- Boost - lists.preview.boost.org

Re: [boost] Asynchronous library now in Boost Library Incubator
by Christophe Henry 29 Nov '16

29 Nov '16

Hi Michael, >Thank you for submitting this library to the the incubator in >preparation for a review. I saw your presentation in 2014 >and I believe that this library solves a common problem. >My first question: I see that the library is listed as C++11 and also >has dependencies on Boost.System, Boost.Thread, Boost.Chrono, >and Boost.DateTime. Much of the functionality provided by those >Boost libraries is now part of the C++11 or C++14 standard. >Can you comment on the specific Boost dependencies >and if the standard versions would suffice? A good question which deserves a rationale, which I will add. In theory, the standard version would suffice, except for Thread. Boost.Thread provides interruption points, which the standard does not. With the standard version, I would not be able to offer task interruption. I rely for the other parts also on Boost because the standard version had bugs on earlier versions of compilers, and quite some projects are still stuck with gcc-4.9 or mingw. At some points, it seemed easier to maintain using Boost (which also happens to be faster than the standard, at least on my tests). HTH, Christophe

2 1

[ssl] SSL handshake takes forever sometimes
by Shuchen He 28 Nov '16

28 Nov '16

Hi, I have an application written in boost with SSL. I noticed that when the client network is down for a while and come back again. The SSL handshake sometimes take forever. I tried to kill the client application and run it again, the problem still there. The only way to make it work again was to reboot my client device. I'm using boost 1.60 on the client side and 1.58 on the server side. The SSL handshake code was copied from Boost SSL server and client example I modified a bit to suit my application needs. Does anyone experience the similar issue? If so, can you please let me know how do you fix it? It would be much appropriated if you could help me out on this issue. Below is the SSL handshake code. chat_client(boost::asio::io_service& io_service, boost::asio::ssl::context& context, tcp::resolver::iterator endpoint_iterator) : io_service_(io_service), socket_(io_service, context) { filesize_=0; isfileopen_=false; socket_.set_verify_mode(boost::asio::ssl::verify_peer); socket_.set_verify_callback( boost::bind(&chat_client::verify_certificate, this, _1, _2)); boost::asio::async_connect(socket_.lowest_layer(), endpoint_iterator, boost::bind(&chat_client::handle_connect, this, boost::asio::placeholders::error)); } bool verify_certificate(bool preverified, boost::asio::ssl::verify_context& ctx) { char subject_name[256]; X509* cert = X509_STORE_CTX_get_current_cert(ctx.native_handle()); X509_NAME_oneline(X509_get_subject_name(cert), subject_name, 256); log_time(); fprintf(logfile,"%s: Verifying %s\n",logtime.c_str(), subject_name); return preverified; } void handle_connect(const boost::system::error_code& error) { if (!error) { socket_.async_handshake(boost::asio::ssl::stream_base::client, boost::bind(&chat_client::handle_handshake, this, boost::asio::placeholders::error)); } else { log_time(); fprintf(logfile, " %s Connect to server failed: %s\n ",logtime.c_str(), error.message().c_str()); fflush(logfile); delete this; } } void handle_handshake(const boost::system::error_code& error) { if (!error) { ssl_handshake_ok=1; do_read_header(); } else { log_time(); fprintf(logfile,"%s: Handshake failed:%s\n ",logtime.c_str(), error.message().c_str()); } }

1 0

[poly_collection] request for help at performance testing
by Joaquin M López Muñoz 28 Nov '16

28 Nov '16

Hi, Some days ago I published the first version of (candidate) Boost.PolyCollection at https://github.com/joaquintides/poly_collection Performance tests are available at http://rawgit.com/joaquintides/poly_collection/website/doc/html/poly_collec… but GCC and Clang were run in a Travis VM and the resulting figures are noisy and not sufficiently accurate. I'm looking for volunteers willing to run the performance suite in one or more of the following environments: * Recent GCC, 32 bit * Recent GCC, 64 bit * Recent Clang 32 bit * Recent Clang, 64 bit The full suite in one environment takes 1.5-2h to execute. If you're interested, please contact me privately (so that we don't pollute the mailing list) and I'll provide you with further instructions on how to do the testing. Thank you! Joaquín M López Muñoz

1 0

[iterator] Orbiting iterators
by Jeremy Murphy 27 Nov '16

27 Nov '16

Hi, just wondering what the mood in Boost.Iterator is for new features? At work we recently found use for an orbiting iterator, specifically a circular one, which effectively wraps at the start/end of a bounded range. (The domain is digital signal processing.) Implementation details aside, does this feature sound desirable to the library? I call it an "orbiting" iterator because it is a specific (and useful) case of the more general rho-shaped orbiting iterator. The theory behind it is in chapter 2 of Elements of Programming, which happens to be the free sample chapter available: http://elementsofprogramming.com/032163537X_Sample.pdf#page=6 Anyway, if there is interest, I'll make a pull request. Cheers. Jeremy

6 10

Re: [boost] [beast] Separation between parse and generation and asio
by Vinnie Falco 24 Nov '16

24 Nov '16

On Wed, Nov 23, 2016 at 10:45 PM, Felipe Magno de Almeida <felipe.m.almeida(a)gmail.com> wrote: > I'm trying Beast with avr-gcc to see how much footprint it would take > on embedded applications. Thank you for your interest in Beast! > Unfortunately, stream_concepts uses asio > types, which requires asio include which requires network includes > which will obviously fail in most microcontrollers. > > Would it be possible to make parse and read implementation less > dependent on each other? I see what you mean. Yes, I believe I can do that - would you mind creating a new issue here? https://github.com/vinniefalco/Beast/issues Thanks!

1 0

[outcome] Requesting pre-review of Boost.Outcome tutorial
by Niall Douglas 24 Nov '16

24 Nov '16

Dear Boost, I am hoping to bring Boost.Outcome, a C++ 14 library providing a factory and family of policy driven lightweight monadic value-or-error transports, into the Boost peer review queue by March 2017 followed shortly thereafter by an ACCU talk in April if my talk proposal passes their programme committee. To that end, I've written an explanation and a sort of tutorial explaining the history and purpose of Outcome at https://ned14.github.io/boost.outcome/ and I'd greatly appreciate if people could tell me: 1. Does it make sense? 2. Do you think you could use Outcome in your own code after reading it? 3. Do you think you would want to use Outcome in your own code after reading it? 4. Are there any missing parts which are showstoppers? My thanks in advance. Be aware the install stuff described at the start doesn't work yet. The reference documentation is also quite patchy thanks to doxygen not understanding CRTP. All that stuff and a fair bit more remains to fix before it's ready to enter the Boost peer review queue. Niall -- ned Productions Limited Consulting http://www.nedproductions.biz/ http://ie.linkedin.com/in/nialldouglas/

6 14

[beast] Separation between parse and generation and asio
by Felipe Magno de Almeida 24 Nov '16

24 Nov '16

Hello, I'm trying Beast with avr-gcc to see how much footprint it would take on embedded applications. Unfortunately, stream_concepts uses asio types, which requires asio include which requires network includes which will obviously fail in most microcontrollers. Would it be possible to make parse and read implementation less dependent on each other? Boost.Spirit x3 works very well on microcontrollers. I'd expect a different type for messages which could actually receive iterator pairs for strings with Beast might give a very good option for HTTP/WS implementation on microcontrollers. Regards, -- Felipe Magno de Almeida

1 0

[sort] Parallel sorting sub-library mini-review conclusion
by Christophe Henry 22 Nov '16

22 Nov '16

> I would like to know the memory used and the best, >normal and worst case of the Asynchronous sorting algorithms. The sort versions are mergesort so they use double memory. The quicksort versions require no extra memory while partitioning. When they degrade after some partitioning iterations, they require the extra memory used by the underlying sort algorithm (spreadsort, std:sort, intro_sort, etc.). Thank you Francisco for this interesting library which I shamelessly used into my own ;-) It's probably a bit late but I vote for acceptance. Cheers, Christophe

2 1

Re: [boost] [sort] Parallel sorting sub-library mini-review. Performance tests.
by Christophe Henry 22 Nov '16

22 Nov '16

>> >> OTOH the single-threaded ones are interesting, especially the stable_sort >> and intro_sort for cases where usage of spreadsort is not possible. >> >I challenge you to find a sorting problem where it isn't possible to use >spreadsort. string_sort can sort anything that std::sort can; > you just need to convert the sort key comparison sequence into a sequence of bytes > and feed string_sort the correct functors. That said, it may not be >efficient to sort some problems with string_sort, >and it'll probably be too much work to be worth the effort for many problems. This is what I meant. Sorry for the lack of precision. I'll add it to my performance tests, it might be an interesting variant as the algorithm is really fast. >> >> A short summary: >> >> 100000000 uint64_t elements already sorted: >> Boost parallel sort : 0.064965 secs >> >> Asynchronous parallel sort : 0.0167134 secs (same with other >> algorithms) >> >> Asynchronous provides a special optimization for this case. >> >If this optimization is practical to incorporate in Boost parallel sort, >it is a common case and your solution seems noticeably faster, >so I would like Francisco to consider it. That said, .06s isn't bad. Sure. >> >> I added this one: >> 100000000 uint64_t elements reverse sorted >> >> Boost parallel sort : 0.581381 secs >> >> Asynchronous parallel sort : 0.0478701 secs >> >> Asynchronous provides a special optimization for this case. >> I this the library should do it too. This one is pretty common and a >> popular DOS attack. >> > This optimization would be nice to have, if it's cheap. A parallel reverse is in O(n) so it's cheap compared to a sort. >> >> 100000000 uint64_t elements randomly filled >> >> Asynchronous parallel quickspreadsort: 0.587432 secs >> Asynchronous quick_intro_sort : 0.728393 secs >> >> This mixing of quicksort degrading into a spreadsort works here best. >I find this algorithm interesting; Frankly, there is nothing special. It's simply a parallel partition, which degrades into a "normal" parallel merge sort (in this case spreadsort) after a while. >have you considered using MSD radix-based sorting to break up the data > for the threads in parallel? >For example, take T threads, have each thread take 1/T of the data > and bucket it into N MSD-based buckets (just like Spreadsort), >where T < N <= ~2^11, then merge the equivalent buckets, >and as you merge them hand them off to another thread to sort. >This will probably divide better if you find the max and min before bucketing >using a parallel min/max search. I admit I didn't. But it sounds really interesting and I think I'll have a go at it. I'm not a sort expert though. I approached this from the parallel perspective. >> 10000000 strings randomly filled >> >> OMP parallel sort : 1.05803 secs >> Boost parallel sort : 0.933055 secs >> >> OMP parallel stable sort : 1.12269 secs >> Boost sample sort : 0.889216 secs >> Boost parallel stable sort : 1.56564 secs >> >> Asynchronous parallel quickspreadsort: 0.788856 secs >> Asynchronous quick_intro_sort : 0.893652 secs >> >> Asynchronous parallel stable sort : 1.23495 secs >> Asynchronous boost::stable sort : 1.21817 secs >> >> Similar results. >> >> >> Let's move on to big objects: >> >> 1562500 elements of size 512 randomly filled >> H E A V Y C O M P A R I S O N >> >> OMP parallel sort : 0.308923 secs >> Boost parallel sort : 0.342992 secs >> >> Asynchronous parallel_quicksort : 0.246709 secs >> Asynchronous quick_intro_sort : 0.269666 secs >> >> Both quicksorts are best, with a slight advantage to intro_sort. >> >> The light comparison is similar. >> >Do we have your permission to copy and/or adapt your quick_intro_sort >algorithm into the Boost.Sort library? Please feel free to copy whatever might be useful. The library has a Boost license as it's intended for inclusion in Boost.I also took the freedom of adding the single-thread versions of Fernando's algorithms as basis for my parallel versions. The algorithm is quite simple. It's the same as quick_spreadsort, a quicksort which degrades into Francisco's intro_sort after some iterations. I just found it interesting to add a wrapper to Francisco's algorithm for the mini-review. It was just a quick test (no time for cleverer algorithms) but I'm pretty satisfied with the result and I'll keep these versions. >How is the worst-case performance relative to Francisco's library? The sort versions have O(n log n) like any merge sort. The quicksort have O(n log n) to O(n^2) as expected from a quicksort. Cheers, Christophe

1 0

Re: [boost] [sort] Parallel sorting sub-library mini-review. Performance tests.
by Christophe Henry 22 Nov '16

22 Nov '16

Hi, > The threads are created only 1 time*. Have I missed something? I checked sample_sort and I saw several sets of calls to std::async which is likely to create and destroy threads several times. >*parallel_sort, sample_sort* : this algorithm is commonly accepted as the >fastest for the parallel stable sort. >The implementation is *notoriously faster than the provided by the GCC compiler* >( based on Open MP ) using less memory. TBB don?t >provide parallel stable sort, and only have an experimental code in their web pages Indeed, I found it highly interesting. Cheers, Christophe

1 0