On 10 Sep 2014 at 20:00, Antony Polukhin wrote:

...

I would just love to have you manage the review of AFIO. Unfortunately, I am just about to begin its monadic continuations framework which integrates ASIO, AFIO, future-promises and expected<T, E> into one, so I suspect that work will transform any review of AFIO significantly.

I'd gladly manage the review of AFIO.

When I was looking through the sources of AFIO last time I've got an impression of ongoing heavy development. That's why there's no [afio] in the title of this message.

There has been almost no new code since it entered the review queue last October. What was done was a very substantial *reorganisation* to fit modular Boost so it would behave as any other Boost submodule. That took weeks :( and probably appeared as if there was heavy development when in fact no new code was written. The engine is currently at v1.22, and the v1.2x engine will be the last to support VS2010. v1.3 will use the new concurrent_unordered_map I just recently finished, and the new no-alloc future-promise I am just about to start work upon. I am hoping this new v1.3 engine will reduce average op latency from 75 microseconds (~300,000 cycles) by a factor of 10, with a bit of luck by a factor of 20 down to about 15,000 cycles average. Moreover, there will be no memory allocation and deallocation at all in the v1.3 engine, which would be neat. However, as you're ready to review manage now, and the v1.3 engine is likely to not arrive before the end of 2014, please do look through the docs now relocated to http://boostgsoc13.github.io/boost.afio/. I have made some improvements based on what Robert said about them. If you think the library is ready for review now, do let me know. Niall -- ned Productions Limited Consulting http://www.nedproductions.biz/ http://ie.linkedin.com/in/nialldouglas/

Antony Polukhin wrote

2014-09-10 20:55 GMT+04:00 Niall Douglas <

s_sourceforge@

>: <...>

...

However, as you're ready to review manage now, and the v1.3 engine is likely to not arrive before the end of 2014, please do look through the docs now relocated to http://boostgsoc13.github.io/boost.afio/. I have made some improvements based on what Robert said about them. If you think the library is ready for review now, do let me know.

I'll take a closer look to the library and write a personal note with comments/notes as soon as I get some free time.

But first of all, I'll take care of the Compute review :-)

I'm wondering if perhaps you've got confused about the differing roles of review manager vs reviewer. IRC a review manager doesn't review libraries himself, he reviews the reviews of others, weighs them and arrives at a acceptance/rejection. Think the US court system judge - review manager jury - reviewers. If you actually want to review these libraries, consider posting you're review to the boost library incubator and encouraging others to do the same. Getting a review manager is a separate task. Hopefully, this would be easier of there were several reviews already prepared. Robert Ramey -- View this message in context: http://boost.2283326.n4.nabble.com/process-sorting-singularity-compute-other... Sent from the Boost - Dev mailing list archive at Nabble.com.

Andrey Semashev-2 wrote

On Saturday 13 September 2014 13:25:30 Robert Ramey wrote:

...

I'm wondering if perhaps you've got confused about the differing roles of review manager vs reviewer. IRC a review manager doesn't review libraries himself, he reviews the reviews of others, weighs them and arrives at a acceptance/rejection.

I don't think there is a rule on this. The review manager is free in choosing means to form an opinion of the library, including reviewing it himself. In fact, I'd say, the review manager is expected to acquaint himself with the library to form educated opinion.

I looked at http://www.boost.org/community/reviews.html and it describes the duties of the review manager and reviewers a as separate roles. On the other hand, it doesn't specifically proscribe one person performing both roles in the same review. So technically you might be correct. But in 12 years of being associated with boost, I've never seen this happen. (not that It might have - I can't watch everything). And I believe that there is a good reason for not mixing these roles. There fundamentally different - one is to advocate for a position while the other is to weigh the differing opinions and select the best representation of a consensus. It's very similar to the legal system in English speaking countries and though I wasn't around when it was invented, I'm doubting that it's a coincidence. But maybe we can avoid having to resolve this now. If you add reviews to these libraries in the Boost Incubator, you might be in a position to encourage other people to add their own reviews. With enough reviews ready, the job of review manager will likely be easy to fill. So lets start with this. Robert Ramey -- View this message in context: http://boost.2283326.n4.nabble.com/process-sorting-singularity-compute-other... Sent from the Boost - Dev mailing list archive at Nabble.com.

On 9/13/2014 4:52 PM, Andrey Semashev wrote:

On Saturday 13 September 2014 13:25:30 Robert Ramey wrote:

...

Antony Polukhin wrote

...

2014-09-10 20:55 GMT+04:00 Niall Douglas <

s_sourceforge@

>: <...>

...

However, as you're ready to review manage now, and the v1.3 engine is likely to not arrive before the end of 2014, please do look through the docs now relocated to http://boostgsoc13.github.io/boost.afio/. I have made some improvements based on what Robert said about them. If you think the library is ready for review now, do let me know.

I'll take a closer look to the library and write a personal note with comments/notes as soon as I get some free time.

But first of all, I'll take care of the Compute review :-)

I'm wondering if perhaps you've got confused about the differing roles of review manager vs reviewer. IRC a review manager doesn't review libraries himself, he reviews the reviews of others, weighs them and arrives at a acceptance/rejection.

I don't think there is a rule on this. The review manager is free in choosing means to form an opinion of the library, including reviewing it himself.

Using Robert Ramey' analogy this would be the same as a judge in a court case also serving as a juryman and casting his vote. I do not believe that should ever be the situation either in a law court or during a Boost library review.

In fact, I'd say, the review manager is expected to acquaint himself with the library to form educated opinion.

Of course, but this does not mean that the review manager should be reviewing the library. A good review manager's responsibility is to be as impartial as he can be in deciding, based on the reviews and his knowledge of the library functionality, whether it should be accepted or not.

2014-09-14 0:25 GMT+04:00 Robert Ramey <ramey@rrsd.com>:

Antony Polukhin wrote

...

I'll take a closer look to the library and write a personal note with comments/notes as soon as I get some free time.

But first of all, I'll take care of the Compute review :-)

I'm wondering if perhaps you've got confused about the differing roles of review manager vs reviewer. IRC a review manager doesn't review libraries himself, he reviews the reviews of others, weighs them and arrives at a acceptance/rejection. Think the US court system

judge - review manager jury - reviewers.

I'm not going to write a full review of the library, I'm just trying to satisfy the first bullet for Review manager: http://www.boost.org/community/reviews.html#Review_Manager In other words, there is a need to look through the sources, docs and build system, make sure that installation notes are sufficient, tell the author about missing/weak parts. For example: Compute library looks good, however its docs have no performance section. This is important, because in most cases users will use Compute library to speedup their code. If a lot of notes appear during such "preliminary investigation", then it's worth writing a review/comment to Boost Incubator. Compute library is not such case. -- Best regards, Antony Polukhin

Rob Stewart-6 wrote

...

...

judge - review manager jury - reviewers.

There's a problem with the analogy because we do expect the Review Manager to ensure the case is ready for court and the defense is fully prepared, so to speak.

where's the problem? In the legal world the judge makes sure that the parties have time to prepare, will grant them more time if he believe its justified, rule on what evidence will be admitted etc, etc. I decides, but generally doesn't participate in the argument. It's an explicit attempt to separate the roles of advocate from those of judge. This analogy has worked well for us. I don't remember any case where authors of rejected libraries have blamed the reviewer for unfair treatment. To me its an indication that someone "got it right". But of course there are still opportunities for improvement. I would like to see the following rule added. A review can't proceed until there are N reviews already prepared and sitting in the incubator. I'm thinking a reasonable N would be 4? This would mean a) more people have incentive to use the incubator. b) the preliminary effort the reviewer has to go through would be a little easier. c) the incubator has a low bar of non-controversial requirements. A library which can't pass this bar is not really reviewable. I think this tiny addition would make the reviewers job a lot easier and save wasted effort with no discernible downside. One more thing. reviewers recommend that the library be rejected, accepted, or accepted subject to conditions. I think lots of people see this as a "vote". It suggests that the decision is made by vote. This is not at all true and stating this is very misleading. One of the key points of the review process is that one person weighs all the considerations and is personally responsible for making the decision. It's a heavier burden than just counting votes. Its clear from the review results emails that the reviewers take this responsibility seriously and that Boost has gained mightily from this. I would like to discourage reviewers from referring to their recommendations as "my vote" rather than "my recommendation". I realize that I'm blowing this out of proportion, and I can't really understand myself why I'm picking on such a nit. So please forgive me for indulging in a very personal quirk here. Robert Ramey -- View this message in context: http://boost.2283326.n4.nabble.com/process-sorting-singularity-compute-other... Sent from the Boost - Dev mailing list archive at Nabble.com.