Re: [safe_numerics] needs a mini-review

From: Robert Ramey (ramey_at_[hidden])
Date: 2020-10-09 15:21:54

On 10/8/20 9:28 PM, Andrzej Krzemienski via Boost wrote:
> czw., 8 paź 2020 o 19:18 Robert Ramey via Boost <boost_at_[hidden]>
> napisał(a):
> 
>>
>> As author of two boost libraries, I can offer a little perspective.
>>
>> My second offering is the safe numerics library. It had some loose ends
>> but was considered sufficiently implemented to accept. And so it was.
>> Apparently, it was not that interesting for enough people to craft
>> objections to.  WHat happened next?
>>
>> You guessed it.  cleaning up the "loose ends", unraveled the whole
>> library implementation.  Even worse, it impacted that type requirements
>> (concepts) which the library specified.  It was almost like starting
>> from scratch again. About the only thing left unscathed was the
>> documentation. When I got it back together, I merged into boost
>> development branch.  .... and the it is.  No one objected to this.  In
>> fact, no one even noticed. Did I do the right thing? Should I have asked
>> for another review?  You decide.  I'm wondering if it got accepted in
>> large part due the documentation.  As usual, I don't have point here -
>> just like to keep the pot boiling.
>>
> 
> The review result of Safe Numerics library was that it is conditionally
> accepted:
> https://lists.boost.org/Archives/boost/2017/03/233511.php
> 
> My understanding of the process is that as the next steps:
> 1. The author applies the indicated conditions and reports readiness for a
> mini-review.
> 2. A mini-review is scheduled. Its purpose is to determine if the
> conditions have been satisfied
> 3. If the review passes, the author has green light to release their
> library as part of Boost libraries.

Of course this is the question at hand right now.  If there are 
conditions, how should the be enforced?  In many/most cases the 
conditions are pretty simple and straight forward so one has confidence 
that the author will simply make the updates and check in the library. 
I believe that this is the most common case.

But sometimes, even though the conditions might seem straight forward, 
implementing them might end up being a lot more work than first meets 
the eye.  In the case of safe numerics, I did manage (in my personal 
estimation) to implement all the stated conditions, but the effort 
required to do so was far beyond what I had anticipated.  I also ended 
up refining ideas which were actually incorrect but never detected/noted 
in the review.  I fixed that stuff also.

So to my mind, the system worked well and as intended.

a) I made the prototype library including documentation and carefully 
specified API
b) It got reviewed
c) It got accepted subject to some conditions
d) I re-implemented it to fulfill all the conditions, fix discovered 
mistakes, fill in missing pieces.
f) checked it in.

The "missing step" would be e) - get a mini-review.  Honestly this just 
never occurred to me.  I had included all the conditions as issues on 
github and documented my changes there so I'm pretty confident that I 
addressed them all.  I probably should have asked you to verify/certify 
that the conditions were met.  But it didn't occur to me.

In short, I'm arguing that a "mini-review" is generally not necessary 
and I don't think it's documented as as a step in the boost "process".

> 
> I read your message as saying that safe_numerics is ready for a
> mini-review. Is that correct?
> 
In this particular case, I'd be happy to respond to anyone who wants to 
subject the current version of the safe numerics library to a 
mini-review.  In general, I love talking about my stuff.

> Regards,
> &rzej;
> 
> _______________________________________________
> Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
>