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

[review][fixed_string] review period finishes today Dec 8
by Joaquin M López Muñoz 08 Dec '19

08 Dec '19

The review period of candidate Boost.FixedString finishes today, Dec 8. Many interesting subjects have been discussed about during these two weeks and I have quite a bit of material to digest before publishing the final review result plus further action points. I expect to be able to do so during the following week. Thank you to all the people who have contributed their reviews and comments and to Krystian and Vinnie for submitting their work and enduring the famously severe review ordeal. Best regards, Joaquín M López Muñoz Review manager for candidate Boost.FixedString

2 1

Re: [boost] Boost GIL
by Siddhartha Sen 08 Dec '19

08 Dec '19

Thanks a lot for forwarding my email to the boost GIL list. I will check it out. What I meant by conversion, however was primarily converting an image to negative and greyscale. As you may already know, the two pf them are pretty similar and OpenMP has shown how incredibly fast the process can be using more threads. I just feel GIL would benefit a lot from using parallel programming. On Sun, 8 Dec, 2019, 5:30 PM , <boost-request(a)lists.boost.org> wrote: > Send Boost mailing list submissions to > boost(a)lists.boost.org > > To subscribe or unsubscribe via the World Wide Web, visit > https://lists.boost.org/mailman/listinfo.cgi/boost > or, via email, send a message with subject or body 'help' to > boost-request(a)lists.boost.org > > You can reach the person managing the list at > boost-owner(a)lists.boost.org > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of Boost digest..." > > > The boost archives may be found at: http://lists.boost.org/Archives/boost/ > > > Today's Topics: > > 1. Re: [fixed_string][review] Review of fixed_string (Rob Stewart) > 2. Re: Extending GIL (Mateusz Loskot) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Sat, 07 Dec 2019 13:23:51 -0500 > From: Rob Stewart <rstewart(a)ptd.net> > To: boost(a)lists.boost.org,Phil Endecott via Boost > <boost(a)lists.boost.org> > Cc: Phil Endecott <spam_from_boost_dev(a)chezphil.org> > Subject: Re: [boost] [fixed_string][review] Review of fixed_string > Message-ID: <142EED34-B802-4381-9FF9-C9FE999BA0F9(a)ptd.net> > Content-Type: text/plain; charset=utf-8 > > > > On December 5, 2019 9:35:35 AM EST, Phil Endecott via Boost < > boost(a)lists.boost.org> wrote: > > Peter Dimov wrote: > > > - a function that checks whether the remaining capacity is at least > > n should > > > be added. I suggest the name > > > > > > bool can_fit( std::size_t n ) const noexcept; > > > > > > although my previous choice of "has_capacity_for" has merit too. > > I suggest the simpler "fits". > > > Make this a free function template and you can use it with other > > containers: > > > > template <typename CONTAINER> > > bool can_fit( const CONTAINER& C, typename CONTAINER::size_type n ) > > noexcept > > { > > return C.capacity() - C.size() >= n; > > } > > That's a nice idea, especially with the name "fits". :) However, if you > reverse the parameters, "fits_in" would be a better name as it indicates > the argument order. > -- > Rob > > (Sent from my portable computation device.) > > > ------------------------------ > > Message: 2 > Date: Sun, 8 Dec 2019 01:25:04 +0100 > From: Mateusz Loskot <mateusz(a)loskot.net> > To: boost(a)lists.boost.org > Subject: Re: [boost] Extending GIL > Message-ID: > < > CABUeae_cupOMOhCNvvHaeruX2ksycHhVKkHJbtCSCLr-kW-XkA(a)mail.gmail.com> > Content-Type: text/plain; charset="UTF-8" > > On Sat, 7 Dec 2019 at 01:12, Siddhartha Sen via Boost > <boost(a)lists.boost.org> wrote: > > > > Having looked up this repository : > https://github.com/boost-gil/gil_threaded, > > I feel the idea was the same, but a lot of improvement can be made. > > Good, sounds like you may have a plan in mind > > > Also, there was no provision for image conversion, which I feel > > Please, you need to be more specific here. > What kind of conversion do you mean: color, channel/pixel type, format,...? > > p.s. I forwarded your post to the GIL mailing list > https://lists.boost.org/boost-gil/2019/12/0362.php > > Best regards, > -- > Mateusz Loskot, http://mateusz.loskot.net > > > ------------------------------ > > Subject: Digest Footer > > _______________________________________________ > Unsubscribe & other changes: > http://lists.boost.org/mailman/listinfo.cgi/boost > > > ------------------------------ > > End of Boost Digest, Vol 6271, Issue 1 > ************************************** >

2 1

[release] 1.72.0 post-beta merges
by Michael Caisse 08 Dec '19

08 Dec '19

The master branch is is now open for post-beta merges, but only as described in the Post-Beta Merge Policy. See <https://github.com/boostorg/boost/wiki/Releases%3A-Beta-Merge-Policy>

11 31

[range] Trying to extend boost::range::combine to std::tuples
by Andres Llopis 08 Dec '19

08 Dec '19

Hi everyone, I am trying to make my own version of combine so that it works also with std::tuples. Someone on slack suggested that the zip iterator should be able to handle it properly. My attempt however fails when trying to use the metafunction tuple_of_references::type for the std::tuple of iterators. This is what I tried (https://godbolt.org/z/Ymjtvn) (Please excuse any style problems, this is my first post) namespace myns { using boost::iterators::zip_iterator; using boost::iterator_range; template<typename IterTuple> class combined_range : public iterator_range<zip_iterator<IterTuple> > { typedef iterator_range<zip_iterator<IterTuple> > base; public: combined_range(IterTuple first, IterTuple last) : base(first, last) { } }; template<typename... Ranges> auto combine(Ranges&&... rngs) -> combined_range<decltype(std::make_tuple(std::begin(rngs)...))> { return combined_range<decltype(std::make_tuple(std::begin(rngs)...))>( std::make_tuple(std::begin(rngs)...), std::make_tuple(std::end(rngs)...)); } }

3 5

Extending GIL
by Siddhartha Sen 08 Dec '19

08 Dec '19

Having looked up this repository : https://github.com/boost-gil/gil_threaded, I feel the idea was the same, but a lot of improvement can be made. For example, by using OpenMP, I feel the computational performance can be massively improved and that too, not requiring any extremely sophisticated hardware. Also, there was no provision for image conversion, which I feel is a very good extension. What do you all think?

2 1

[fixed_string][review] Review of fixed_string
by Peter Dimov 07 Dec '19

07 Dec '19

This is my review of FixedString. * Summary The library should be ACCEPTED. It's useful, for more than one domain, and specifically where compile-time string manipulation is concerned, everyone needs to reinvent it. The name fixed_string is fine as is. Alternatives exist, but I don't see any of them as unquestionably better, and one becomes accustomed to "fixed_string" after a very brief period. Throwing on exceeding capacity is fine as is. A function should be provided to check the remaining capacity, because writing this check by hand is very easy to get wrong, and in fact the library throughout its implementation consistently gets it wrong. * Documentation The documentation is adequate, but I don't find the reference convenient. I would prefer a synopsis that contains the entire class definition of `fixed_string`, so that I can see the various overloads at a glance. Instead, the synopsis only gives me the declaration of `fixed_string`, without even telling me what namespace it's in. (It also gives me `basic_string_view`, also without mentioning the namespace.) The current layout is fine if I want to look up a specific function by name. The format should ideally follow the standard library, with Throws and Expects clauses (the latter instead of "undefined behavior unless".) It's not documented when constructors and assignment operators throw. * Design - static_capacity should be removed; capacity() and max_size() should be static constexpr. - all nonthrowing functions should be marked `noexcept`. - initializer_list<CharT> overloads are of dubious utility. I know that the design copies std::string here, but I think that we should deviate and not provide them. - substr() should return a fixed_string, subview() a string_view (as already implemented.) - operator+ should be provided, only for arguments with capacity known at compile time (fixed_string<N>, CharT and CharT const (&)[ N ]), with the capacity of the result being the sum of the capacities of the arguments. This is not very useful for long strings, but it's fine for short strings, useful for compile-time manipulation, and useful in simple cases such as to_fixed_string( n ) + " bytes written". - a function that checks whether the remaining capacity is at least n should be added. I suggest the name bool can_fit( std::size_t n ) const noexcept; although my previous choice of "has_capacity_for" has merit too. This is, in my opinion, an absolutely necessary addition, because of my already-stated observation that the check is almost always gotten wrong. Here for instance is an example: if(size() + count > max_size()) BOOST_FIXED_STRING_THROW(std::length_error{ "size() + count > max_size()"}); The correct way to write it is max_size() - size() > count, which avoids the integer overflow in size() + count. In this case, passing count = (size_t)-1 will pass the check (when size() > 0) and then lead to undefined behavior. This makes the utility of reserve() dubious, because it basically invites one to get it wrong, but I think we should retain it for consistency. - everything should ideally be constexpr. The problem here is that constexpr requires initializing all elements, and this will heavily penalize runtime uses of f.ex. fixed_string<512>. On compilers with __builtin_is_constant_evaluated (gcc 9, clang 9) we should use that; otherwise, it might be worth it to create a specialization for N < some suitable upper limit. Again, any nontrivial string manipulation at compile time basically requires a fixed_string. * Implementation The physical separation of the library into three headers is inconvenient. A single header would make it possible to include the library directly on Godbolt via https, which is useful for both demonstrations and analysis of the generated code. In addition, hunting down the definitions is irritating for people reading the code. The reverse is true for the tests; they should be split into files each testing a logical group, such as constructors, assignment, append, and so on. The helper functions "testAS", "testI" and so on make reading the tests unnecessarily cryptic (and is quite annoying for reviewers.) They should have been named "test_assign", "test_insert". It's generally better to use `BOOST_TEST_EQ(s.size(), 0);` instead of `BOOST_TEST(s.size() == 0);` because the former prints the values on failure, which often gives clues as to where the error is. The test functions shouldn't be in the boost::fixed_string namespace. Users don't write their code in this namespace, so the tests don't test what users use. All capacity checks should be either audited for correctness and fixed, or use the proposed can_fit addition. Instead of repeating BOOST_FIXED_STRING_THROW(std::length_error{ "n > max_size()"}); in every function, which leads to it being inlined there repeatedly, this could use a helper function void throw_length_error( char const * msg, boost::source_location const& loc ) { boost::throw_exception( std::length_error( msg ), loc ); } taking advantage of the new boost::throw_exception overload that takes a source location as documented in https://www.boost.org/doc/libs/develop/libs/throw_exception/doc/html/throw_…. A variation could take `n` as an argument and compose a message containing it, f.ex. void throw_length_error_fmt( boost::source_location const& loc, char const * fmt, ... ); #define BOOST_FIXED_STRING_USE_BOOST should go, as already mentioned. Since the library requires C++11, there's no need to use BOOST_STATIC_ASSERT. static_assert is fine. Unless of course we backport it to C++03, which might be doable, although of questionable utility. Checking for overlap and using Traits::copy instead of Traits::move isn't needed. Traits::move does the same check anyway, so there's no gain, just an opportunity to get the check wrong and introduce undefined behavior. "The behavior is undefined if `count >= npos`" doesn't seem necessary. In this case, count will be bigger than max_size() and the constructor will throw std::legth_error. There's no reason to gratuitously introduce undefined behavior. All "undefined behavior" cases should consistently BOOST_ASSERT.

4 4

[nowide] Future
by Artyom Beilis 07 Dec '19

07 Dec '19

The nowide library was reviewed and accepted into Boost almost 3 years ago. I implemented most of fixes required under Review Fixed branch https://github.com/artyom-beilis/nowide/tree/reviewfixes Unfortunately I don't have any capacity to do maintenance for it on all various platforms: MSVC, mingw. cygwin etc mostly because virtually I don't develop on Windows and any time I need to go there I need to figure it out from the beginning (installations, environment, msvc) and I don't even have windows machine in my hands (all done in VM) I realized that I'm not windows developer any more and it would be impossible for me to complete the mission of maintaining the code mostly due to personal reasons and lack of development resources. I deeply regret wasting the time and effort of so many boost developers who invested to review. I understand that it is highly desired library but I just can get the job done I seek for your advice. Do I need to seek for an another maintainer or withdraw the library entirely? Once again sorry for disappointing the community and people who were invested in this library. Artyom Beilis

6 10

[review][fixed_string] End of review period approaching and a plea
by Joaquin M López Muñoz 06 Dec '19

06 Dec '19

The review period for candidate Boost.FixedString finishes tomorrow, Dec 4. There's an ongoing lively discussion on many aspects of the library, which is very good, but only three of you have cast your acceptance/rejection vote, which is not good. This is a gentle petition to the people engaging in the review who haven't voted yet: Vinícius, Nevin, Mike, Andrzej, JeanHeyd, Hans, Peter, Julien, Andrey, Tim, Gavin, Emil, (and anyone who might be still lurking) to please submit a review with an explicit vote. If you feel like you need some more time to do the work, I think we are in a position to extend the review period, but please express your intentions before I ask the review wizards. Thank you! Joaquín M López Muñoz Review manager for candidate Boost.FixedString

13 35

[release] Boost 1.72.0 Release Candidate 1
by Marshall Clow 05 Dec '19

05 Dec '19

The first release candidates for the 1.72.0 release are now available at: <https://dl.bintray.com/boostorg/release/1.72.0/source/> The SHA256 checksums are as follows: fcd8b598bc63ae7bbd3e72635d95b9d749ef65f4e5342c24445ea1fb9d417ac3 boost_1_72_0_b1_rc1.7z f6a527a261e273bfd2b6161d359ab9fb53c1795239c696f501a62f575cd30bda boost_1_72_0_b1_rc1.tar.bz2 8e591146d39e8cb45c74540aae26ae6321b1e83ab1b12e236dd117f93475ceb9 boost_1_72_0_b1_rc1.tar.gz 3425e7bfd49a6ad70aa7e82c1c740f7821eb58d14604b133be4dfbcdc0c58e3a boost_1_72_0_b1_rc1.zip As always, the release managers would appreciate it if you download the candidate of your choice and give building it a try. Please report both success and failure, and anything else that is noteworthy. -- The Boost Release Managers

3 10

[review][fixed_string] Andrzej's feedback
by Andrzej Krzemienski 05 Dec '19

05 Dec '19

Hi Everyone, This can hardly be called a review, as I didn't look thoroughly at the implementation. (I only checked what happens on the resize() over the capacity(), as this was not clear from the docs.) I think that ultimately the library would be a useful addition to Boost, however I would not like to rush it before its design is fully baked, and my impression is that it isn't yet. Therefore my recommendation would be to REJECT the library at this point, and encourage the authors to resubmit it for the second review once the design goals and design decisions have been clearly settled and communicated. My vision of a review is that we: 1. First evaluate the design goals, whether they are worth pursuing 2. Then we evaluate the design *against* those goals. 3. Then we evaluate the documentation and implementation *against* the design and goals. I cannot even go to #3, as to me #1 is missing some parts. There are different possible goals: 1. Treat this new string as many other STL sequence containers with variable size. In particular, users use it as if its max_size() were infinite, and are prepared that at some point an exception might be thrown. 2. Treat it as a "drop-in replacement" for std::string with no-heap-alloc guarantee. 3. Make something suitable for embedded systems where the user provides the guarantee: "I will never make the size() of this string go above this value. If I did, it would be plain clear that the program is not doing what it was supposed to." These goal 3 is in clear conflict with either 1 or 2. This becomes clear when we consider what should be the response to doing a resize() over capacity(): a precondition violation or a normal course of action signaled by an exception. The authors should make a choice which of these goals is not supported and communicate it clearly from the outset in the docs. I think this decision has already been made: the library was extracted from Boost.Beast and there the purpose must have been clear. It is just that it is not communicated in fixed_string library. Any decision is good. In C++ performance trade-offs are part of the contract (interface), so there is room in Boost for many strings: small strings, short codes, fixed size strings, fixed capacity strings. But for each library it should be clear what trade-offs were chosen and why. We have a situation that there exists static_vactor in Boost, which defines the resize() over capacity() to be a precondition violation. And fixed_string needs to provide some response to this fact: why did you choose a different design? I consider the choice of name to be part of the design: you cannot assign the name before you have decided on the goals. "static_string" might be a good name if you chose the same trade-offs as static_vector. But if you make different trade-offs, it might be better to use a different name. Making a resize() over capacity() a precondition violation is a *feature* useful for bug detection and I do not consider it a valid argument that "library will throw exceptions and if you never resize() over capacity() you will never see exceptions or std::abort()". If this is a precondition, then I expect of a library to put some BOOST_ASSERT() or _builtin_unreachable() in those paths to enable better bug detection. I find the goal "a drop-in replacement for std::string" invalid. You surely mean only a drop in replacement for some subset of usages. I guess this subset comes from the usage in Beast, and I would expect that it is clearly outlined what this subset is. One of the operations on strings that I use most often is operator+, so when I hear that this operation is not implemented in a class that claims to be a drop-in replacement for string, I see a contradiction. There might be valid reasons not to provide operator+, but the documentation cannot state that this is a drop-in replacement for std::string. I think that only after the above issues have been sorted out can we have a review of the implementation and the documentation. One remark about documentation is that I was not able to find what `resize()` over `capacity()` does. This is the most important information that I was after. anything else is more less intuitively clear from the analogy to std::string. I hope that this feedback doesn't sound discouraging. I appreciate the work Krystian and Vinnie have put into making this submission, and I would like to see this library ultimately included in Boost. Regards, &rzej;

7 27