Reduce Callback.h using C++11

[kjbracey]

Description

The bulk of Callback.h was auto-generated as 6 specialisations, handling zero to five arguments.

This can now be handled without specialisation using C++11 variadic templates, reducing the file from 4,900 lines to 900 lines.

This should reduce compilation time, and offset potential increases from use of <type_traits> or a local mbed_cxxsupport.h equivalent.

Several other improvents to Callback are possible and/or desirable with C++11, such as the ability to store lambdas, but this commit is purely the variadic simplification.

Pull request type

[ ] Fix
[X] Refactor
[ ] Target update
[ ] Functionality change
[ ] Docs update
[ ] Test update
[ ] Breaking change

Reviewers

@pan-, @bulislaw, @geky

[kjbracey]

The header file is actually somewhat understandable after this - it is complex, but the previous version further obfuscated it by just repeating everything 6 times, meaning it was rather hard to visualise the structure.

C++11 could possibly simplify it further using type traits to cut out a bunch of the const/volatile overloading repetition.

[kjbracey]

Could get it down to 500 lines if all the deprecated stuff was dropped...

[pan-]

Look goods to me.
I've been working on a new implementation yesterday, that reduces the code size and simplify most of the (unnecessary) overloads and add other goodies. I will submit it once this first simplification goes in.

[ciarmcom]

@kjbracey-arm, thank you for your changes.
@bulislaw @geky @pan- @ARMmbed/mbed-os-core @ARMmbed/mbed-os-maintainers please review.

[evedon]

That's great, thanks Kevin!
I was discussing with the team this optimisation just a few days ago as this file was really huge! Did not know that you had plan to work on it :)

[kjbracey]

I've been working on a new implementation yesterday, that reduces the code size and simplify most of the (unnecessary) overloads and add other goodies. I will submit it once this first simplification goes in.

Cool. I'll hold off on my temptation to fiddle further then, and wait to review yours.

[pan-]

@evedon events/Event.h would also benefit from such simplification.

[kjbracey]

It would make sense to do the same thing to Event.h at the same time. I can do that now, if no-one else has already done it.

[evedon]

It would make sense to do the same thing to Event.h at the same time. I can do that now, if no-one else has already done it.

We have not worked on it so feel free.

[kjbracey]

Hmm, Event.h and EventQueue.h are much more complex, as they don't just have to forward variadic arguments, they have to store them (in the EventQueue::context<x><y> structure). Nowhere near as straightforward - solutions involve a bunch of extra standard library machinery like std::tuple, which we don't have for ARM C 5.

[evedon]

In that case, it is not worth doing it as part of this PR.

[geky]

Glad to see this come in 👍

[kjbracey]

Glad to see this come in 👍

No doubt this was what you were yearning to write in the first place. The hard part was generating the C++98 version!

Hmm, Event.h and EventQueue.h are much more complex,

Actually, they're not so bad. I think I can remove 95% of them - it's just the innermost struct context that will remain specialized. I'll make a separate PR for them.

[artokin]

@kjbracey-arm , what is status of this PR?

[kjbracey]

I think it's ready. Got a pile of design changes to follow, but this one's fine.

[artokin]

CI started

[mbed-ci]

Test run: SUCCESS

Summary: 11 of 11 test jobs passed
Build number : 1
Build artifacts