Re: [boost] usage of auto in tutorials

Subject: Re: [boost] usage of auto in tutorials
From: Stephan Menzel (stephan.menzel_at_[hidden])
Date: 2019-03-19 10:59:05

Hello Andrzej,

On Tue, Mar 19, 2019 at 10:26 AM Andrzej Krzemienski via Boost <
boost_at_[hidden]> wrote:

>
> The initial page you linked to has the following two usages of auto:
>
> auto process(const string& content) noexcept -> outcome::result<int>;
> >
> > if (auto rslt = read_data_from_file("config.cfg"))
>
>
> Are you referring to these?
>
> I agree that inside the if-statement it is better to spell the type
> explicitly.
> However, the auto in the function declaration doesn't seem to do any
> harm: it only allows me to put the return type at the end,
> so there is no deduction that is taking place.
>

Mostly, yes. I do understand what the code does but it kind of forces me to
do the mental indirection when reading. I do understand the trailing return
type syntax and why one could use auto there.
But for some people this looks very unusual. For example, because they are
not aware of this new syntax or they have not applied it and still write
functions the old fashioned way. Some (like me) work in environments where
the use of auto is banned altogether.
I'm not arguing pro or contra auto or trailing return types or indeed even
the runtime implications.
All I'm saying is, for a tutorial I feel one should choose a simple way of
expressing things with a high likelyhood of being understood by many
immediately.
Especially here where the actual return type is the most interesting part
as the lib is essentially about returning things. Having it as the first
part of a statement would help.

> Are there any other usages of auto that cause your concern? In the
> examples that I was able to have a look at autos are avoided, like:
>
> https://www.boost.org/doc/libs/1_70_0_beta1/libs/outcome/doc/html/tutorial/essential/result/inspecting.html

Yes, I just had a look atother parts of the tutorial and I find others
easier to read. OK, I'm an old fashioned guy and I know these days some
folk have other ways of coding but the quick start tutorial should be as
simple as possible.

> There is one auto in the explanation of OUTCOME_TRY:
>
> https://www.boost.org/doc/libs/1_70_0_beta1/libs/outcome/doc/html/tutorial/essential/result/try.html
>
> But in this case we are explaining what the macro expands to, and part
> of the contract of the macro is that type deduction is performed,
> so I guess the auto fits better.
>

Yes it does. I think parts of the tutorial may and should go into the
'fancy new stuff' and take advantages of newer language features. No
problem there. Thanks for understanding my concerns. I'm going to have a
close look at outcome as soon as 1.70 releases. I was looking forward to it
ever since I used the outcome implementation in the AWS SDK. Which is
incredibly header intensive and cumbersome in my opinion but interface wise
not too far from what you do. I was curious to see Boost's take on this.

Cheers,
Stephan