Re: [scope] Scope review starts on November 26th

From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2023-11-30 15:52:31

On 11/30/23 00:35, Janko Dedic via Boost wrote:
> On Wed, Nov 29, 2023 at 4:56 PM Andrey Semashev via Boost <
> boost_at_[hidden]> wrote:
> 
>> I don't see the current definition of BOOST_SCOPE_FINAL as a deficiency
>> of the interface compared to passing the capture list in the macro
>> arguments. The latter doesn't offer any advantages compared to the
>> current definition, and in fact is more limiting as it only allows for
>> lambda functions for the scope guard action.
> 
> It offers better syntax.

Better in what way, other than your personal preference? Can you
describe objective advantages of your proposed syntax and why the
current BOOST_SCOPE_FINAL doesn't work for you?

>> Besides, note that the lambda definition syntax has evolved over time
>> and may now include additional elements such as template parameters,
>> noexcept and mutable qualifiers and the trailing return type. There may
>> be new syntax additions in the future, which could be incompatible with
>> the macro definition. Although some of the syntax elements are useless
>> for the purpose of scope guards, at least noexcept and mutable are
>> meaningful and would look awkward if BOOST_SCOPE_FINAL accepted the
>> capture list in its parameters.
> 
> I only suggested adding [&] to the macro, or adding a macro that includes
> the [&].

Sorry, but no. That would limit user's options for no reason, and
binding variables by value is not uncommon. Users of Boost.Scope should
have the full range of function object definition ways that is permitted
by the language.

>> scope_success isn't as widely useful as e.g. scope_fail, but you do
>> occasionally want it to schedule some commit steps when the enclosing
>> function succeeds, if there are multiple exit points and a lot of the
>> function-local state that is difficult or inconvenient to pass to a
>> separate function. You can sometimes work around it with a goto, and
>> scope_success is just another, possibly cleaner way to do this.
> 
> Do you have a code example? I'm really curious about this use case.

I had something like this in mind:

  void foo()
  {
    auto private_data = do_something();
    scope_success commit_guard{[&private_data]
    {
      commit_data(private_data);
    }};

    if (condition1)
    {
      do_something_on_condition1(private_data);
      return;
    }

    if (condition2)
    {
      do_something_on_condition2(private_data);
      return;
    }

    bool succeeded = complete_processing(private_data);
    if (!succeeded)
    {
      commit_guard.set_active(false);
      return;
    }
  }

In the above pseudocode, private_data may be any number of data items of
various types, including those defined locally in foo(), and function
calls like commit_data, do_something_on_condition1, etc. would be pieces
of code working on that and other data.

>> Also, and this pertains to some other questions you asked in your
>> review, Boost.Scope tries to provide components that are defined in the
>> Library Fundamentals TS with a compatible interface. scope_success is
>> defined in the TS, so Boost.Scope also provides it.
> 
> I very much dislike this reasoning. Why should Boost copy std even when std
> is making a mistake? Boost should rely on its review process to judge
> libraries. "std does the same" is not a valid argument (especially in the
> light of recent discussions on the list about the quality of the WG21
> "review process"). I can understand if it's the stated goal of the library
> to copy std, but (1) scope guards are not even standardized yet and (2) for
> copying std there is Boost.Compat.
> 
> At the end of the day, users are robbed of better libraries because "std
> has already done it this way."

If you read earlier commits in this review, you'll see that there are
people who would prefer Boost.Scope to *exactly* reflect the TS with no
extensions or deviations. Although I disagree with that point of view, I
still think there is value in providing interface that is *compatible*
with the TS, if possible, so that users, at least potentially, have the
option to switch between Boost.Scope and std, should they need to for
whatever reason. Another reason is that this compatibility reduces the
element of surprise and makes learning the library (whether Boost.Scope
or std) easier.

If this review shows that no such compatibility is desired by the
community, I can remove some elements of the library that are deemed
unwanted or detrimental to its quality.

>> If you want to enforce that your action never throws, you can declare
>> its operator() as noexcept:
>>
>>   BOOST_SCOPE_FINAL []() noexcept { /* I guarantee I won't throw */ };
> 
> Does anyone really want to write code like this?

I see nothing wrong with this code. I actually wrote code like this,
where the no-throw guarantee was important.

>> Movability is necessary for the factory functions, at the very least.
> 
> I can see why. C++17 fixes this "defect". You can technically rely on
> reference lifetime extension, but that doesn't look as good.
> 
> auto&& guard = make_scope_exit(...);

Reference lifetime extension has nothing to do with it. Without RVO, in
order to return the scope guard from the factory function by value, its
copy or move constructor needs to be invoked.

Besides, TS scope guards are move-constructible.

>> And I do specifically prefer set_active() over a pair of methods like
>> activate()/deactivate() because passing a bool is fundamentally more
>> flexible than having two distinct methods. You can receive that bool
>> from elsewhere and just forward to the method instead of testing it
>> yourself. Like in the example above, you could have written it like this:
>>
>>   void add_object(std::shared_ptr< object > const& obj)
>>   {
>>     // Create a deactivated scope guard initially
>>     std::set< std::shared_ptr< object > >::iterator it;
>>     boost::scope::scope_fail rollback_guard{[&, this]
>>     {
>>        objects.erase(it);
>>     },
>>     false};
>>
>>     bool inserted;
>>     std::tie(it, inserted) = objects.insert(obj);
>>
>>     // Activate rollback guard, if needed
>>     rollback_guard.set_active(inserted);
>>
>>     obj->on_added_to_collection(*this);
>>   }
>>
>> which wouldn't be as simple with two different methods.
>>
> 
> This looks much better honestly. Why isn't this variant used for the
> example?

I can update the example in the docs.

>>> I did not like the locked_write_string example. Reading this code
>> locally:
>>>
>>>     // Lock the file
>>>     while (flock(fd, LOCK_EX) < 0)
>>>     {
>>>         err = errno;
>>>         if (err != EINTR)
>>>             return;
>>>     }
>>>
>>> it's hard to get what's happening, why we're setting err and how big of
>> an
>>> impact that has on function behavior. A better way to write this code
>> would
>>> be to simply extract this piece into its own function:
>>>
>>>         err = errno;
>>>         if (err != EINTR)
>>>             return;
>>>
>>> together with the ec = std::error_code(err, std::generic_category()),
>> which
>>> doesn't require a scope guard.
>>
>> Sorry, I don't understand your suggestion. How would this help with
>> other return points of the function?
> 
> It helps with all the return points in the example function.

I don't see how. Can you provide a code example?

Note that the checks for EINTR are made in different contexts. Some of
them cause return, other - restart loop iteration, third - affect
error_code initialization.

>>> I think a similar utility would be more useful in form of a class
>> template,
>>> for example:
>>>
>>> resource<HANDLE, INVALID_HANDLE_VALUE, ::CloseHandle> windows_handle;
>>> resource<SDL_Window*, nullptr, ::SDL_DestroyWindow> sdl_window_handle;
>>>
>>> This would be very useful for concisely implementing RAII wrappers over C
>>> handles (you don't have to manually write destructors, move constructors
>>> and move assignment operators).
>>
>> This would require C++17, if I'm not mistaken (or when were the auto
>> non-type template parameters introduced?). I don't think it offers much
>> advantage compared to the current unique_resource with resource traits.
> 
>  The advantage is doing the same thing in 1 line instead of 20 lines of
> code, at which point writing the destructor and moves manually has a better
> ROI than implementing traits classes.

This would also prohibit resource types with non-constexpr constructors,
as well as non-default-constructible resources. I don't like this tradeoff.