On Tuesday, 16 January 2018 at 22:23:41 UTC, Walter Bright wrote:
> On 1/16/2018 10:32 AM, qznc wrote:
>> On Monday, 15 January 2018 at 01:10:58 UTC, Walter Bright wrote:
>>> PRs banished to the Phantom Zone would have to have an explanatory comment saying why.
>> 
>> I would suggest an alternative approach: Automate it.
>
> Banishment to the PZ requires judgement.

Oh my. Banishment, phantoms, judgement day... It's not even Halloween!

On Monday, 15 January 2018 at 01:10:58 UTC, Walter Bright wrote:
[snip]
> Andrei has suggested calling it "Limbo".

I'd like to throw my hat in the ring with "PRgatory".

On Tuesday, 16 January 2018 at 11:13:35 UTC, Walter Bright wrote:
> On 1/15/2018 8:41 PM, Brad Roberts wrote:
>> What I'd like to see as a first step is to draw a line in the sand, every new pull request as of 1/1/2018 should be handled to resolution, period.  Don't let them get stale.
>
> We've tried that again and again. It presupposes that all PRs have a resolution. They do not - I mentioned 4 common reasons why in the opening post.
>
> There is another problem here. I just got back from POPL 2018 with a list of ideas I want to get going on. I have made zero progress on any of them, because I spend all the time reviewing other peoples' work, dealing with can you please look at these bug reports

We all know this by heart. This is counter-productive to the extend of questioning GitHub as development model. Don't get me wrong, contributions are great, and we received an amazing amount of talented work this way.

It becomes problematic though when it prevents us from setting an agenda and follow that through. Focus is the most precious resource we have, but contributions tend to come as random streams of particular interests.

What works for me, is to batch issues and work on one and only one domain at a time for a prolonged period (~2 weeks). Working 2 weeks only on PRs, but then having 4 or 6 full weeks for planned work would be a useful approach.
For this to work we need to align better, so that most of the time we have someone working on PRs.

Another barrier with this approach, our current review pipe is so clogged that it's hard to sustain pace during a focused effort. To overcome this we also need to plan for review time alongside focused work. For example I already know that my planned work on limited lifetimes will need a lot of feedback from you, and that I'll roughly start to work on it early February. Likewise I gave Sönke a heads-up to soon expect plenty of dub and dub-registry pulls.
Just join us on dlang.slack.com, it's a very undisturbing way for an informal notice.

To address the backlog we prolly should prioritize every PR that cannot be reviewed within a minute.
This is actually rather simple work once you start to batch it. I went through dlang-bot's bug tracker yesterday, took less than an hour to prioritize https://github.com/dlang-bots/dlang-bot/issues (on topic https://github.com/dlang-bots/dlang-bot/issues/47).

And lastly a reminder that we should give early approvals when there are only nits and clearly request specific changes or close PRs with fundamental design issues.
There is nothing worse than getting a bit of review, only to get the rest of it once you come back with the requested changes.

On Tuesday, 16 January 2018 at 20:04:35 UTC, Sebastian Wilzbach wrote:
>
>
> On 2018-01-16 19:32, qznc via Dlang-internal wrote:
>> On Monday, 15 January 2018 at 01:10:58 UTC, Walter Bright wrote:
>>> PRs banished to the Phantom Zone would have to have an explanatory comment saying why.
>> 
>> I would suggest an alternative approach: Automate it.
>> 
>> 1. When there is no activity on a PR for one month, the bot pings the
>> author and asks for the status.
>> 2. When there is no activity on a PR for four months, the bot pings
>> three random other people and asks them to adopt the PR. Maybe it can
>> be more clever then "random" and look at what files are touched or
>> something.
>
> The bot currently marks a PR has stalled if there hasn't been any activity within three months
>
>> 3. When there is no activity on a PR for half a year, the bot leaves a
>> short comment "no activity for half a year." and closes the PR.
>
> This would be trivial to implement, but I haven't done this so far, because there hasn't been a consensus on this.
>
>> If someone wants to keep a PR alive it only takes single comment every
>> six months.
>> 
>> I doubt that there is any PR which got zero activity for six months,
>> yet was successfully merged later.
>
> Hehe, it does happen sometimes, e.g.
>
> https://github.com/dlang/dmd/pull/4988
>
> Also Rainers recently rebased #2480:
>
> https://github.com/dlang/dmd/pull/2480
>
> But obviously not too often.

It actually quite common (https://github.com/dlang-bots/dlang-bot/pull/109 ;)).
Someone has a good and jumps to work on it, then gets distracted, becomes busy with sth. else, or the problem turned out much more complicated. Now the initial motivation is gone and the PR gets abandoned.
But a year or so later the topic comes up again and the PR is revived.

See dub watch for example, it's almost 3 years old https://github.com/dlang/dub/pull/446.
The initial design was wrong, but the PR still contains lots of useful information and code.
It could also help someone else to pick up the task.
Sure we could just close it, but it feels like abandoning that idea altogether.

On 1/16/2018 2:22 PM, Walter Bright via Dlang-internal wrote:
> On 1/16/2018 12:35 PM, Brad Roberts wrote:
>> What's your goal with the PZ?
> 
> The current system is a binary "it's good enough to merge now" and "it has zero merit whatsoever and should never be looked at again."

What?  It's far from binary today.  I don't understand this assertion at all.  Particularly since you directly counter it immediately.  Did you mean something closer to:  The current system has two extremes... ?  If so, then I don't agree with the starting point of "good enough" either.

> There's a continuum in between those poles, and that is what the PZ is for. I listed 4 reasons a PR may fall in that continuum.

The PZ formalizes an option which effectively adds to the set of paths a PR can take.  The issue is that if it just becomes a new place to house roughly the same set of pull request then all that's been accomplished is hiding the problem a layer deeper and pissing off submitters even more.  Now, not only is their hard work not being looked at, it's actively (be it by a human or automation) been pushed even further into the shadows.

For what it's worth, I'm not trying to say that there's NO pulls for which closing is the right answer, clearly some need to be.  I'm also not saying that for some of that set that there's enough value in the pull requests that making sure it's easy to look back isn't a good idea.  What I'm trying to say is that those should be extremely rare.  Rare enough that they don't make up a meaningful percentage of the current queue.

>> Is it to have the github pull queues kept to a tiny number of open and active requests?
> 
> One is to keep the autotester from wasting resources on those.

No resources are wasted in the auto-tester due to old requests.  Stale requests are further down in the queue.  Time spent on them is time that if it wasn't building those it would sit idle.

> The other is the political/marketing issue of "what are those old PRs doing there? They should just be merged or closed!"

Much like the size of the bugzilla db, from a basic PR standpoint, I don't care all that much.  I care a whole lot more about the effect to the author of a PR who's work has not progressed.  The end results might be the same, but the motivation is very different.

That said, I'm not an active developer and my opinion shouldn't be weighed all that heavily here.  Go ahead, try it.  I suspect you're not going to be satisfied with the results.

On 1/16/2018 6:57 PM, Brad Roberts wrote:
> What I'm trying to say is that those should be extremely rare.

Let's frame it another way.

There have been 7733 PRs for dmd. Of these, 140 are open. Let's say 100 of those are stale and candidates for the PZ.

100/7733 is 1.3% accumulated over a period of what, 10 years? Isn't that "extremely rare"? But people do not look at the resolved PR counts, just the count of unresolved ones, and perceive it as a big problem.

On Tuesday, 16 January 2018 at 20:35:30 UTC, Brad Roberts wrote:

>> There is another problem here. I just got back from POPL 2018 with a list of ideas I want to get going on. I have made zero progress on any of them, because I spend all the time reviewing other peoples' work, dealing with can you please look at these bug reports, and one entire evening rebasing PRs due to the torrent of refactorings.
>
> Yup, pretty standard life of a senior developer.  I'm sure you're used to it.

+1000

>> You're asking me to redouble my efforts at looking at PRs. It's just not possible.
> Every single place I've ever worked getting code reviewed and checked in has taken precedence over writing new code.

+1000

/Paolo

On Tuesday, 16 January 2018 at 23:02:09 UTC, Andrew Benton wrote:
> I'd like to throw my hat in the ring with "PRgatory".

Very nice idea :). We should stick to Italian though, PRgatorio.

http://www.worldofdante.org/media/images/more/full/Purgatorio.jpg

On Wednesday, 17 January 2018 at 02:57:38 UTC, Brad Roberts wrote:

> Much like the size of the bugzilla db, from a basic PR standpoint, I don't care all that much.  I care a whole lot more about the effect to the author of a PR who's work has not progressed.  The end results might be the same, but the motivation is very different.
>
> That said, I'm not an active developer and my opinion shouldn't be weighed all that heavily here.  Go ahead, try it.  I suspect you're not going to be satisfied with the results.

I agree with Brad: Dlangland desperately need to motivate active developers, and grow them to core developers.

Again, I strongly suggest to concentrate the effort on that, instead of loosing time in what Walter called "marketing".

But as Brad, I'm not an active developer, neither a contributor, I've just followed the D communiti since a decade: but I've the same suspect, you're not going to be satisfied with results.

/Paolo

On Wednesday, 17 January 2018 at 03:23:31 UTC, Walter Bright wrote:
> On 1/16/2018 6:57 PM, Brad Roberts wrote:
>> What I'm trying to say is that those should be extremely rare.
>
> Let's frame it another way.
>
> There have been 7733 PRs for dmd. Of these, 140 are open. Let's say 100 of those are stale and candidates for the PZ.
>
> 100/7733 is 1.3% accumulated over a period of what, 10 years? Isn't that "extremely rare"? But people do not look at the resolved PR counts, just the count of unresolved ones, and perceive it as a big problem.

It is a problem, reducing it to a percentage doesn't do it justice. Let's say you are in a house and in that house there are 50 zombies and those are the only zombies that exist on Earth. Does it really matter if that's only 0.0000006718624% of the population? No, you are still going to have to fight through 50 zombies to get out of the house.

Clearing 140 pull requests is no small task, especially for how few people there are. That's time and effort that is going to have to be redirected from other tasks. Not only that, but because some of these pull requests are years old, it's going to take even longer for them to be fixed up and brought up to date.

Percentages have their place, this isn't the place for it right now.