Open Stack

On Tue, Jan 28, 2014 at 4:45 AM, John Garbutt <john at johngarbutt.com> wrote:
> On 27 January 2014 10:10, Daniel P. Berrange <berrange at redhat.com> wrote:
>> On Fri, Jan 24, 2014 at 11:42:54AM -0500, Joe Gordon wrote:
>>> On Fri, Jan 24, 2014 at 7:24 AM, Daniel P. Berrange <berrange at redhat.com>wrote:
>>>
>>> > Periodically I've seen people submit big coding style cleanups to Nova
>>> > code. These are typically all good ideas / beneficial, however, I have
>>> > rarely (perhaps even never?) seen the changes accompanied by new hacking
>>> > check rules.
>>> >
>>> > The problem with not having a hacking check added *in the same commit*
>>> > as the cleanup is two-fold
>>> >
>>> >  - No guarantee that the cleanup has actually fixed all violations
>>> >    in the codebase. Have to trust the thoroughness of the submitter
>>> >    or do a manual code analysis yourself as reviewer. Both suffer
>>> >    from human error.
>>> >
>>> >  - Future patches will almost certainly re-introduce the same style
>>> >    problems again and again and again and again and again and again
>>> >    and again and again and again.... I could go on :-)
>>> >
>>> > I don't mean to pick on one particular person, since it isn't their
>>> > fault that reviewers have rarely/never encouraged people to write
>>> > hacking rules, but to show one example.... The following recent change
>>> > updates all the nova config parameter declarations cfg.XXXOpt(...) to
>>> > ensure that the help text was consistently styled:
>>> >
>>> >   https://review.openstack.org/#/c/67647/
>>> >
>>> > One of the things it did was to ensure that the help text always started
>>> > with a capital letter. Some of the other things it did were more subtle
>>> > and hard to automate a check for, but an 'initial capital letter' rule
>>> > is really straightforward.
>>> >
>>> > By updating nova/hacking/checks.py to add a new rule for this, it was
>>> > found that there were another 9 files which had incorrect capitalization
>>> > of their config parameter help. So the hacking rule addition clearly
>>> > demonstrates its value here.
>>>
>>> This sounds like a rule that we should add to
>>> https://github.com/openstack-dev/hacking.git.
>>
>> Yep, it could well be added there. I figure rules added to Nova can
>> be "upstreamed" to the shared module periodically.
>
> +1
>
> I worry about diverging, but I guess thats always going to happen here.
>
>>> > I will concede that documentation about /how/ to write hacking checks
>>> > is not entirely awesome. My current best advice is to look at how some
>>> > of the existing hacking checks are done - find one that is checking
>>> > something that is similar to what you need and adapt it. There are a
>>> > handful of Nova specific rules in nova/hacking/checks.py, and quite a
>>> > few examples in the shared repo
>>> > https://github.com/openstack-dev/hacking.git
>>> > see the file hacking/core.py. There's some very minimal documentation
>>> > about variables your hacking check method can receive as input
>>> > parameters
>>> > https://github.com/jcrocholl/pep8/blob/master/docs/developer.rst
>>> >
>>> >
>>> > In summary, if you are doing a global coding style cleanup in Nova for
>>> > something which isn't already validated by pep8 checks, then I strongly
>>> > encourage additions to nova/hacking/checks.py to validate the cleanup
>>> > correctness. Obviously with some style cleanups, it will be too complex
>>> > to write logic rules to reliably validate code, so this isn't a code
>>> > review point that must be applied 100% of the time. Reasonable personal
>>> > judgement should apply. I will try comment on any style cleanups I see
>>> > where I think it is pratical to write a hacking check.
>>> >
>>>
>>> I would take this even further, I don't think we should accept any style
>>> cleanup patches that can be enforced with a hacking rule and aren't.
>>
>> IMHO that would mostly just serve to discourage people from submitting
>> style cleanup patches because it is too much stick, not enough carrot.
>> Realistically for some types of style cleanup, the effort involved in
>> writing a style checker that does not have unacceptable false positives
>> will be too high to justify. So I think a pragmatic approach to enforcement
>> is more suitable.
>
> +1
>
> I would love to enforce it 100% of the time, but sometimes its hard to
> write the rules, but still a useful cleanup. Lets see how it goes I
> guess.

I am weary of adding any new style rules that have to manually
enforced by human reviewers, we already have a lot of other items to
cover in a review.

>
> John