process. It will likely turn out that the release process fails in some way, and it may take all the alphas to get the issues resolved - so it is *expected* that the release is done despite known issues.
FWIW, I don't consider the failure of the zipimport and signal tests that relevant, even on Windows. It doesn't mean that the release is completely useless for users (it could even be that the modules actually work, just the test runner is flawed).
We hope that end users (and in particular library developers) test the release, to find out what got broken since 3.3. Again, for that, it isn't critical that all tests pass, since those users are expected to have different test cases, anyway.
Regards, Martin
participants (4)
-
Eli Bendersky
-
Larry Hastings
-
martin@v.loewis.de
-
Nick Coghlan