April 2014 - Boost - lists.preview.boost.org

[context] msvc /SAFESEH
by Jeff Flinn 12 Aug '15

12 Aug '15

I need to build my application with /SAFESEH which uses boost coroutine which uses context. I'm getting the following link errors: 1>libboost_context-vc100-mt-gd-1_53.lib(jump_i386_ms_pe_masm.obj) : error LNK2026: module unsafe for SAFESEH image. 1>libboost_context-vc100-mt-gd-1_53.lib(make_i386_ms_pe_masm.obj) : error LNK2026: module unsafe for SAFESEH image. I've built boost static libs with the same MSVC vesion and options (sans /SAFESEH as it's not a compile/lib option) Compiler/linker versions: Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 16.00.40219.01 for 80x86 Microsoft (R) Incremental Linker Version 10.00.40219.01 Has anyone run into this and resolved the error? Any insight is appreciated. Thanks, Jeff

3 9

Interactive metaprogramming shells
by Abel Sinkovics 28 Dec '14

28 Dec '14

Hi, I've built interactive shells for preprocessor and template metaprogramming. I've been using them for development and debugging of metaprograms and I think Boost developers could also find them useful. You can type in (metaprogramming) expressions which get evaluated and the result is displayed immediately. For example you can type boost::mpl::push_front<boost::mpl::vector<int, char>, double>::type and you get boost_::mpl::vector<double, int, char> as the result (the boost_ namespace is used because there are some tricks there to make the output more readable). As an example for the preprocessor shell you can type BOOST_PP_REPEAT(3, foo BOOST_PP_TUPLE_EAT(3), ~) and you get foo foo foo as the result. The tools are based on the Boost.Wave library and Clang. You can get them here: Preshell for preprocessor: https://github.com/sabel83/preshell Metashell for template metaprograms: https://github.com/sabel83/metashell You can also try them from your browser here: http://abel.web.elte.hu/shell Feedback is welcome. Regards, Ábel

6 16

[coroutine] x86 msvc11 stack corruption
by Tobias Loew 12 Dec '14

12 Dec '14

Hi, I use Boost 1.55 uni-directional coroutines to build an x86 MSVC11 application. I have a simple pull-type (boost::coroutines::coroutine<void>::pull_type) that under certain reproducable conditions destroys the stack after returning from the coroutine. Here is the corrupted stack as shown by VS2012 > ebs2000.exe!CScriptMainFrame::DoRun(CPascalEditorDoc * 0x00000000, int 0x00000000, int 0x00000000) Line 3372 C++ ebs2000.exe!CScriptMainFrame::OnDebugStepover() Line 4916 C++ mfc110u.dll!_AfxDispatchCmdMsg(CCmdTarget *, unsigned int, int, void (void) *, void *, unsigned int, AFX_CMDHANDLERINFO *) C++ 00000002() Unknown [Frames below may be incorrect and/or missing] mfc110u.dll!CCmdTarget::OnCmdMsg(unsigned int, int 0x00000000, void *, AFX_CMDHANDLERINFO *) Line 364 C++ mfc110u.dll!CWnd::OnCommand(unsigned int 0x00018263, long 0x00000000) Line 2785 C++ mfc110u.dll!CFrameWnd::OnCommand(unsigned int 0x00018263, long 0x00000000) Line 389 C++ ... This only happens in windows x86 32-bit release-builds when the coroutine finishes upon a user-action that also includes a windows WM_SETFOCUS message on my menu-bar. If at least one of the following is true, the stack-corruption does NOT happen: - debug-build - x64 build - using a different pull-type, e.g. enum status_enum{...}; boost::coroutines::coroutine<status_enum>::pull_type - not returning form the coroutine, but pushing just before the end of the coroutine and swapping the pull-type-variable with a default constructed "not-a-coroutine" Even though I have a lot of workarounds, I still would like to know, why this happens. regards Tobias PS: I had to recompile ml /safeseh /c make_i386_ms_pe_masm.asm ml /safeseh /c jump_i386_ms_pe_masm.asm to get SAFESEH-libraries. -- View this message in context: http://boost.2283326.n4.nabble.com/coroutine-x86-msvc11-stack-corruption-tp… Sent from the Boost - Dev mailing list archive at Nabble.com.

2 3

[winrt support] Adding support for Windows 8 store/phone to Boost libraries
by stgates 22 Jul '14

22 Jul '14

Hi, I work for Microsoft and we’ve been taking a look at making a some of the libraries in Boost work in the Windows 8 store and phone 8 runtime environment (WinRT). The main issue is use of Win32 desktop APIs that are banned in WinRT. The fixes generally involve updating to a newer non-banned version, like CreateEvent to CreateEventEx, or replace areas that call banned APIs using ifdef macro techniques. Similar to what already exists in some places of Boost when targeting different platforms or operating system versions. Files which do not use any banned APIs are largely unaffected. So far we’ve made the thread, chrono, system, and date_time libraries work in WinRT. To facilitate building and testing, changes are also required to the build and test libraries. Listed at the end of this message is a summary of the major changes made. I’m unfamiliar with the process for contributing back to Boost and would like to know what is the best way to work, to contributing back these changes to each of the individual libraries? I see the documentation for working with the git submodules (https://svn.boost.org/trac/boost/wiki/StartModPatchAndPullReq) Right now the changes for each library are in a forked repository on a new branch called ‘winrt’ from the ‘develop’ branch. For example the changes we've made for the chrono library can be located here: https://github.com/MSOpenTech/chrono/tree/winrt. Is the best way to proceed, or should I to just initiate a pull request to the develop branch on each of the individual libraries? Thanks, Steve Boost Build To support deciding which Windows API to target I added a new feature <windows-api> with values desktop, store, and phone. When utilizing the Visual Studio 2012 on Windows 8 or later the store and phone options are available, otherwise desktop is the default. Most of the changes in msvc.jam are to deal with selecting the correct vcvars setup script since with phone it is in a different location. I also added ARM support to the msvc toolset. For libraries which support the new Windows Runtime they now produce both the desktop and store binaries by default. This is done by calling the supported-windows-api rule I added in toolset.msvc which is called in the default-build section of each library’s Jamfile to indicate if targeting WinRT is supported. When building for the Windows 8 store or phone a new macro is now defined BOOST_WINAPI_FAMILY, set with the same value as WINAPI_FAMILY. This is then used in the individual Boost libraries to detect if targeting the new Windows Runtime and avoid certain banned APIs. To handle running tests against WinRT in an automated fashion that integrates well into the Boost workflow of running tests as part of the build, if windows-api is set to store, the test and all its binary dependencies need to have the APPCONTAINER requirement stripped. This then allows execution outside of the sandboxed Windows store environment. To do this all dependent dlls are copied into the tests executable’s folder, if using <link>shared. This is necessary to avoid modifying the APPCONTAINER flag on the actual product binaries. Just let me know if this needs more explanation, I’ll be glad to go into more detail if necessary. With this change the tests for Windows store run just like the existing Windows tests. I also added a new feature to control whether or not test binaries are actually executed, disable-test-execution. This is used to avoid running the ARM tests when building since Visual Studio 2012 doesn’t install on ARM devices. All these changes are in testing.msvc. Chrono The GetProcessTimes and GetThreadTimes functions doesn’t exist for WinRT and there are no replacements so the thread_clock and process user, system, and cpu clock features are disabled. Only process_real_cpu_clock is supported. Thread For the thread library most of the changes are around the fact that there is no way to directly create a thread in WinRT. To work around this I implemented the boost::thread class itself on top of std::thread. This actually leads to pretty clean code, much cleaner than directly targeting the WRL thread pool like in the C runtime. Other changes include moving from TLS to FLS and basic banned API replacements like WaitForSingleObjectEx for WaitForSingleObject. In general where it makes sense updates were made to utilize new APIs on platforms when available. Test For the Boost test library I added a new macro for turning on/off debugging support (BOOST_TEST_HAS_DEBUG_SUPPORT). WinRT doesn’t support creating processes so attaching a debugger to execution like this is disabled. There also are no APIs for accessing registry keys, so this feature has been disabled as well. -- View this message in context: http://boost.2283326.n4.nabble.com/winrt-support-Adding-support-for-Windows… Sent from the Boost - Dev mailing list archive at Nabble.com.

8 22

Proposal: Vector-like container, which takes O(log(N)) for insert/erase, feedback
by Alexander Kuprijanov 12 Jul '14

12 Jul '14

Hi, everybody! The most important feature of my container is performance. However, other features will appear later. https://github.com/alexkupri/array/blob/master/description/description.txt Today I compared my container with rope. My container outperformed it 10 times. > Date: Thu, 3 Apr 2014 16:39:29 -0700 > From: Glen Fernandes <glen.fernandes(a)gmail.com> > I wasn't sure at first why my name was in parenthesis in the e-mail > subject. :-) > Glen It's my fault, sorry about it. > Date: Fri, 04 Apr 2014 13:41:04 +0200 > From: Adam Wulkiewicz <adam.wulkiewicz(a)gmail.com> > Subject: Re: [boost] Proposal: Vector-like container, which takes > O(log(N)) for insert/erase > AFAIU just like a B-tree can be seen as a generalization of a binary > search tree in that a node can have more than two children, your > container can be seen as a similar generalization of a Rope. Or it can > be seen as an augumented B-tree (not really because the values are no > longer stored in internal nodes and the tree isn't sorted, correct me > if I'm wrong). Your You understood right. > proposal is very similar to Counted B-tree > (http://www.chiark.greenend.org.uk/~sgtatham/algorithms/cbtree.html) > which is a B-tree with random access. The difference is that your > container can't be used to search for a stored value in O(logN) > without sorting and binary search. This may be ok if someone require > only random access, the size of a node is smaller which probably > improves the cacheing. Again, right. There are a lot of situations when only random access is necessary. You probably will not sort characters in string. > > In general, B-tree was introduced to improve the performance of the > access to data stored in persistent memory (lower performance). In > this case it was better to load the whole page into memory (and cache > it) than load many small, scatered parts. But in the case of > RAM-only-access, a binary search tree should be faster. As far as we > don't consider spatial indexing, interval trees, etc. where > values/nodes may "overlap" somehow and we must search in more than one > subtree. Therefore I'm wondering if a classic Rope would be faster > than your container in the use-case you're testing - simple insert and > random It outperformed Rope 10 times. It's expected result. Please visit my link, I added verbose explanation under "ROPE TEST" section. Shortly speaking, four reasons exist: cache, pages, balancing and malloc/free. > access to a in-memory container. Have you test it? I'd perform 2 tests: > 1. Very big number of values + truly random insert/access to prevent > cacheing > 2. insert/access in the same area I've tested with truly random tests. Please see results and code of SingleOperationPerformanceCheck and MultipleOperationsTest functions. Tell me, if I'm wrong. > > IMHO the benefits related to the use of this container should be more > noticable if the persistent storage were used. In memory/cache there are the same relationship as in HD/memory. > > As for the name, it's definietly not an array. I'd call it B-rope. Thanks, your option is accepted. I'll collect more options and we'll vote. I like brope. > > Regards, > Adam > Message: 9 From: Lars Viklund <zao(a)acc.umu.se> > Date: Fri, 4 Apr 2014 14:10:06 +0200 > If it doesn't have contiguous storage, it's not a replacement for > vector, however much you want it to be the universal container that > everyone uses for everything. > It may be a better general-purpose container compared to the > usual default containers, deque and vector, but it isn't a replacement > for either of them. Well, I would see it as first choise container for non-sorted data. It's comparable with vector/deque on small arrays and outperforms them on large sizes. It's more preferrable than list at least in terms of malloc/free calls. And it's probably better for strings. So one should use it by default in all cases with unsorted data and use list, vector, deque and string only if he/she really understands what he/she does. > You use the term 'intrusive' in the documentation, but it doesn't seem > to carry the usual meaning that it requires the types involved to > participate in the container structure. You are right. I've corrected the documentation. Now I call "fat leaves" leaves which contain elements, and "thin leaves" leaves which contain pointers to elements. > Your tests doesn't seem to exercise much else than container<int> > either. Suggest your tests. > I'm not quite sure I like the explicit L/M non-type template parameters > either, considering that it leaks implementation detail into every > single user of the type. Well, you are right in principle. It must have default parameters. But it's platform dependent, they should be chosen so that branch and leaf would take roughly 1 cache line, minus information required for malloc overhead. > As for the Copy-on-Write semantics, isn't the general consensus that CoW > is a horrible horrible thing and is very hard to get right+fast in the > modern world with concurrency? I disagree with your estimation. There are different situations, different machines. Embedded and mobile machines want to save memory. Large machines can also want to save memory, if they amount of data is large. Finally, they can even speed-up by avoiding copying the whole array. Imagine you are writing text editor. (DNA editor, sound editor, etc.) You make backup copy and continue editing, again backup copy and continue editing. And you keep only forward commands (without undo). You can easily rollback to any state you need. Ofcourse, user should decide, whether CoW option is necessary for him/her, because it wastes resources when enabled and not used. > > Message: 11 > Date: Fri, 4 Apr 2014 23:23:09 +1000 > From: Vadim Stadnik <vadimstdk(a)gmail.com> > Could you please provide a figure that shows data association with > nodes in > your data structure? I updated the doc, follow the link above.

7 8

[random] new threefry random engine
by Thijs van den Berg 13 Jun '14

13 Jun '14

Hi all, I've completed a nice new random engine that I would like to propose for addition to boost/random. The engine is called "threefry" and it's based on the paper "Parallel random numbers: as easy as 1, 2, 3" by Salmon, John K. and Moraes, Mark A. and Dror, Ron O. and Shaw, David E. I've been using it myself a lot for the last couple of years, and now I've rewritten it so that it will fit in the boost random framework and adhere to the concepts. It has some really nice properties, in particular for large scale parallel computing: * High quality randomness (according to the BigCrush tests) with a cycle length of 2^258 * Fast (comparable to Mersenne Twister) & little memory usage (13x 64 bit integers) * O(1) discard and guarantees for non-overlapping sequences when the (up to 256 bit) seeds are unequal I've uploaded the code, description and tests to https://github.com/sitmo/threefry The actual engine is at https://github.com/sitmo/threefry/blob/master/random/include/boost/random/t… I would really appreciate any feedback about the idea for inclusion and of course comments about the quality of the code,tests,docs kind regards, Thijs Sitmo Consultancy B.V. Financial Modelling & Data Science + 31 6 24110061 thijs(a)sitmo.com P.O. Box 1059, 2600BB, Delft, The Netherlands

10 60

[test] Looking for co-developer/maintainer
by Gennadiy Rozental 29 May '14

29 May '14

Hi, I think this is somewhat overdue. It's obvious various forces pulling me in different direction won't allow me to continue to do this all by my own. I am still plan to contribute and maintain the library, but I do believe it can be served better if someone can volunteer to share the load with me. There are both maintenance and development tasks in pipe. I have in mind quite a few very interesting features we can extend Boost.Test with (aside from those already requested by users and waiting to be implemented). These include: MT testing support performance testing support Mocking library GUI/DE integration and more Short term though I need help pushing to the release features which are already implemented in trunk and waiting to be released. Major showstopper is absence of proper documentation for these. So one of the first things we'll need to do is to write one. We also need to improve unit testing and finally push this release though some form of boost review. I do not have any specific requirements for the applicant, aside from interest in the library and willingness to work with me. I do not mind several people helping out with documentation, but longer term I'd prefer not very large group of maintainers for the library. If you are interested please shoot me email off list to rogeeff at gmail. Regards, Gennadiy

14 49

Is there any interest in a library for actor programming?
by Charousset, Dominik 28 May '14

28 May '14

Hi, is there any interest in a library for actor programming? The library we've developed is a C++11 open source library named "libcppa". It is currently released under LGPL2, but re-releasing the library under the boost license is not an issue. You'll find all important links and ressources either under http://libcppa.org or on the GitHub project page: https://github.com/Neverlord/libcppa, including a user manual as HTML or PDF. We've also submitted a paper to the C++Now conference. What is actor programming? For those of you who aren't yet familiar with the actor model: actors basically represent tasks or services that communicate via message passing. In libcppa, actors are very lightweight (you can spawn literally millions of them) and scheduled cooperatively using a thread pool (although actors are allowed to opt-out of the cooperative scheduling). Some of the main strengths of this programming paradigm are: (1) network transparency, (2) race conditions are avoided by design, (3) a strong failure model that enables developers to build reliable distributed systems, and (4) the actor model addresses both concurrency (multiple actors on one host) and distribution (any number of actors on any number of hosts). Here's the classical hello world example: ---------------------------------------------------------------------------------- #include <string> #include <iostream> #include "cppa/cppa.hpp" using namespace std; using namespace cppa; void mirror() { // wait for messages become ( // invoke this lambda expression if we receive a string on_arg_match >> [](const string& what) { // prints "Hello World!" via aout (thread-safe cout wrapper) aout << what << endl; // replies "!dlroW olleH" reply(string(what.rbegin(), what.rend())); // terminates this actor ('become' otherwise loops forever) self->quit(); } ); } void hello_world(const actor_ptr& buddy) { // send "Hello World!" to our buddy ... sync_send(buddy, "Hello World!").then( // ... and wait for a response on_arg_match >> [](const string& what) { // prints "!dlroW olleH" aout << what << endl; } ); } int main() { // create a new actor that calls 'mirror()' auto mirror_actor = spawn(mirror); // create another actor that calls 'hello_world(mirror_actor)' spawn(hello_world, mirror_actor); // wait until all other actors we have spawned are done await_all_others_done(); // run cleanup code before exiting main shutdown(); } ---------------------------------------------------------------------------------- A message is simply a tuple of values that is pattern matched at the receiver. For example, " on_arg_match >> [](const string& what) {...}" defines a message handler that is called whenever a tuple with one single string element arrives. "on_arg_match" simply defines a pattern that matches the signature of the given lambda. Please have a look at the examples folder on GitHub (https://github.com/Neverlord/libcppa/tree/master/examples) or at the user manual (http://neverlord.github.io/libcppa/manual/) if you want to see more examples. Best regards, Dominik Charousset

17 62

[Filesystem] Proposal: make filesystem generic-programming friendly
by Alexander Lamaison 23 May '14

23 May '14

Dear all, I'm increasingly finding that I need to write programs that can operate over multiple kinds of filesystem, including the local filesystem, network filesystems such as (S)FTP, filesystems in archives such as zip files, and mock filesystems that exist only in memory for unit testing. This is not as easy as it could be. The difficulty is that the Boost.Filesystem API makes it difficult to write generic code that uses the local filesystem as just one of several API-compatible filesystems. Therefore, I've written a proposal for making a backward-compatible addition to the Boost.Filesystem API that would make it much more friendly to generic programming. It is a first draft and I welcome your comments, criticism and suggestions for improvement. The proposal is available here: http://alamaison.github.io/2014/01/09/proposal-generic-filesystem/ and I've included a text-only version below. Alex Proposal: Generic filesystem API ================================ [FIRST DRAFT] C++ streams provide a common interface to operate on file-like data, regardless of how that data it is actually stored. However, programs often need to operate not just on data from different file sources, but also on entirely separate filesystems. The problem ----------- There is, currently, no way to operate on different filesystems generically, if Boost.Filesystem is to be one of those filesystems. The local-filesystem operations in the Boost.Filesystem API make it difficult to call them generically because they, inadvertently, defeat ADL, so generic code cannot resolve them to specific implementations. For example, how do you write code that calls `path temp_directory_path();` and can operate on both the local filesystem, via Boost.Filesystem, and on a filesystem over FTP, via an FTP library? If both libraries declare the function, how do you resolve the correct implementation? You can't use a namespace as a type. Normally, ADL is the solution to the problem; it resolves the correct namespace based on the namespace of the operation's argument(s). But `temp_directory_path()` doesn't take an argument. The source of the problem is a misconception that the free-functions performing FS operations are part of the `class path` API, when really they are part of the API of an implicit local-filesystem object. The Boost.Filesystem FAQ asks: > **Why are paths sometimes manipulated by member functions and > sometimes by non-member functions?** > > The design rule is that purely lexical operations are supplied as > class path member functions, while operations performed by the > operating system are provided as free functions. This is wrong because the majority of the non-member operations don't manipulate paths at all. They are functions that manipulate the filesystem, some of which use a path, and, therefore, are really part of the API of an implicit local filesystem object. The proposed solution (later) just makes this explicit, in a backward-compatible way. Is modifying Boost.Filesystem necessary? ---------------------------------------- Before we discuss our proposal, let us explore how much can be achieved without any changes to Boost.Filesystem, using the typical ADL approach to generic programming. ### Limited solution: ADL on filesystem-specific path If each filesystem were to declare a `path` class conforming to the `boost::filesystem::path` interface, ADL could resolve a *subset* of the filesystem operations. The example below implements a simple generic algorithm, `remove_if_temporary` over both Boost.Filesystem and an imaginary `ftp_filesystem`. #include <iostream> #include <boost/filesystem.hpp> namespace ftp_filesystem { class path { public: path(const std::string& server, const std::string& location) : server_(server), location_(location) {} std::string string() const { return location_; } /* ... */ private: std::string server_; std::string location_; }; void remove(const path& target) { std::cout << "Removing " << target.string() << " over FTP" << std::endl; }; } template<typename Path> void remove_if_temporary(const Path& file_path) { if (file_path.string()[file_path.string().size() - 1] == '~') { remove(file_path); } } int main() { remove_if_temporary(boost::filesystem::path("/tmp/bob")); remove_if_temporary(boost::filesystem::path("/home/bob~")); remove_if_temporary( ftp_filesystem::path("ftp.example.com", "/tmp/bob")); remove_if_temporary( ftp_filesystem::path("ftp.example.com", "/home/bob~")); } The key to this solution is that each filesystem's namespace maintains a separate `path` type. The algorithm is instantiated with different path types, allowing ADL to resolve the necessary implementations. However, this is not a good solution because it is continues the pretence that operations are part of `class path`, but this pretence only stretches so far. The result is that: 1. Operations that don't take a `path` argument can't be used in generic algorithms. Some operations, such as `temp_directory_path()` just *return* a `path`. ADL won't resolve those. 2. `path` becomes coupled to a particular filesystem. `boost::filesystem::path` is a currently a good abstraction of many filesystem's path representation, but making it as the ADL key that resolves operations to their implementation means that other filesystems can't use it as their path representation (would `BOOST_STRONG_TYPEDEF` solve this?). 3. `path` must carry filesystem instance data. The local filesystem is a singleton so its path only needs the location-on-disk string. But other types of filesystem may be instantiated with host names, authentication information, running network sessions, etc. The path instances have to carry this extra baggage. In addition, important parts of the API, namely file opening/creation and directory iteration, can still not be used generically, without changes to Boost.Filesystem, because they are *classes* constructed with a `path`, rather than free functions taking a `path` argument. Solving this will need additional factory functions for each class so that ADL can resolve them. **In summary**: generic filesystem programming is possible with no changes to Boost.Filesystem if generic algorithms don't open files, create files, iterate directories or use any nullary filesystem operations. Even accepting these limitations, the API is not ideal because of undesirable coupling between `path` and the specifics of a filesystem. ### Proposed solution: Add local filesystem object to Boost.Filesystem The main problem with the above solution is that nullary operations cannot be resolved by ADL. In effect, these operations take a hidden local filesystem argument, but, as the argument is implicit, ADL cannot use it to resolve them The suggestion in this proposal is to make the filesystem object explicit so that: - ADL can resolve all the operations using the namespace of this object; or - the operations can be resolved as methods of this object. Either change is backward-compatible. The existing functions remain in place, and code that only uses the local filesystem need never refer to the local filesystem object explicitly. The example below demonstrates the second option. This is my preferred option because all filesystems, except the local one, are likely to implement their operations in terms of private data in the filesystem object. Passing this object to free functions, instead of calling a method, means this private data must be exposed. The example below demonstrates a partial implementation of the additional new `class filesystem`, and shows how to use it generically. #include <boost/filesystem.hpp> namespace ftp_filesystem { class filesystem { public: typedef boost::filesystem::path path; filesystem(const std::string& server) : server_(server) {} path temp_directory_path() const { std::cout << "Querying " << server_ << " for $TMPDIR\n"; return "/tmp"; } void remove(const path& target) { std::cout << "Removing " << target.string() << " on " << server_ << " over FTP" << std::endl; }; private: std::string server_; }; } /* Added to existing Boost.Filesystem API */ namespace boost { namespace filesystem { class filesystem { public: typedef boost::filesystem::path path; path temp_directory_path() const { return temp_directory_path(); } void remove(const path& target) { remove(target); }; private: filesystem() {} friend boost::filesystem::filesystem& local_filesystem(); }; /* Local filesystem is singleton */ inline boost::filesystem::filesystem& local_filesystem() { static filesystem instance; return instance; } }} template<typename Filesystem> void remove_if_temporary( Filesystem& fs, const typename Filesystem::path& file_path) { std::string name = file_path.filename().string(); if (name[name.size() - 1] == '~' || file_path.parent_path() == fs.temp_directory_path()) { fs.remove(file_path); } } int main() { ftp_filesystem::filesystem ftp_fs("ftp.example.com"); boost::filesystem::filesystem& local_fs = boost::filesystem::local_filesystem(); boost::filesystem::path in_temp_dir("/tmp/bob"); boost::filesystem::path temp_by_name("/home/bob~"); remove_if_temporary(ftp_fs, in_temp_dir); remove_if_temporary(ftp_fs, temp_by_name); remove_if_temporary(local_fs, in_temp_dir); remove_if_temporary(local_fs, temp_by_name); } As well as the technical reasons for introducing this change, I believe it better reflects the division of responsibilities between `class path` and the filesystem. Paths are not coupled to the filesystem and can be shared between supporting filesystems. If a particular filesystem needs a different path implementation, that is supported too and it is up to the generic algorithm whether it works with all path implementations or just `boost::filesystem::path` instances. Filesystem details are encapsulated in the filesystem object alone and it performs all the operations using those internal details. The example above implements <code>class filesystem</code> in terms of the existing operations functions so that it will compile with no further changes. A better idea may be to implement the existing operations in terms of the new filesystem object's methods because that better reflects their meaning. Impact on Boost.Filesystem -------------------------- The impact on existing users of Boost.Filesystem should be minimal. The only change they may notice is extra documentation for generic use of the API. ### Backward compatibility No existing code using Boost.Filesystem will need any changes. The existing free functions remain in place as a sensible convenience for the common case where only the local filesystem is needed. ### Additions A new object `boost::filesystem::filesystem` is introduced, along with its singleton accessor function, `local_filesystem`. Depending which of the two API options is chosen: - the operation free-functions are overloaded to take an instance of `boost::filesystem::filesystem`; or - `class boost::filesystem::filesystem` is given methods that perform each of the operations currently in free functions. Even functions `absolute()` and `canonical()`, which may seem to be part of the path API, are really part of the filesystem object's API: they can be implemented solely in terms on `class path`'s public API but not solely in terms of `class filesystem`'s API. Finally, the classes in the API that perform operations on the filesystem, e.g. `fstream`, `directory_iterator` require one of the following changes: - Add factory functions for each class to `namespace boost::filesystem`. Allows them to be resolved by ADL. Generic callers use auto as return type. - Add static factory functions for each class to the `path` class. Allows them to be resolved by ADL. Generic callers use auto as return type. - Add typedefs for each class to `path` class. Callers construct classes directly. Factory functions for streams will require a workaround in C++03 because the standard streams are not movable. TODO ---- - Define filesystem-path concept - Define filesystem concept -- Swish - Easy SFTP for Windows Explorer (http://www.swish-sftp.org)

6 12

[ptr_container] pull request: test values were swapped
by Ben Pope 22 May '14

22 May '14

https://github.com/boostorg/ptr_container/pull/1 Ben

3 12