On 09/21/2013 07:31 AM, Amos Jeffries wrote:
> On 10/09/2013 7:18 a.m., Kinkie wrote:
>>> Another [debatable] flaw is that Squid build fails on harmless warnings.
>> This is an interesting point; let's debate :)
>>
>> Now that our CI infrastructure has grown quite a bit, we may have to
>> rely less on outside reporting of trivial issues. If that is the case,
>> what do you think about dropping the default level of paranoia in the
>> build flags, leaving that as optional for the build farm to use?
> I consider that improved testing setup a good reason to *keep* the
> strict error checking enabled by default. Our setup still cannot test
> everything conclusively and while the need for user feedback is lessened
> it has not gone away. Certainly it is still very present in the more
> closed-source of the downstream build systems which we do not or cannot
> test for. The later user experience building Squid is better for the
> small amount of pain early-adopters are (knowingly) faced with.
Our choice is between build failures on harmless warnings and ignored
warnings that expose a serious bug. Given that the former problems are
far more frequent and often very annoying to our users, I think we
should prefer the latter as the default. It is also conceptually wrong
to violate GCC classification by default.
Developers can use a different build setting, and perhaps we can even
tell trunk to fail on warnings while all numbered branches leave it up
to the compiler to determine the severity of the build problem.
While I consider the current behavior flawed, I personally do not insist
on changing it. Whether our users should be treated as guinea pigs in
this context is a matter of taste, Neither side can prove the other one
wrong IMO.
Thank you,
Alex.
Received on Sun Sep 22 2013 - 23:18:59 MDT
This archive was generated by hypermail 2.2.0 : Mon Sep 23 2013 - 12:00:12 MDT