January 2004 - Boost - lists.boost.org

Re: [boost] Re: impl(ementations) - keeping it simple
by Robert Ramey 28 Jan '04

28 Jan '04

Reece Dunn wrote: >John Torjo wrote: >>Maybe to simplify the process, in boost-root (where boost-build.jam file >>exists), we could have a makefile which automatically builds bjam from the >>source files and the invokes it with a given configuation. >> >>like: >>make msvc >>(instead of bjam "-sTOOLS=msvc") >> >>Unfortunately, I don't have any makefile knowledge. ahhh - we've come full circle. >I am knowledgeable in makefiles, but this leads to a problem: you need to >configure the makefile to handle all the various compiler and platform >combinations - something bjam already handles, so why duplicate this effort? >Is there a bjam file for bjam?! That way it could be bootstrapped. But then >why not just download a pre-built version of bjam? Bascally I think the whole purpose of bjam is to address the concerns of the original poster. Unfortunately, this is not obvious from the information on bjam that comes with the system. I realise that the information is available - its just that is spread around in several places. Bjam is a large subject and requires almost a book to explain it. BUT it (or something like it is necessary for boost - as was obvious to the original poster. The central purpose of the bjam system is to specify how to build programs in a generic way independent of any particular platform while at the same time specify how each platform handles particular aspects of the application developement. This is a very ambitious concept and has resulted (not surprisingly) in a system that is pretty elaborate. A very ambitiuos effort. In one important way, its not ambitious enough. Unlike the rest of boost, it seems directed a bit narrowly to building and testing boost libraries. I get this impression from the fact that much effort is being invested in the code but relatively little in producing a complete, coherent and well organised explanation of how to make best usage of it. Maybe we can sucker someone in writing an O'Reilly book on this which would include the bjam system? Of course this would need a more polished system. I'm sorry go on about this, the thing just drives me crazy. John Torjo - bjam is exactly what you're looking for - its just not clear from the explanation. Robert Ramey

2 1

Re: [administrative] Formal Review Process
by Dave Gomboc 28 Jan '04

28 Jan '04

] David Abrahams wrote: ] ] ] Dave Gomboc <dave(a)cs.ualberta.ca> writes: ] ] ] ] It's a good idea, but I have to ask: is this a bicycle shed ] ] discussion? Does tweaking the seconding process really ] solve any of ] ] Thomas' problems? ] ] At least it's my shed. I think it can reduce the workload for ] the review wizard. ] ] Thomas Quoting issue here: that was Dave A.'s text, not mine. Dave

1 0

Re: [administrative] Formal Review Process
by David Abrahams 28 Jan '04

28 Jan '04

Dave Gomboc <dave(a)cs.ualberta.ca> writes: > ] I suggest that such a requirement be added to the formal > ] review process, with a specific (low, ie, 3) number of > ] "formal seconds" required for a formal review request in > ] order for it to reach the schedule. It should be explicitly > ] noted that "seconding" a review request does not imply a vote > ] to accept, nor a commitment on the part of the person > ] seconding to participate in the review. > > I also like the idea, but with a twist: it should be explicitly > noted that seconding a review request _does_ imply a commitment to > review the library should a review be scheduled for it. It's a good idea, but I have to ask: is this a bicycle shed discussion? Does tweaking the seconding process really solve any of Thomas' problems? -- Dave Abrahams Boost Consulting www.boost-consulting.com

2 1

Re: [boost] Re: [administrative] Formal Review Process
by Jeff Garland 28 Jan '04

28 Jan '04

On Mon, 26 Jan 2004 22:10:23 -0800, Thomas Witt wrote > > Jeff Garland wrote: > > I don't like the idea of a mailing list to solve this problem. > > Why? Really I just want to know. Because the primary issue is a lack of time and I don't see how complicating things with another mailing list solves that problem. Also, if review comments either intentionally or accidently wind up on a review list then there is a third list I now have to search when looking into the archive -- I've had to look on the dev and user groups a few times because I can't remember where a particular discussion occurred. > There is more to the filtering than meets the eye. The usual process > up to know was that somebody posted a request. That request usually > lead to some discussion. As a result in quite a few cases the Which is one reason why I believe this needs to stay on the dev list... > library author decided to tackle the issues before the request. I've > been following these discussions in order to see whether the request > is likely to be withdrawn or not. The net effect of all this a lot > of postings use review in the subject line and are neither requests > nor formal review related. I am not criticising poeple for doing > this I just try to present the current situation. Yep. > Another issue that is related to this problem is that we are already > close to having more review requests than we can handle. (I am not > talking about the wizard here, just the list and the reviewers). I don't agree. There were more reviews in 2002 (15) than 2003 (~10) by my count. My guess is that about 1 full review per month is about what can be reasonably done. I, for example, have not been a review manager for a year so I don't feel over-worked. >In the past we already had reviews where there was very little feedback. That is an issue for the review manager to handle. I would leave it up to the judgement of the review manager to decide if there has been enough participation. If the participation is too low then the library should probably be rejected for lack of interest. But again that is the judgement of the review manager to make. As for the number of review requests, > In my opinion we need some kind of staging in order to streamline > the influx of libraries in the formal review process and to improve > the quality of reviewed libraries. Thus making failure for trivial > reasons more unlikely. I don't recall any specific examples of this. We already have a multi-stage review process. Sure there are examples of people just joining and throwing out a library for formal review before it is ready, but someone then provides guidance on the process. > I think the idea of a separate mailing list fits into this staging > strategy. I.e. the review list would be reserved for review and > review only. One idea might be to require a pre review on the > developers list or required 2-5 people to second the review request. Sure, but it complicates life for users trying to decide which lists to subscribe to, post to, and search. There won't be anything to stop someone from posting a half-baked library on the new list or accidentally posting a review request on the dev list... > > I think some simple additions to the submission process would solve the > > 'heads-up' to the review manager problem: > > A copy of the Review Request posting should be sent directly to > > the review wizard (email here). If the review wizard does not > > respond with an acknowledgement within 48 hours another request > > should be made. > > As explained not every request leads to a review. Sure, but at least the submitter would be aware that you are paying attention. So the acknowledgement is not a date for the review, but just a little "hello, I see your request, have you read the submission guidelines" sort of mail. > > Also, note that the entire discussion of the Review Wizard is on > > the Formal schedule the review. > > I would really appreciate it if you could make that change. Sure. > > Finally, it seems to me the real solution is to either split the duties or > > find another review wizard volunteer. > > As of now I don't have a good idea how things can be split up. Perhaps you can offload the monitoring or organization of a particular upcoming review -- I'm sure we can figure something out. > > I'm willing to spend a couple hours a > > month on this if it would help, so Thomas send me an email > > off-list if you have something you want to offload on me. > > Thanks Jeff! I'll shoot you an email as soon as I see clearer. Let me know. Jeff

3 3

Re: [boost] Re: [administrative] Formal Review Process
by Douglas Gregor 28 Jan '04

28 Jan '04

On Tuesday 27 January 2004 01:20 am, Thomas Wenisch wrote: > This idea of seconding a review request is an extremely good idea. It > adds formality to what is captured in the current process as "post to list > and see if there is interest". The fact that someone seconds a review > request demonstrates that there is interest in the library, and that there > are people who are following the review list that might be inclined to > review. > > I suggest that such a requirement be added to the formal review process, > with a specific (low, ie, 3) number of "formal seconds" required for a > formal review request in order for it to reach the schedule. It should be > explicitly noted that "seconding" a review request does not imply a vote > to accept, nor a commitment on the part of the person seconding to > participate in the review. It just is an indication that a library is > interesting and worthy of review (and worthy of consuming time on the > schedule that could be used reviewing other libraries). I too like the idea of seconding a formal review request, but I'd rather that "seconding" imply that someone has looked over the library and found that it is in reasonably good shape for a formal review. Otherwise, we'll still have the problem Thomas W. is describing where a formal review is requested, several people "second" because they are interested, but the formal review won't happen because the library isn't ready. Doug

2 1

bccx and lambda (possible config problem)
by Russell Hind 28 Jan '04

28 Jan '04

Using 1.31-RC1, I have successfully made use of some of boost with Borland's bccx compiler. (I think John Maddock has done the work for this, so thanks). But I can't use the bind from Lambda with the new compiler with functions that return void. e.g. this won't compile (error is "lambda_functors.hpp": a void function may not return a value at line 159) #include <boost/function.hpp> #include <boost/lambda/lambda.hpp> #include <boost/lambda/bind.hpp> using namespace boost::lambda; void test(int a, int b) { } int main(int argc, char * argv[]) { boost::function<void (int a, int b)> f(bind(test, 10, _1)); return 0; } Has anyone any ideas if the config for bccx needs updating, or is this just a bug in the compiler and so we'll have to wait for an update to it? Thanks Russell

1 0

Re: Detection of wchar_t for Intel compiler
by Giovanni Bajo 28 Jan '04

28 Jan '04

Guillaume Melquiond wrote: > There was a recent change to the config/compiler/intel.hpp file in > order to change (one more time) the way the detection of wchar_t works. I did it, and discussed it on this very list. > So I've committed a trivial fix to CVS head. However it was a bit > annoying to note that a patch had been committed to such an important > place carefreely (otherwise the bug would have been found). This is my > first concern: the configuration of a compiler had been changed > without even being tested. You're implying way too much. There was a typo on my part, sure, but it wasn't detected under Windows. I *did* test it, and we did discuss the patch on the list. I never commit untested patches. > Second concern: since the patch was not tested on this > compiler/platform configuration, is it sure it behaves sanely? Please, review our discussion on the topic. > Sincethere was a lot of > changes in this area, shouldn't a specific test file be added in order > to finally ensure the correct detection of wchar_t? I can also > directly exercise some testcases if provided; I can test with ICC 7 > and 8 for Linux. We need to do a detection through the preprocessor. > Finally, my third concern: shouldn't such changes be committed to the > release branch since they directly involve the configuration of an > existing compiler? Again, please review the discussion. I told that I would commit it to the release branch after a few days, to not destabilize the branch. In fact, I was awaiting fallouts on different platforms, just like the one you fixed. If I had committed it to the release branch immediatly, I would have broken it as well. > Sorry for this long mail, but since release time draws near, I wanted > to point out this problem. Thank you for the fix. -- Giovanni Bajo

2 1

Re: [constant] New version + YANL- tests with MS 7.1
by David Abrahams 28 Jan '04

28 Jan '04

"Paul A. Bristow" <boost(a)hetp.u-net.com> writes: > However, the major snag is MS lack of __typeof__. Until this is solved, it is a > non-starter :-(( The problem isn't with the MS compiler, but with the library. Libraries that rely on nonstandard language extensions are non-starters at Boost. -- Dave Abrahams Boost Consulting www.boost-consulting.com

2 1

Re: class static constants, mpl, and linker problems
by David Abrahams 28 Jan '04

28 Jan '04

Matthias Troyer <troyer(a)itp.phys.ethz.ch> writes: > Dear Daniel, > > If you use BOOST_STATIC_CONSTANT you need to additionally provide a > definition on conforming compilers. The correct usage is like in: > > // declaration > class A { > BOOST_STATIC_CONSTANT(int , x=1); > }; > > .... > > //definition > #ifndef BOOST_NO_INCLASS_MEMBER_INITIALIZATION > int A::x; > #endif > > > I hope that clarifies this issue, Another approach is to use mpl::int_<n> or integral_c<T,n> which supply the definitions themselves. -- Dave Abrahams Boost Consulting www.boost-consulting.com

2 3

Re: class static constants, mpl, and linker problems
by Daniel Krügler 28 Jan '04

28 Jan '04

Hello Matthias, Matthias Troyer schrieb: > Dear Daniel, > > If you use BOOST_STATIC_CONSTANT you need to additionally provide a > definition on conforming compilers. The correct usage is like in: > > // declaration > class A { > BOOST_STATIC_CONSTANT(int , x=1); > }; > > .... > > //definition > #ifndef BOOST_NO_INCLASS_MEMBER_INITIALIZATION > int A::x; > #endif > > > I hope that clarifies this issue, No, it doesn't, please read the original thread carefully. I am aware of the fact, that an integral or enumeration type static member needs a corresponding definition and I hope my contribution has made that clear to everyone. Your answer implies, that users of the boost library are responsible for providing these definitions for boost constants. If so, I think that would be bad news. Consider those many hidden constants in whatever boost::details namespace or other implementation-dependent data items. How should the boost user correctly provide the necessary definitions? To my opinion boost should at least provide in the libs section source files with possibly necessary definitions of all externally accessible constants, which the user can translate if necessary. I also want to add, that although my example seems artifical, it should show a basic problem. Consider a simple "algorithm" like std::max which takes constant references as arguments. You cannot demand, that boost users must watch out for those or similar expressions in their code and take proper resolutions by hand?? Second, your response did not take into account the problems with the effects of the /Ze and /Za flags of the current VC7.1 compiler which I also think is a very important issue. Daniel

2 2