- Boost - lists.boost.org

[review] Formal review of the Smart Container library starts today
by Pavol Droba 08 Oct '04

08 Oct '04

Hi all, The review of the Smart Container library, written by Thorsten Ottosen starts today (September 26th, 2004) and runs for 10 days (until October 5th, 2004). The latest version of the library can be found at http://boost-sandbox.sourceforge.net/smart_container.zip or http://groups.yahoo.com/group/boost/files/pointer%20container/smart_contain… The package contains all you need to sucessfuly use the library with the Boost version 1.31 or later, however library tests require the latest Boost.Test, that is available from the main Boost CVS. What does the library provide? The library documentation states: >>>>>>>>>>>>>>>> Whenever a programmer wants to have a container of pointers to heap-allocated objects, there is usually only one exception-safe way: to make a container of smart pointers like boost::shared_ptr. This approach is suboptimal if 1. the stored objects are not shared, but owned exclusively, or 2. the overhead implied by smart pointers is inappropriate This library therefore provides standard-like containers that are for storing heap-allocated or cloned objects (or in case of a map, the mapped object must be a heap-allocated or cloned object). For each of the standard containers there is a smart container equivalent that takes ownership of the objects in an exception safe manner. In this respect the library is intended to solve the so-called polymorphic class problem. The advantages of smart containers are 1. Exception-safe and fool proof pointer storage and manipulation. 2. Notational convenience compared to the use of containers of pointers. 3. Can be used for types that are neither Assignable nor Copy Constructible. 4. No memory-overhead as containers of smart pointers can have (see 11 and 12). 5. Usually faster than using containers of smart pointers (see 11 and 12). 6. The interface is slightly changed towards the domain of pointers instead of relying on the normal value-based interface. For example, now it is possible for pop_back() to return the removed element. The disadvantages are 1. Less flexible than containers of smart pointers <<<<<<<<<<<<<<< As an example of how the new containers look like, consider that <code> typedef boost::shared_ptr<Foo> foo_ptr; std::vector<foo_ptr> vec; vec.push_back( foo_ptr( new Foo ) ); (*vec.begin())->bar(); </code> now becomes <code> boost::ptr_vector<Foo> vec; vec.push_back( new Foo ); vec.begin()->bar(); </code> In your reviews, please include the answers for the following questions: * What is your evaluation of the design? * What is your evaluation of the implementation? * What is your evaluation of the documentation? * What is your evaluation of the potential usefulness of the library? * Did you try to use the library? With what compiler? Did you have any problems? * How much effort did you put into your evaluation? A glance? A quick reading? In-depth study? * Are you knowledgeable about the problem domain? And finally, every review should answer this question: * Do you think the library should be accepted as a Boost library? Be sure to say this explicitly so that your other comments don't obscure your overall opinion. Also, in the documentation you will find questions from the submitter to the reviewer; whenever you feel you have something to contribute on these issues, please don't hessitate to include that in your review. Thank you, Pavol Droba, review manager for the Smart Containers library.

13 57

people page
by Fernando Cacciola 08 Oct '04

08 Oct '04

Hi, Should mantainers of released libraries (like me) add themselves to the people page, or we need to be explicitely invited for this? And in that case, should we upload ourselves the photos or should we send them to one of the moderators instead? TIA Fernando Cacciola SciSoft

4 7

Re: Removing dots from toolset names
by Aleksey Gurtovoy 08 Oct '04

08 Oct '04

Jürgen Hunold writes: > Hi ! Hi Jürgen, > > On Tuesday 05 October 2004 15:10, Aleksey Gurtovoy wrote: > > Hi All, > > > > As you probably already know, we also need to take care of these: > > Why ? Can someone point me to the rationale for this ? The original rationale was "to ensure file and directory names are relatively portable": http://www.boost.org/more/lib_guide.htm#Directory_structure. Beman might be able to clarify whether ISO 9660/Level 2 requirements in particular were one of the motivating factors back then or not, but currently those are the primary driving force for getting Boost codebase to conform to these rules: We want to be able to put a Boost distribution on a CD in the unpacked form, and for that CD to be readable on the maximum number of platforms. > Where can I find out which platforms have which limits ? WWW :). Here, for instance -- http://www.zplace.com/crashtips/filename.html. > > > intel-win32-7.1-vc6-stlport-4.5.3-tools.jam: filename > 31 chars, filename > > contains more than one dot character ('.') intel-win32-7.1-vc6-tools.jam: > > Some of these toolsets exist in current 1.31.0 with _more_ than one dot. > Take vc7.1-stlport-tools.jam for example. Sure. The requirements weren't strictly inforced until now. > > > I suggest we either replace the dots with underscores, or get rid of them > > completely (besides the last one, of course), i.e. either: > > > > borland-5_5_1-tools.jam > > borland-5_6_4-tools.jam > > como-win32-4_3_3-vc7-tools.jam > > -1 > > > borland-551-tools.jam > > borland-564-tools.jam > > como-win32-433-vc7-tools.jam > > ... > > Preferences? > > +1 Thanks! > > > [I also know that there is more to this than just renaming of the files; > > any help with the whole matter would be greatly appreciated.] > > Just one curious question: > AFAIK, all of the platforms these toolsets are _for_ support both filenames > with more than 31 characters and more than one dot. > I just wonder if we should break and repair the whole V1 build system when > the offending toolsets are for platforms which don't _have_ these limits. The changes are not that drastic, but it's a reasonable question to ask. Ignoring the fact that we already have nonconforming names fixing which requires some work, IMO all the current requirements except 31-character limit are not hindering us in any significant way, are easy to satisfy, and in general have a good cost/benefit ratio. I don't feel strongly about maximum length of 31, in particular since MacOS 9 is pretty much dead, and we _are_ somewhat restrained by the limit. Being able to read a Boost CD on, let's say, Linux versions released before 1999 is definitely going to be appreciated by _somebody_, but there is a certain cost to that, and personally I'm starting to feel that may be it's not worth it. But then, as a release manager who has to take care of this issue, I'm biased. > > Just my .02€, developing on linux and win32. Thanks, they are appreciated. -- Aleksey Gurtovoy MetaCommunications Engineering

3 3

Re: Removing dots from toolset namesHo
by Aleksey Gurtovoy 08 Oct '04

08 Oct '04

Jürgen Hunold writes: > On Wednesday 06 October 2004 17:35, Aleksey Gurtovoy wrote: > > The changes are not that drastic, but it's a reasonable question to > > ask. Ignoring the fact that we already have nonconforming names > > fixing which requires some work, IMO all the current requirements > > except 31-character limit are not hindering us in any significant > > way, are easy to satisfy, and in general have a good cost/benefit > > ratio. > > > > I don't feel strongly about maximum length of 31, in particular since > > MacOS 9 is pretty much dead, and we _are_ somewhat restrained by the > > limit. Being able to read a Boost CD on, let's say, Linux versions > > released before 1999 is definitely going to be appreciated by > > _somebody_, but there is a certain cost to that, and personally I'm > > starting to feel that may be it's not worth it. But then, as a > > release manager who has to take care of this issue, I'm biased. > > I've got this impression, too. > > My old PPC-6100-Mac is rotting in a corner ... ;-((( > > I don't want to spoil the release, but did you take a look at the > BoostBook generated docs ? > At least BBv2 generates files like > "bbv2.reference.buildprocess.alternatives.html" Yep, I'm aware of the issue. Or, rather, I was aware. It's taken care of now -- http://article.gmane.org/gmane.comp.lib.boost.documentation/661. > This will not work when someone tries to ship pre-built documentation on > pure ISO-9660 CD, either. > And I thought that a release tarball should contain a pre-built > documentation because no-one can be expected to install all the > necessary components to build it. Yep, it should and it will. > Do you/we want to change this, too ? > And have *lots* of cryptic, human-unreadable file names ? Well, the name length is still partly an issue, but certainly not to the extent it was before the subdirectories fix. And even the overly long names don't necessarily have to be turned into cryptic ids. For instance, something like "static_local_time_adjustor.html" doesn't have to become "id413791.html"; it can be "static_local_time_adjust00.html", or "static_local_time_00.html", or something else along these lines. Having said that, I'm going to repeat myself -- while I think all other requirements are worth sticking to, I don't feel strongly about keeping the 31-character limit. But now that we have not that many names to fix, it's tempting to go with it to the end at least for this release. > When we discuss this further, we should do it on Boost-Mailing-List Transferred; Cc'd to jamboost. -- Aleksey Gurtovoy MetaCommunications Engineering

1 0

Re:Re: [boost] [serialization] test_level_class_info_load*...
by Robert Ramey 08 Oct '04

08 Oct '04

< .... Thank for a very clear and concise explanation to something that is just about impossible to understand. >The solution I suggested before does the dependency: ><directory>test_*_load_*.test --> <->test_*_save_*.test >Which make that last line of the DAG above: >test --> *.test -> *.run -> *.exe, *.obj, etc -> *_load_*.test directory >-> *_save_*.test >Make sense? More or less. If I understand this correctly, then on a second run if the directory is already created, the desired dependency won't be enforces. That is the following scenario won't work a) test_..._save builds but fails at runtime. This creates all directory structures for test...save. b) test...._load will build and run but will fail because test...save failed to create an temporary file that test...load needs. >PS. If you are sure you are only going to have run tests. You could do >things like: >local test-part-2 = [ test-bsl-run_files test_class_info_load ] ; >local >test-part-1 = [ test-bsl-run_files test_class_info_save ] ; >DEPENDS $(test->part-2:S=.run) : $(test-part-1:S=.run) ; Actually I think this latter is more a kin to what is required. The dependency is that test..save must be executed successfully to create the appropriate output file to test the run of the corresponding test...load. Even if the test..save fails to build or run, it is valuable to build (though not run) to check for compilation and linkage errors) Your explanation has motivated to spend a little time with this. Robert Ramey

1 0

Re: [boost] Re: [multi_index] problem with gcc 3.4.2 for Compaq Tru64UNIX
by JOAQUIN LOPEZ MU?Z 08 Oct '04

08 Oct '04

----- Mensaje original ----- De: Markus Schöpflin <markus.schoepflin(a)comsoft.de> Fecha: Viernes, Octubre 8, 2004 5:04 pm Asunto: [boost] Re: [multi_index] problem with gcc 3.4.2 for Compaq Tru64 UNIX > Markus Schöpflin wrote: > > > Joaquín Mª López Muñoz wrote: > > > >> mips-tfile, /tmp//cctFjnCI.s:10965 string too big (35326 bytes) > >> line:? #.stabs?"and_<boost::mpl::not_<boost::is_same<boost::mpl: > >> [very long expression] > > > > A quick google search indicates that this is a problem with gcc > on alpha > > when compiling with debug symbols. A possible workaround might > be to use > > -gstabs instead of -g, I'll try it ASAP. > > I did some experiments and the only way to compile these files it > to > disable debug symbols completely. I have done that on the toolset > for gcc > used for the regression runs, I hope it helps to clean up some of > the errors. > Thanks for your help! Indeed, Boost.MultiIndex generates horribly long symbol names, it's a pita when trying to decode compiler errors (comparatively, regular STL errors are even pleasant to deal with). I'll have to think some way to shorten them in future releases. Joaquín M López Muñoz Telefónica, Investigación y Desarrollo

1 0

[multi_index] problem with gcc 3.4.2 for Compaq Tru64 UNIX
by Joaquín Mª López Muñoz 08 Oct '04

08 Oct '04

Seems like some multi_index tests are failing due to lack of resources or something like that, for COMSOFT reports. http://tinyurl.com/66t2h mips-tfile, /tmp//cctFjnCI.s:10965 string too big (35326 bytes) line:? #.stabs?"and_<boost::mpl::not_<boost::is_same<boost::mpl: [very long expression] Anyone has an idea what this could be about? Thank you, Joaquín M López Muñoz Telefónica, Investigación y Desarrollo

2 2

Problematic inclusion of "pthread.h".
by Markus Schöpflin 08 Oct '04

08 Oct '04

When compiling the smart pointer tests with gcc 3.4.2 on Tru64, two tests fail with an error message. (See http://tinyurl.com/4jngu ) As far as I can see, the shared pointer implementation at some point includes the "pthread.h" system header file but gcc on Tru64 requires that -pthread must be specified on the command line when including this header file. When running this test with gcc 3.4.2 on Linux the file "pthread.h" is also included but gcc doesn't require the addition of -pthread in the command line on this platform. This leads to the question whether it is ok to include "pthread.h" without specifying -pthread on the command line. Does anyone know an answer to this? Markus

5 26

Re: uBLAS broken on MSVC 6.5
by Michael Stevens 08 Oct '04

08 Oct '04

Alekskey, > Seems like some recent changes brought MSVC to its knees -- > http://tinyurl.com/3vmlv. Fix is already committed. Tomorrow we should be all green again. My plan is for this release is to get maximum compatibility with the older compilers. In particular VC6.5 7.0, GCC 2.95, Intel 7.0, and CV8.3 Then after this release I can drop many of the multitude of workaround necessary for these. With regard to Code Warrior I was hoping to develop a fix for the debug mode compilation errors show up on you metacomm test. I didn't come up with anything satisfactory however. Do you have any idea why there is such a large difference between your cw-8.3 and the RSI's cw-8.3-release tests? I can only see the command line options in the failing build. The only option I can see which could be of relevance is the '-iso-template on'. If you are both using stock versions of the build system this same option should be set for the release build as well. Thanks for you interest (and hard work), Michael

2 2

Re: Re: [boost] [serialization] test_level_class_info_load*...
by Robert Ramey 08 Oct '04

08 Oct '04

>First, you need to "fix" your Jamfile to return the test targets: >... OK >=================================================================== >Having that working, lets you define the tests so that they depend on >each other (indirectly). For example: >=================================================================== >local test-suite-serialization-part-1 = > [ test-bsl-run_files test_array ] > [ test-bsl-run_files test_binary ] >; >local test-suite-serialization-part-2 = > [ test-bsl-run_files test_contained_class ] > [ test-bsl-run_files test_cyclic_ptrs ] >; >test-suite serialization : > $(test-suite-serialization-part-1) > $(test-suite-serialization-part-2) >; >DEPENDS $(gLOCATE($(test-suite-serialization-part-2):G=directory-grist) > : $(test-suite-serialization-part-1) >; Now - out of curiosity, most of the rule are of the form: rule test-bsl-run_polymorphic_archive ( test-name : sources * ) which suggests to me that I could just use: test-suite serialization : [ test-bsl-run_files test_level_class_info_load test_level_class_info_save ] Which would make ...load dependant on ...save. Or this is this answered by the following: >That last DEPENDS makes the directory for the part 2 depend on the >completion of part 1. You may ask why the directory and not the test >itself? Strangely it's because to properly chain the tests you need to >make the earliest step of the second test depend on the latest part of >the next test. And directories are the only sure thing that you can know >a test needs first. Robert Ramey

2 1