Re: [boost] [afio][v1.30 "C++ Now edition" released]

Subject: Re: [boost] [afio][v1.30 "C++ Now edition" released]
From: Niall Douglas (s_sourceforge_at_[hidden])
Date: 2015-03-18 11:11:08

On 18 Mar 2015 at 9:41, Pete Bartlett wrote:

> > This release took far longer than I originally intended mainly due to 
> > the 
> > extensive debugging of weird filing system races, and for six weeks 
> > now I've 
> > been working till 5am every night after my day job. I cannot tell you 
> > how glad I am to see this release out!
> 
> You are obviously very proud of this library and I believe rightly so.
> You've worked hard on it and in terms of things like soak testing it
> probably sets a new high watermark for Boost libraries.

I'm proud of what I've been able to do with the very limited 
resources available to me, but I feel not especially proud of 
shipping a library which isn't well tested in absolute terms. For 
example, almost all the testing is functional rather than unit 
testing (lack of time to write more), plus the soak tests run for 
minutes instead of the 6 or 24 hours I'd much prefer and on only quad 
core hardware with local drives rather than the 16-32 core 
distributed over Samba/NFS network shares that it should be. I'm also 
not happy with the testing of the (too) many build options, the lack 
of testing of std::bad_alloc correctness (right now I claim exception 
safety except for bad_alloc), and there is no testing of localised 
resource leaks, only global testing.

Ok, so like most Boost maintainers I see all that I failed to do 
before any of the achievements. Still, it could be a lot better than 
it is, and a bug in AFIO means data loss for users. In the end, we 
all have day jobs and families.

> You're also a
> highly active member of this community. So why is the library 
always
> "proposed".... When is there going to be a review? :) 

When a review manager volunteers themselves!

Antony nearly did a few months ago, but this release was in progress 
and at that time I had no ABI nor API versioning in AFIO, so I could 
not propose an even semi-stable API at that time. In the v1.3 
release, both ABI and API are versioned and there is a unit test 
which compiles two intentionally dimorphic and ABI incompatible 
copies of AFIO into the same translation unit and runs both sets of 
unit tests for each to make sure they don't collide. They don't, so I 
am now sure v1.3 of AFIO can coexist without issue with v1.4 of AFIO 
or any other version.

So I now have no objections, and furthermore no major work will be 
done to AFIO until after CppCon in September. Still, asynchronous 
file i/o and race free filesystem is an extremely niche use case. I 
am not aware of anyone who uses AFIO in their code. That has surely 
to count against the wisdom of including it into Boost official as 
without a user base, who can really say if a design or implementation 
is any good during review?

Niall

-- 
ned Productions Limited Consulting
http://www.nedproductions.biz/ 
http://ie.linkedin.com/in/nialldouglas/