monarch dodra granted write access to phobos, druntime, and tools

[Don]

On Tuesday, 23 July 2013 at 19:24:10 UTC, Andrei Alexandrescu 
wrote:
> On 7/23/13 8:27 AM, Leandro Lucarella wrote:
>> You don't need push access to the blessed repository to 
>> contribute,
>> THAT's why git exists! Only people merging stuff needs push 
>> access and
>> is good to keep that team as small as possible (and if there 
>> is a review
>> bottleneck then it is too small and needs to be expanded). 
>> Right now,
>> fortunately, the lack of review doesn't seem to be a huge 
>> bottleneck,
>> and while having committed, smart people helping could be 
>> beneficial,
>> I think is wise not to give every contributor push access to 
>> the repo
>> right now.
>
> I'm very surprised by your outlook. My perception is that the 
> long queue of pending pull requests not being reviewed is the 
> single most important bottleneck at this point in history in 
> the path of D.

Yes, but OTOH, I'd like to see more people who ONLY review Phobos 
requests, with the distraction of also handling DMD requests.

My impression is that Phobos requests currently take even longer 
to be merged, than DMD requests. (Far more DMD pull requests are 
generated than for Phobos, yet the queues are a similar length).


> By my estimates I think we'd improve the speed of D's 
> development by at least one third if we solve this one issue. 
> There's no other issue offering so much impact.

I agree.

> I also think it may transform into a major crisis (an 
> inflection point in pull requests rate followed by a decline) 
> if we leave this unresolved.

Yes. I suspect that may already apply to Phobos to some extent. 
Potentially there there should be many more contributors to 
Phobos than to DMD. But it has not happened so far.

> We must find a solution to reviewing pull requests, and fast.

There are still a couple of easy practical issues. The build time 
is still not quick enough, which means we have a 
last-in-first-out structure. When the top few pull requests are 
controversial or harder to review, they tend to clog the request 
queue.

The simplest immediate improvement we could do is add another 
FreeBSD box to the build farm. FreeBSD takes so long, and gives 
so many spurious failures, that we'd be currently be better off 
without it.