Subject: Re: [boost] Really dumb off topic question regarding variant
From: Niall Douglas (s_sourceforge_at_[hidden])
Date: 2019-02-22 23:20:30
> I see optional, expected, outcome and .. (monad?) as just special cases
> of variant. for example
> template<typename T>
> using optional = variant<false_type, T>;
> Why is it necessary to have all these types separately implemented?
> Can't there be some sort of "base" type which can be used to implement
> all these others?Â Wouldn't this approach make things much simpler to
> review, maintain, and use?
The reason is that there are different tradeoffs at work. I cover this a
bit in https://ned14.github.io/outcome/history/#outcome-v2 where I
explicitly mention variant2, but to summarise:
1. Outcome has a simple implementation with minimum SFINAE, a fixed ABI
and aims for low impact on build times in order to maximise usefulness
in public interfaces in large codebases.
2. variant2 has a less simple implementation, enabling more flexibility,
better code density, but at the cost of build time impact.
3. A fully conforming Expected can be implemented easily using either
Outcome or variant2. See
4. Optional is an interesting one. Outcome uses an internal optional
implementation because reusing Optional wasn't efficient. Specifically,
I needed more than a single bit discriminant, and Optional wasn't
reusable for that with doing UB.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk