- Boost - lists.boost.org

Do Tools Need Formal Reviews?
by Vinnie Falco 10 Oct '24

10 Oct '24

Traditionally, tools used by authors and maintainers are quietly integrated with oversight from owners and release managers. Examples include saxon-he and docca (https://github.com/boostorg/docca) Recently there have been noises made expressing the desire for the formal review process to be applied to tools as well. The purpose of this post is to explain why this is a bad idea. There are two kinds of tools. External tools, brought into use by the project through the scripts which create the release, and internal tools which are in repositories added to the superproject. An example of an external tool is asciidoctor: https://asciidoctor.org/ This was quietly added and the command line tool is required to be installed in order to build a Boost release, as libraries such as Boost.mp11 or Boost.Unordered need it for producing their documentation. The Asciidoctor author provides the installation package and provides all support. The only documentation requirement for Boost libraries is that they render into HTML and that "b2 doc" works. A formal review to adopt "asciidoctor" as an available external tool could result in REJECT. This serves no useful purpose other than to make things more difficult for the author or maintainer. They can still use asciidoctor but would have to do it locally and commit the resulting HTML output to the repository, as the tool would not be available to the release process. This is objectively worse for everyone. An example of an internal tool is BoostLook: https://github.com/boostorg/boostlook This is a set of CSS files and Ruby scripts used by some libraries to style their Asciidoctor documentation. This was quietly added to the superproject as the release will need it for some of the libraries to build. Discussion with owners happened beforehand; the files for this tool were originally in the website-v2-docs repo (https://github.com/boostorg/website-v2-docs) but the objection was made (rightly so) that the website-v2-docs repo was too large to add to the superproject. Thus the files were put into their own small repository. A formal review to adopt BoostLook as an internal tool could result in a REJECT outcome. This serves no useful purpose other than to make things more difficult for maintainers. They would manually inline the boostlook files in their repository and be responsible for keeping it up to date (submodules in libraries are intentionally not cloned by the release script). Tools are not user-facing (where "user" is defined as a Boost user and not an author or maintainer). I've seen a lot of intellectual gymnastics trying to justify why tools are the equal of libraries but they fail in light of an obvious fact: tools no longer in use can be removed from the superproject without a tsunami of mailing list protest. It is true that tools which Boost end-users find useful can be adopted, but they do not come with the same promise of forever support the way that libraries do. They are only supported to the extent to which the Boost libraries need them. My maybe-unpopular opinion: b2 was correct to move to an independent repository, as this can come with implied lifetime support. There's a recurring theme: tools are not libraries, despite Robert Ramey's sincere efforts to formally review them as such. A tool is an administrative construct designed to facilitate the efforts of authors and maintainers when they wish to make use of the tool in more than one library. The use of the tool is still possible in a Boost library, without the tool becoming part of the superproject, yet with more effort and opportunity for mistakes. It is for this reason that we do not use the formal review process for tools, as rejecting a tool has only disadvantages for existing libraries. The very last thing we need to do now is to make things harder for authors and maintainers. An argument used to justify a formal review is that the resulting feedback can improve the tool. I agree that feedback is generally helpful, this is true even if there is no formal review. That is, there is no requirement for a formal review to be ongoing for someone to provide feedback. Just provide the feedback. Tools are boring, and unlikely to attract a significant number of review responses. Formal reviews of proposed internal tools which receive a very small number of reviewers do not signal strong consensus and I believe harm morale. The informal process used for oversight on the internal and external tools Boost uses to produce its releases has worked well, and no convincing rationale has been put forth for why it should change. We should not invoke the formal review process for tools. Thanks

8 11

Who can fix CI ?
by hermann＠stamm-wilbrandt.de 09 Oct '24

09 Oct '24

5 days ago I created my first PR: "new planar_vertex_six_coloring() algorithm with examples, tests and doc" https://github.com/boostorg/graph/pull/387 CI helped me to fix things I was responsible for. But after that CI still fails in code I am not responsible for. From release notes: https://www.boost.org/users/history/version_1_86_0.html "...Graph: Major update: C++14 is the new minimum standard; this was partly dictated by dependencies (at least to C++11) and partly by choice. If you require support for an older standard, please contact the maintainer. ..." Somebody (who?) forgot to tell CI that minimum requirement. Example error caused by this: gcc.compile.c++ ../../../bin.v2/libs/graph/test/gursoy_atun_layout_test.test/gcc-5/debug/x86_64/cxxstd-11-iso/threading-multi/visibility-hidden/gursoy_atun_layout_test.o In file included from ../../../boost/math/constants/constants.hpp:11:0, from ../../../boost/graph/topology.hpp:16, from ../../../boost/graph/gursoy_atun_layout.hpp:34, from gursoy_atun_layout_test.cpp:10: ../../../boost/math/tools/config.hpp:28:6: warning: #warning "Boost.Math requires C++14" [-Wcpp] # warning "Boost.Math requires C++14" ^ In file included from ../../../boost/math/tools/mp.hpp:15:0, from ../../../boost/math/policies/policy.hpp:11, from ../../../boost/math/constants/constants.hpp:16, from ../../../boost/graph/topology.hpp:16, from ../../../boost/graph/gursoy_atun_layout.hpp:34, from gursoy_atun_layout_test.cpp:10: ../../../boost/math/tools/type_traits.hpp:208:12: error: 'std::is_final' has not been declared using std::is_final; Since this is my first PR, will there be someone helping me in PR 387? Who owns CI? Can issues be created against CI? Or do I understand this completely wrong? Regards, Hermann.

2 2

(no subject)
by Christophe Henry 08 Oct '24

08 Oct '24

> Should one or more of these libraries be marked as "deprecated" in their > corresponding meta/libraries.json? > > Array > Assign > Chrono > Compatibility > compressed_pair > enable_if > ForEach > functional > mem_fn > move > MPL > Ratio > ResultOf > Typeof Yes for marking old libraries as obsolete, as it does not convey a "might be removed" tag. No for marking as deprecated for this reason. Yes for hiding obsolete libraries by default. Strong no for ever even thinking of removing a Boost library from 1.x. Strong no for marking as obsolete libraries which are using obsolete libraries in the implementation. This one would force me to drop parts of MSM which might be used in production code and I strongly oppose the idea. Christophe

7 20

Scheduled review of Decimal library
by Mateusz Loskot 08 Oct '24

08 Oct '24

Folks, On behalf of the Review Wizards Team, formal review of proposed Decimal library has now been scheduled: - https://github.com/boostorg/website/pull/878 The review details should be soon displayed on the website - https://www.boost.org/community/review_schedule.html Best regards, Mateusz Loskot

4 6

[context] Build error with context-impl
by Tim Haines 08 Oct '24

08 Oct '24

git -c advice.detachedHead=false clone --branch develop --single-branch --depth 1 https://github.com/boostorg/boost.git git -c advice.detachedHead=false submodule update --init --recursive ./bootstrap.sh --with-toolset=gcc --with-libraries=context ./b2 -d+2 --debug-configuration --debug-building --debug-generator headers context-impl=fcontext toolset=gcc cxxstd=20 notice: found boost-build.jam at /boostorg/boost-build.jam notice: loading B2 from /boostorg/tools/build/src/build-system.jam notice: Searching <redacted> for site-config configuration file 'site-config.jam'. notice: Configuration file 'site-config.jam' not found in <redacted>. notice: Searching <redacted> for user-config configuration file 'user-config.jam'. notice: Configuration file 'user-config.jam' not found in <redacted>. notice: Searching '.' for project-config configuration file 'project-config.jam'. notice: Loading project-config configuration file 'project-config.jam' from '.'. notice: will use 'g++' for gcc, condition <toolset>gcc-11 notice: using gcc libraries :: <toolset>gcc-11 :: /usr/bin /usr/lib /usr/lib32 /usr/lib64 notice: using gcc archiver :: <toolset>gcc-11 :: /usr/bin/ar warning: toolset gcc initialization: can not find tool windres warning: initialized from project-config.jam:12 notice: using rc compiler :: <toolset>gcc-11 :: /usr/bin/as /boostorg/tools/build/src/build/feature.jam:327: in validate-feature from module feature error: unknown feature "<context-impl>" /boostorg/tools/build/src/build/feature.jam:361: in expand-subfeatures-aux from module feature /boostorg/tools/build/src/build/feature.jam:422: in feature.expand-subfeatures from module feature /boostorg/tools/build/src/build/build-request.jam:20: in apply-to-property-set from module build-request (builtin):-1: in sequence.transform from module sequence /boostorg/tools/build/src/build/build-request.jam:32: in build-request.expand-no-defaults from module build-request /boostorg/tools/build/src/build-system.jam:778: in module scope from module build-system The feature is present: grep -A5 -n "feature.feature context-impl" libs/context/build/Jamfile.v2 52:feature.feature context-impl 53- : fcontext 54- ucontext 55- winfib 56- : propagated 57- composite I'm not familiar with the Jam DSL, so I'm not sure how to debug this further. Any advice is very welcome. Thanks. - Tim

2 2

[Boost] runtime+memory evaluation of 6 BGL planar graph algorithms
by hermann＠stamm-wilbrandt.de 06 Oct '24

06 Oct '24

Half a year ago I copied together Boost examples making use of: boost/graph/make_connected.hpp boost/graph/make_biconnected_planar.hpp boost/graph/make_maximal_planar.hpp boost/graph/boyer_myrvold_planar_test.hpp boost/graph/planar_canonical_ordering.hpp boost/graph/chrobak_payne_drawing.hpp I added "read_leda_graph()" for processing graph files. And added code to write a graphiviz file from the generated straight line drawing. Use of "layout=neato" with fixated vertex coordinates with exclamation mark (pos="3,2!") misuse GraphViz to just draw the BGL determined straight line drawing. I did evaluation on random maximal planar graphs with 10,000 up to 100 million vertices in 10x steps. In 1993 I wrote a technical report on how to create maximal planar random graphs. I published the 1994 code two years ago (self written graph lib, not LEDA nor Boost): https://github.com/Hermann-SW/randomgraph All details of the evaluation including runtime table in this "straight_line_graphviz.cpp" gist comment: https://gist.github.com/Hermann-SW/99d151a273d290ee0d843c79b2da26a8?permali… Total runtime factors are 13.5x and 20.9x between 1, 10 and 100 million vertex graphs. Since the input graphs were maximal planar, the algorithms make_connected / make_biconnected_planar and make_maximal_planar did not have much to do. The resident RAM usage is perfectly linear, 2.8 / 28.1 / 281.4 GB for 1 / 10 / 100 million vertices. 100 million vertices, 300 million edges, edge property, planar ordering, planar straight line drawing and 8 byte storage for numbers on 64bit OS cause a factor of 2800 comparing resident RAM used with number of vertices. Regards, Hermann.

1 0

Scheduled review of MQTT library
by Mateusz Loskot 06 Oct '24

06 Oct '24

Folks, On behalf of the Review Wizards Team, formal review of proposed MQTT library has now been scheduled: - https://github.com/boostorg/website/pull/877 The review details should be soon displayed on the website - https://www.boost.org/community/review_schedule.html Best regards, Mateusz Loskot

2 4

[optional] Feedback on the range interface for Boost.Optional
by Andrzej Krzemienski 05 Oct '24

05 Oct '24

Hi Everyone, I have just recorded the following message in the Mailing List. sob., 14 wrz 2024 o 02:44 Alan de Freitas via Boost <boost(a)lists.boost.org> napisał(a): > - As people seem to imply, the WG21 issue goes way beyond the motivation > behind the proposals or the difficulty behind the Boost review process. Not > all Boost libraries are intended to go through WG21. Not all WG21 proposals > would make good Boost libraries (like begin/end iterators for > std::optional). We are in the process of implementing the range interface for Boost.Optional, as proposed in P3168R2 <https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2024/p3168r2.html> and to the extent possible in C++11. I would like to hear opinions from the people in this list, on if and why this would be a bad idea. Regards, &rzej;

9 18

[Boost][BGL] "is_straight_line_drawing()" is buggy for large graphs
by hermann＠stamm-wilbrandt.de 05 Oct '24

05 Oct '24

A while back I combined all BGL planarity functions (make_"connected / biconnected_planar / maximal_planar" + straight_line_drawing) prepended with "read_leda_graph()" and writing GraphViz output. The Graphviz output uses neato engine because that allows for fixed vertex coordinates with exclamation mark: 2 [pos="3,2!" label="2"] So GraphViz is just misused to display BGL straight line drawing Recently I enhanced it to time each relevant BGL function call in order to verify linear runtime as well as to investigate RAM usage. In 1993 I created a technical report about creation of random maximal planar graphs and published the code 2 years ago: https://gist.github.com/Hermann-SW/99d151a273d290ee0d843c79b2da26a8 I created maximal planar randomgraphs on 10,000, 100,000, 1million and 10million vertices, and used them as input for above timing code. All was fine until and including 1million vertex graph. But for 10million vertex graph final "is_straight_line_drawing()" incorrectly asserted: hermann@7950x:~$ randomgraph/randomgraph 10000000 -t maximal_planar -o 10000000.u ------------------------------------- malloc: 18 malloc-free: 0 hermann@7950x:~$ time ./straight_line_graphviz 10000000.u t read_leda_graph(g, argv[1]); 9.63147 make_connected(g); 1.82002 assert(boyer_myrvold_planarity_test(boyer_myrvold_params::graph = g, boyer_myrvold_params::embedding = &embedding2[0])); 73.5683 make_biconnected_planar(g, &embedding2[0]); 25.9196 assert(boyer_myrvold_planarity_test(boyer_myrvold_params::graph = g, boyer_myrvold_params::embedding = &embedding2[0])); 88.7905 make_maximal_planar(g, &embedding2[0]); 42.7211 assert(boyer_myrvold_planarity_test(boyer_myrvold_params::graph = g, boyer_myrvold_params::embedding = &embedding2[0])); 96.7233 assert(boyer_myrvold_planarity_test(boyer_myrvold_params::graph = g, boyer_myrvold_params::embedding = embedding)); 68.5428 planar_canonical_ordering(g, embedding, std::back_inserter(ordering)); 24.2465 chrobak_payne_straight_line_drawing( g, embedding, ordering.begin(), ordering.end(), straight_line_drawing); 4.49772 assert(is_straight_line_drawing(g, straight_line_drawing)); a.out: straight_line.cpp:163: int main(int, char**): Assertion `is_straight_line_drawing(g, straight_line_drawing)' failed. Aborted (core dumped) real 7m25,836s user 7m15,357s sys 0m10,356s hermann@7950x:~$ I added code into "graph/include/boost/graph/is_straight_line_drawing.hpp" to write the coordinates of the two edges that intersect: hermann@7950x:~$ cat intersect.txt a 4143438 86426 4064945 7932 4064944 7931 4143438 86426 hermann@7950x:~$ As you can see, both edges share vertex at coordinate "4143438,86426", but they do not intersect. Until that point in code the coordinates are std::size_t. But the called function uses double, causing false report: inline bool intersects(double x1, double y1, double x2, double y2, double a1, double b1, double a2, double b2, double epsilon = 0.000001) { I played with different epsilon values including 0, but the valid straight line drawing was always stated to be wrong. There are several algorithms for segment intersection with integer coordinates that don't need division and are correct for integer coordinates that are result of chrobak_payne_straight_line_drawing( g, embedding, ordering.begin(), ordering.end(), straight_line_drawing); https://en.wikipedia.org/wiki/Line%E2%80%93line_intersection#Given_two_poin… or https://www.dcs.gla.ac.uk/~pat/52233/slides/Geometry1x1.pdf#page=4 1) I want to create a BGL issue on this, is that OK? At least the doc does not state any limit on graph sizes: https://www.boost.org/doc/libs/1_86_0/libs/graph/doc/is_straight_line_drawi… 2) I want to create a pull request with new implementation for "intersects()" with integer coordinates. 2 days ago I created pull requests with 7 commits ahead on new "planar_vertex_six_coloring()": https://github.com/boostorg/graph/pull/387 Do I have to wait until that pull request was decided on? Or can I submit separate pull request not interfering with that request on my graph fork? Regards, Hermann.

2 7

Boost.Decimal Seeking Endorsment
by Matt Borland 05 Oct '24

05 Oct '24

Hello, Chris Kormanyos and I are seeking endorsement for our library proposal, Decimal (https://github.com/cppalliance/decimal) What is Decimal? It is an implementation of IEEE 754 and TR 24733[3][5], along with extensions to both. We provide three IEEE 754 compliant types: decimal32, decimal64, and decimal128, as specified in both documents. Since announcing the beta in the first half of the year, we have added three similar but non-compliant types: decimal32_fast, decimal64_fast, and decimal128_fast. These types perform all operations faster than the compliant types, with identical numerical results, at the cost of extra space. We also provide implementations of the majority of the STL as it apply to numerical types. The library is header-only, has no dependencies, and only requires C++14. Why should we have this in Boost? Decimal Floating Point types were standardized in IEEE 754 in the 2008 revision, but it has not been standardized in C++. We believe this library can serve as a reference implementation. An existing alternative would be one of the authors of the IEEE 754 standard helped write libdfp[4]. This is also available in C with GCC as _Decimal32, _Decimal64, and _Decimal128 [1]. We bring these to C++ with similar or better performance. We also bring interoperability with the STL which is missing in the aforementioned libraries. Since the announcement of our beta in the spring we have improved correctness, performance, and gained users. We added the fast types after conversation with Glen Fernandes, and bit conversions for both encodings of the types [2]. I am aware of at least one US-based trading firm using this library in production so the library is out in the field making money today. Should the library be endorsed John Maddock has graciously offered to be our review manager, and are targeting a post-holiday review period in mid-January 2025. Feel free to ask any questions, and open issues prior to the review period. Matt [1] https://gcc.gnu.org/onlinedocs/gcc/Decimal-Float.html [2] https://cppalliance.org/decimal/decimal.html#conversions [3] https://www.open-std.org/JTC1/SC22/WG21/docs/papers/2009/n2849.pdf [4] https://github.com/libdfp/libdfp [5] https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n3871.html

3 4