Granularity

[Myk Melez]

MattyT wrote:

>What really concerns me here is not the per-component issue, but the
>fact that we're discussing it *after* it's been checked in.  And this is
>hardly the first time it's happened for me.
>
It's common for me, too, but that's the price of working for a large, 
active project and not spending my life reading bug mail.  I regularly 
notice check-ins with flaws I would have corrected, but the patches are 
almost always good enough (TM), and I'd rather they go in than be 
shunted through another bottleneck that I still won't have enough time 
to keep up with.

>Currently everyone is working on their own enhancements, and they only
>get one review.  This means a large number of people are potentially in
>the dark about the implementation of features, and complaints arise when
>it's essentially too late.
>
Even patches that only need one review frequently get several sets of 
experienced eyes on them (and lack of a comment doesn't mean lack of 
attention).  It doesn't matter how many people are in the dark as long 
as a sufficient number aren't, which could be as few as one if that 
person is an accomplished Bugzilla hacker.

>Many freed software projects have a single mind who ultimately controls
>what goes in and what doesn't - Linux for example.  We don't have this,
>Dave probably does not have the time to design review all patches.  But
>it's still important that we have vision and don't let Bugzilla become
>any more crazy and unorthogonal than it already is in some places.
>
I agree, but our code quality has been consistently increasing, so I 
don't think we have a "crazy and unorthogonal" problem, and making flags 
component-specific certainly isn't an example of it, since intentional 
deviation from a norm for a good reason is a good thing.

>What I think needs to be done, is to have ALL enhancements thoroughly
>summarised and the summary posted to this list.
>
In other words, design by committee.  No thanks.

>We have certainly done this sort of thing before, but not for everything
>(or even most things), and I think we should expect it to be.
>  
>
I disagree.  Design discussions on this list have had some very bad 
results, f.e. the "sending mail" patch which got bogged down on design 
so much that nothing has been done, even though all the proposed 
solutions were good enough and any of them would have been better than 
no change at all, which is what we have now.

I'd much rather have Dave approve patches or delegate such approval to 
the component owners, with discussions on the list reserved for cases 
where developers need help or can't agree.

-myk