June 2019 - Boost - lists.preview.boost.org

[asio][coroutine] forbid_suspend blocks
by Vinícius dos Santos Oliveira 07 Aug '19

07 Aug '19

I've implemented a dumb linter using Python + clang API that allows you to mark certain blocks as “atomic in respect to coroutines” using C++11 attributes. Usage example follows: void f() { int yield; g(yield); [[bt::forbid_suspend]] { // Comment this call to make linter accept the code g(yield); t(); } } You can change bt::forbid_suspend to another identifier by tuning settings on the beginning of the script. The prefix forbid was inspired by Rust's linters[1] as you cannot disallow the linter from inside the forbid_suspend block. Some limitations follow: - A suspend point is identified by looking references to any id equals to yield in the last argument passed to function calls. It means that yield[ignored_ec] and alike will be correctly identified, but it needs to be named yield and it needs to be the last argument to the function call. - I haven't found a way to extract attributes related to COMPOUND_STMT nodes with clang API, so I manually look for token gaps and do a simple regex. It means you cannot use macros to [[bt::forbid_suspend]] as the macro won't be expanded. - You better call the script with '-x c++' and also ' -I/usr/lib/clang/5.0.1/include'. Like I've said in the beginning of the email, it's a *dumb* linter, which means it doesn't try to handle all cases. Therefore, I don't hope to do many changes to it or maintain it to work against newer releases of clang and I'm only uploading it to gist (not fullblow Github): < https://gist.github.com/vinipsmaker2/d930fbe5b7597432b021effe618da171>. It should also be easy to hack as it is very small. I'm releasing the linter under public domain. Enjoy. I have been using fibers in a daily basis on the last months of my job to solve networking problems on a gateway project. Some of the technologies I use are: - Boost.Asio and its spawn function. - A few custom synchronization primitives for the fibers (e.g. mutex, semaphore). - Eventually I'll have to write a custom spawn function that returns a joinable handle so I can kill the usage of a fiber::barrier in a few places. I've been using the linter for about a week already and looks fine. My use case are: - To prevent iterator invalidation in regions accessing shared variables. - To guarantee events order in algorithms that handle races (cannot yield while task YYY hasn't finished). The project I use it in is not an open source project, so I cannot talk much more about it. Also, I don't claim my solution is original. I wouldn't be surprised if the very same solution exists in the wild (please let me know if you know of a similar approach). This is not a new problem. It is know for some time. For instance, in P0171, Gor Nishanov noted this same problem: “In coroutines the suspend point is clearly marked with await, which tells the reader that something unusual happens in this function and allows the reader, for example, to confirm whether the lifetimes of the objects of interest align with the lifetime of the coroutine or not, whether some locks need to be acquired to protect some concurrently accessed data, and whether some locks need to be released before the execution reaches the suspend point.” — <http://open-std.org/jtc1/sc22/wg21/docs/papers/2015/p0171r0.html>, 2015 However, the solution he presented was to manually examine whole blocks of code without help from the compiler. I argue that the solution he presented is as error-prone as manually looking for exit paths that don't free acquired resources in languages lacking RAII. And I've been spoiled too much by RAII. Furthermore, I take this comment of his with highly suspicious look: “Since we are on the subject of maintenance nightmares, we would like to offer a conjecture that the absence of the await in P0114 is a likely source of many maintenance nightmares. Without a syntactic marker to signal to the person reading the code that something funny is going on, it is impossible to tell whether the following code is correct or not” — http://open-std.org/jtc1/sc22/wg21/docs/papers/2015/p0171r0.html If I had to give a vote to the proposal he was defending or Chris' proposal, I'd probably have chosen Chris'. [1] https://doc.rust-lang.org/reference/attributes.html# lint-check-attributes -- Vinícius dos Santos Oliveira https://vinipsmaker.github.io/

2 3

Failed reconnection
by JH 11 Jul '19

11 Jul '19

Sorry for keeping posting questions about ASIO TCP socket reconnection regarding to network interface changes, I am very much stuck at the moment. As I described in other emails, when one network interface WiFi failed, the network interface Ethernet takes over, the client application using boost ASIO TCP socket cannot detect the broken of WiFi, it continually sent messages to the remote server via the broken WiFi interface without triggering errors but obviously the server could not receive messages any more. So my solution is prior to send message, it ping to 8.8.8.8 via the network interface WiFi, if it failed, it called the SetConnection again to close the socket binding to WiFi interface and to start reconnection to bind to Ethernet interface: void SetConnection(void) { if (this->mSocket.lowest_layer().is_open()) { std::cout << "Set connection close socket" << std::endl; this->mSocket.lowest_layer().close(); } boost::asio::ip::tcp::resolver resolver(this->mIoService); boost::asio::ip::tcp::resolver::query query(this->mHost, this->mPort); this->mIterator = resolver.resolve(query); this->mContext.load_verify_file("ssl/ca.pem"); this->mSocket.set_verify_mode(boost::asio::ssl::verify_peer); this->mSocket.set_verify_callback(boost::bind(&client::verify_certificate, this, _1, _2)); boost::asio::async_connect(mSocket.lowest_layer(), this->mIterator, boost::bind(&client::handle_connect, this, boost::asio::placeholders::error)); } But it did not work, the error: handle_connect connect failed: Network is unreachable What I am missing here? How can I make the TCP socket reconnection? Thank you. Kind regards, - jupiter

2 3

Mother of all Variants
by Robert Ramey 05 Jul '19

05 Jul '19

I'm way out of my depth in the variant discussion. Seems to me it revolves around all the trade offs regarding design choices. Could we perhaps decide not to decide. Suppose we create a type which looks like: template< bool BasicGuaranteeSupported, bool StrongGuaranteeSupported, bool EmptyStateProhibited, bool AssignmentSupported, bool MoveSupported, class T ... > struct mother_of_variants{ // some trivial mp11 code }; Now we could define: using boost_variant = mother_of_all variants< bool true, //BasicGuaranteeSupported, bool true, //StrongGuaranteeSupported, bool true, //EmptyStateProhibited, bool true, //AssignmentSupported, bool true, //MoveSupported, >; std_variant, std_optional, std_expect, boost_outcome, etc. would be similarly defined. And some "extras" like no_except_variant, trivial_variant, etc. would be added. and of course my favorite candidate: using safe_variant = mother_of_all variants< bool true, //BasicGuaranteeSupported, bool true, //StrongGuaranteeSupported, bool true, //EmptyStateProhibited, bool false, //AssignmentSupported, bool false //MoveSupported, >; In addition, we'd have some type traits: mother_of_all_variants::no_runtime_allocation<mov>; mother_of_all_variants::no_double_instancen<mov>; mother_of_all_variants::is_mother_of_all_variants<mov>; etc. Just think that of all the time we've spent on discussions about optional, expected, std::variant, boost::variant, etc. At this rate, these discussions will keep hounding us forever. Of course I realize that the main purpose of having these variants of variants is to justify the discussions - but perhaps we might be ready for a new topic. Robert Ramey

17 57

Boost Version 1.47 not building with 2017.
by nunna, sarath babu (GE Healthcare) 03 Jul '19

03 Jul '19

Hello, 1.47 boost is not being built with VS2017. We want to get boost 1.47 built with VS2017, So that we can integrate them with our VS2017 code base. Please guide us, it will be a great help. Thanks, Sarath.

4 9

out_ptr review
by Glen Fernandes 30 Jun '19

30 Jun '19

[Edit: Some changes for clarity within, as it was pointed out on Slack that some of the phrasing is confusing. Also updated subject.] Many thanks to JeanHeyd for going through the Boost review process, and to Zach for managing the review. Briefly: I cannot justify acceptance for out_ptr just yet in its current form. I hope to see it return for a new review if the feedback from this review shapes the design so that some of these concerns are resolved. Second: This excludes "inout_ptr" (for which I had even more concerns) and just covers "out_ptr" which is the simpler and more common use case of the two. Below when comparing "what happens under the hood" with out_ptr against the hand written code that a user would write to use a typical C out-pointer API with smart pointers, this is what I'm comparing: [Typical code user would write to use a smart pointer]: Type* raw; Result result = TheCApi(args..., &raw); if (!Success(result)) { // failure (e.g. return, or branch and handle error) } Smart<Type, Deleter> smart(raw, etc...); [What would happen under the hood when using out_ptr]: /* First, 'smart' already has to exist here before TheCApi is called. Either as Smart<Type, Deleter> smart; or Smart<Type, Deleter> smart(nullptr, etc...); depending on deleter */ /* User writes: Result result = TheCApi(args..., out_ptr(smart)); or Result result = TheCApi(args..., out_ptr(smart, etc...)); depending on deleter */ Type* raw = nullptr; Result result = TheCApi(args..., &raw); if (raw != nullptr) { smart.reset(raw); // or smart.reset(raw, etc...); depending on deleter } if (!Success(result)) { // failure (e.g. return, or branch and handle error) } Concern 1. Having to define the smart pointer before TheCApi is called. Ordinarily this might be fine for some smart pointers and stateful deleters, but it may not be fine or desirable for others (depending on the deleter, and depending on the use case). Concern 2. The original user code initializes the smart pointer based on Success(result). The out_ptr version does so based on whether the raw out pointer was changed to something non-null by TheCApi. This might be fine for some cases, but it may not be for others. That it isn't the same as fundamental logic as the user code bothers me, though. Concern 3. The equivalent of the hand written user code is less overhead. Personally I'm of the opinion that this doesn't matter too, but here the author does say he has users who cannot abide by it, and that this facility needs support from e.g. std::unique_ptr to be viable. I'm not a fan of needing unique_ptr implementations to add support for this to make it comparable to some user's hand written code (And this also leads to concern 4 below). Concern 4. The UB optimization that the author wrote because of Concern 3 is a real concern to him. The UB optimization has no place in Boost, and one of which doesn't even work with many C++ library vendors' unique_ptr. (Fortunately the Boost reviews have already echoed this sentiment and I believe this is going away). Concern 5. Supporting other custom smart pointers requires effectively re-implementing out_ptr_t by specialization. I have come to see this as not viable. (Already mentioned in other reviews, and I echo this.) Concern 6. C APIs where the deleter (its state, logic, etc.) are governed by some other result from TheCApi. This is can be worked around in many cases, but the workaround is clumsy. Or insupportable, in other cases. i.e. The stateful deleter in the example above is always initialized with 'etc...' either before-hand via Smart<Type> smart(nullptr, etc...) or at call site with: TheCApi(args, out_ptr(etc...)); before analyzing other results of calling TheCApi. [Concern 7. This concern I had made notes about pre-review has already been eliminated, when the unconditionally noexcept destructor is now only conditionally noexcept.] Other thoughts: The suggestion by Emil lead to an interesting alternative idea to this problem domain, but I had no time to fully explore it. I don't think it would address all of these concerns. It has some other greater appeal, e.g. a way to call C APIs that materializes the smart pointers as results (along with C API return value) of an "Invoke-style" thing. Or even a way to transform a C API with out-pointers into a C++ callable that returns smart pointers (along with C API return value) via a "Bind-style" thing. Glen

2 1

out_ptr's formal review starts now!
by Zach Laine 30 Jun '19

30 Jun '19

Dear Boost community, The formal review of JeanHeyd Meneide's out_ptr library starts Sunday, June 16th and ends on Wednesday, June 26th. out_ptr is a library for making it easy to interoperate between smart pointers and traditional C-style initialization and allocation interfaces. It also enables doing so in a way that allows library authors to opt into speed optimizations for their smart pointers that give them performance equivalent to typical C pointers (see benchmarks ( https://github.com/ThePhD/out_ptr/blob/master/docs/out_ptr/benchmarks.adoc) and the Standard C++ proposal for more details). Online docs can be found here: https://github.com/ThePhD/out_ptr/blob/master/docs/out_ptr.adoc The GitHub repository is available here: https://github.com/ThePhD/out_ptr Also, there is an associated standards proposal: https://thephd.github.io/vendor/future_cxx/papers/d1132.html We encourage your participation in this review. At a minimum, please state: - Whether you believe the library should be accepted into Boost - Your name - Your knowledge of the problem domain You are strongly encouraged to also provide additional information: - What is your evaluation of the library's: * Design * Implementation * Documentation * Tests * Usefulness - Did you attempt to use the library? If so: * Which compiler(s)? * What was the experience? Any problems? - How much effort did you put into your evaluation of the review? All issues discussed during this review will be tracked on the library's GitHub issue tracker at https://github.com/ThePhD/out_ptr/issues. Regards, Zach Laine

11 31

out_ptr review
by David Sankel 29 Jun '19

29 Jun '19

# Whether you believe the library should be accepted into Boost No. Not until some of the major issues are addressed. # Your name David Sankel # Your knowledge of the problem domain I've had to wrap C APIs on multiple occasions. # What is your evaluation of the library's: ## Design The design of the `out_ptr` type is reasonable as a shorthand for calling certain C APIs. In my experience a use case for `in_out_ptr` has never come up. Given the subtleties of its correct (and likely rare) usage, I'd be in favor of leaving this out of the library altogether. Especially given that the semantics of when 'release()' is called are not specified. I agree with the other comments that the customization point is not minimally complete. Ideally a customization point would not require one to replicate all the functionality of the class I'm customizing and, instead, only require a few general purpose functions that that interface can make use of. ## Implementation There are two implementations of `out_ptr`. The one called "clever" invokes undefined behavior as a means to squeeze out some extra performance. I do not think the performance overhead of using this library with the "non clever" implementation justifies the potential portability and and other risks associated with depending on undefined behavior working a particular way. It looks like "clever" mode was #ifdef'ed out, but removing it as cruft would improve maintainability of the implementation and remove the complexity of having an additional `core_out_ptr_t` class. The provided implementation works for smart pointers that follow conventions like the standard, but has some additional logic if the types happen to be `std::shared_ptr` or `boost::shared_ptr`. This hard coded special casing won't work with, say, a company specific shared pointer implementation. This relates to my earlier comment that the customization point should be simplified. ## Documentation The recommendation in the "caveats and caution" to not use out_ptr outside if/else conditionals goes a bit far in my opinion. It is common practice to make function calls that emit failure with return codes in conditionals, as in: ``` if (SQLITE_OK != sqlite3_open("test.db", boost::out_ptr::out_ptr(database, sqlite3_close))) { std::cout << "Problem opening the file"; } ``` For these cases it is safe to use `out_ptr` so I'd suggest rewording the recommendation to allow calls like this. "Almost all well-designed C APIs will set the user-provided output pointer argument to NULL/nullptr upon invalid parameters". This is a very difficult claim to make (what's the sample size?). I'd suggest changing "Almost all" to "Many". I don't understand what the issue is here with APIs not setting the out parameter on failure (which IMO is good practice especially as it relates to strong exception safety guarantees). These APIs have return codes that need to be inspected anyway. The reference documentation reads like a specification rather than what I would typically consider reference documentation. I'd expect to see it written with an intended audience of a casual user with examples and an organization that would be helpful for this reader (e.g.: purpose, description, examples, details) ## Tests The tests look reasonable and comprehensive although there seems to be some `#if 0` conditions that could either be cruft or perhaps the intent is the developer would manually change those to verify compilation failure. The tests are using the 'catch' library which is intended to be a submodule. Is there a precedence for using a third party library like this? For folks using Boost it could be painful to get another dependency approved to run all the Boost tests. It would be preferable IMO to use Boost.Test or something self contained. ## Usefulness I attempted to use the library with g++ 8.3.0 and clang 8.0.0. My example code is below: ```C++ #include <boost/out_ptr.hpp> #include <sqlite3.h> #include <iostream> int main() { std::shared_ptr<sqlite3> database; if (SQLITE_OK != sqlite3_open("test.db", boost::out_ptr::out_ptr(database, sqlite3_close))) { std::cout << "Problem opening the file"; } using Sqlite3Ptr = std::unique_ptr<sqlite3, decltype(&sqlite3_close)>; Sqlite3Ptr database2(nullptr, &sqlite3_close); if (SQLITE_OK != sqlite3_open("test.db", boost::out_ptr::out_ptr(database2, sqlite3_close))) { std::cout << "Problem opening the file"; } } ``` With this `Makefile` ```c++ main : main.cpp g++ -I/home/dsankel/code/out_ptr/include -o $@ $< `pkg-config sqlite3 --cflags --libs` .PHONY: clean clean: $(RM) main ``` I didn't run into any issues. Everything worked okay. I think this library would be handy for wrapping C libraries like sqlite. # How much effort did you put into your evaluation of the review? I spent a few hours on this review. I read through all the documentation, did a cursory review of the code, and came up with the example above. # Minor things * Copyright notices need to have a person's real name to have legal benefits. * In `aout_ptr.adoc` the examples link goes to the overview page. * Some grammar errors in the "C function calls are indeterminately sequenced..." paragraph. * When referring to a Boost library in free flowing text, the correct form is Boost.LibraryName with the same font as the surrounding text.

5 5

out_ptr review is extended until Wednesday, July 10
by Zach Laine 29 Jun '19

29 Jun '19

We don't have enough reviews yet, so we're extending the review by two weeks. Please review this small and useful library! >From the original announcement: The formal review of JeanHeyd Meneide's out_ptr library starts Sunday, June 16th and ends on Wednesday, June 26th. out_ptr is a library for making it easy to interoperate between smart pointers and traditional C-style initialization and allocation interfaces. It also enables doing so in a way that allows library authors to opt into speed optimizations for their smart pointers that give them performance equivalent to typical C pointers (see benchmarks ( https://github.com/ThePhD/out_ptr/blob/master/docs/out_ptr/benchmarks.adoc) and the Standard C++ proposal for more details). Online docs can be found here: https://github.com/ThePhD/out_ptr/blob/master/docs/out_ptr.adoc The GitHub repository is available here: https://github.com/ThePhD/out_ptr Also, there is an associated standards proposal: https://thephd.github.io/vendor/future_cxx/papers/d1132.html We encourage your participation in this review. At a minimum, please state: - Whether you believe the library should be accepted into Boost - Your name - Your knowledge of the problem domain You are strongly encouraged to also provide additional information: - What is your evaluation of the library's: * Design * Implementation * Documentation * Tests * Usefulness - Did you attempt to use the library? If so: * Which compiler(s)? * What was the experience? Any problems? - How much effort did you put into your evaluation of the review? All issues discussed during this review will be tracked on the library's GitHub issue tracker at https://github.com/ThePhD/out_ptr/issues. Zach

5 7

[accumulators] Offer to help with maintenance
by Cromwell Enage 29 Jun '19

29 Jun '19

Hi, everyone! I'm offering to help maintain Boost.Accumulators and have already e-mailed Eric Niebler regarding my intentions. With the deadline for Boost release 1.71 fast approaching, I can take steps to ensure that Boost.Accumulators is in a usable state. I also have a few ideas on how to update and improve Boost.Accumulators (one of them being to make use of variadic templates), but they can wait until after the release for further discussion. Very respectfully, Cromwell D. Enage

1 0

Re: [boost] [logic] Boost::logic (tribool?) actively developed?
by Jacob Lidman 28 Jun '19

28 Jun '19

> It has resently updated to support a change in gcc compilers. So I > would say it's actively maintained. I never asked if it was actively maintained but rather if the library is actively developed, i.e. are new features added, improvements in documentation... > I would say it seems stagnant because it's always been as close to perfection as a > library can be. The reason why the library appear close to perfect is because it clearly has minimal number of features. Improvements can be made to accommodate new needs, e.g. by including other logics. > see https://github.com/boostorg/logic As I mentioned this is the repo for the tribool library, at least that was the name prior to boost 1.70.0

2 1