- Boost - lists.boost.org

Problems compiling Boost.Test example with gcc under Linux
by Noel Llopis 24 Dec '04

24 Dec '04

I suspect the answer to this is going to be really simple, but I've been trying to figure it out for several hours and I'm ready to give up (I couldn't even find a mention of the problem in the mailing list archives). All I want to do is use the Boost Unit Test library with gcc under Linux. I have Boost 1.32 correctly installed (I've used other libraries successfully without any problems). Now I simply try to compile the simplest possible example: http://www.boost.org/libs/test/example/unit_test_example1.cpp This is what I get: gcc -o test boost_test.cpp boost_test.cpp:11: error: `boost::unit_test' has not been declared boost_test.cpp:11: error: expected nested-name-specifier before "test_suite" boost_test.cpp:11: error: `test_suite' has not been declared boost_test.cpp:23: error: expected constructor, destructor, or type conversion before '*' token boost_test.cpp:23: error: expected `,' or `;' before '*' token boost::unit_test has not been declared??? How is that possible? The compiler is correctly including the boost/test/unit_test.hpp file from the first line of the program. What am I missing? It must be something really, really stupid. I can tell :-) Happy holidays everyone! --Noel Games from Within http://www.gamesfromwithin.com

3 3

filesystem lib patch
by François Dumont 24 Dec '04

24 Dec '04

Find attached a patch for the filesystem library. remove and rename symbols are injected within the std namespace before being defined by the inclusion of cstdio header. The patch make the code less Standard library implementation dependent. Francois Dumont

2 1

Some observation re libraries as DLLS
by Robert Ramey 24 Dec '04

24 Dec '04

I'm making adjustments in the build and test Jamfiles to take into account changes made to permit the serialization library to be built and used as a DLL. This as raised a couple of questions in my mind. Re: the Jamfile for test a) It would seem to me that if the test is to be built and run with <runtime-link>dynamic - indicating the usage of C++ runtime dll, that the default build for the test should be to use the dll versions of all the libraries. I've managed to do this in sort of a kludgy way and it might be working. But I'm still working on it. b) the above raises the issue of how one should specifiy a target be built to use the dll vs the static library. <runtime-link>dynamic does something else. I was looking for a variant but I don't find it. I'm thinking that the way its done in the Program Options library is the way to go. c) when using the dll versions of boost libraries, the dll has to be on the path (at least with windows). It seems that bjam takes care of this automatically. So its only a problem when I want to debug the app with my VC IDE. It might be a good idea if the document mentioned this. d) The section on Testing Auto-linking in .../more\separate_compilation.html leaves me extremely confused. If I try to build the test without auto-linking by including the BOOST_???_NO_LIB then the libraries names searched for don't match the onces created by the build Jamfile. That its the above define doesn't get propogated to the build Jamfile and I don't know how it would. If I alter my Jamfile so that it doesn't include common-variant-tag as a target. The when the bjam stage is invoked, the libraries won't have the version specfic stuff. The situation is made even more compilicated by the fact that I have one dll (boost_wserialization.dll) which calls into another one (boost_serialization.dll). I've made it work by using the variant tags all the time - that is never using BOOST_???_NO_LIB. But now I'm concerned the the section at the end of the reference which seems to suggest that this won't work on platforms which don't suppor t auto-link. This seemed very simple when I started, but now I'm trying to get the details right and I'm having lots of nagging issues like the above. Its actually harder to figure out than the serialization library itself. Robert Ramey

1 0

mpl::inherit_linearly reuse to produce loki's tuple?
by Larry Evans 24 Dec '04

24 Dec '04

Could inherit_linearly be used to create something like loki's tuple? The name, inherit_linearly, and a very brief look at the code suggests it does something like loki's GenScatterHeirarchy, which was used to create the loki tuples as shown in: http://cvs.sourceforge.net/viewcvs.py/loki-lib/loki/Reference/HierarchyGene… If so, then I was wondering if it would be worth it to reuse inherit_linearly to implement boost's tuple?

3 14

typeof: what to register?
by Arkadiy Vertleyb 23 Dec '04

23 Dec '04

Hi all, Here is a question: besides primitive types, which types/templates should be registered as a part of typeof library? The first candidate is STL. The containers, functors, binders, etc. The STL iterators is a separate question. In most STL implementations they are typedefs, and therefore their template names are non-standard and different. In Dinkumware they are nested classes, and so are the subject to non-deduced context, which prevents them from being registered in a trivial way. Peder does have a solution for this, based on a bug in VC7.1 that allows to work-around non-deduced context (I am personally against using this, although I have to admit that it does the job). So, if we decide to put #ifdefs, based on STL implementation, and *if* we decide to use Peder's solution, it looks like we can take care of all the standard library (including iterators) on the typeof lib level. I have to note here that, if we decide against iterator registration, iterator instantiations can be always registered as needed, by the end user: BOOST_TYPEOF_REGISTER_TYPE(MyClass) BOOST_TYPEOF_REGISTER_TYPE(std::vector<MyClass>::iterator) I think if other libraries want to be supported by typeof (like Spirit, for example), those libraries should register their types/templates themselves (just because of synchronisation issue). Any help would be provided, of course. I would greatly appreciate any opinions on the subject. Regards, Arkadiy

2 2

boost::noncopyable and DLL usage
by Roland Schwarz 23 Dec '04

23 Dec '04

The thread class derives from boost::noncopyable. While this works for static library builds, there are some ugly warnings when compiling user programs that use the DLL version of the threading lib. I suspect noncopyable not having the appropriate __declspec(export) in a DLL being the culprit. I suspect any library will show this behaviour when doing a DLL build. Is this effect known? Are the warnings avoidable? Thank you, Roland

2 5

Status of VC++.NET 2005 Beta
by Loïc Joly 23 Dec '04

23 Dec '04

Hello all, I would like to know if anybody has successfully built boost with VC++.NET 2005 beta version ? I tried to look the regression tests, but the links from boost web site are broken (This site is defaced!!! NeverEverNoSanity WebWorm generation 25.). By the way, is there any document that explains how to create a new toolset for boost::bjam ? Regards, -- Loïc

3 3

Re: [boost] [gui] rectangles and the CSS box model
by brangdon＠cix.compulink.co.uk 23 Dec '04

23 Dec '04

In-Reply-To: <cqbhda$lq7$1(a)sea.gmane.org> msclrhd(a)hotmail.com (Reece Dunn) wrote (abridged): > However, for efficiency, it is a good idea to expose the > members of point and size. Is that because your compiler cannot inline member functions? -- Dave Harris, Nottingham, UK

1 0

X-Platforming Boost GUI
by Alan Gutierrez 22 Dec '04

22 Dec '04

I've spent quite a bit of time getting this Boost GUI code to build anywhere, but now it builds on both OS X and W2K. I'm using XCode to build on OS X, Jamboost is working on W2K. I'm feeling like I might as well add Linux to the mix, and work toward having an early integeration across these three platforms. It would teach me a lot about Boost and bjam. -- Alan Gutierrez - alan(a)engrm.com

3 3

[gui] rectangles and the CSS box model
by Reece Dunn 22 Dec '04

22 Dec '04

Hi All, In the "thoughts on a GUI component library" thread there have been various discussions about the layout of a rectangle and come relating to CSS. Note that point and size are stable. The CSS3 Values and Units module (http://www.w3.org/TR/2001/WD-css3-values-20010713/) defines various units (e.g. inches, picas and em). This is a good basis for a GUI library metric type. The type should store the values as a floating point to allow things like 2.71cm and also to support platforms that store coordinates as floating points (e.g. Cocoa). The CSS3 relative values (em and ex) are dependant on the inherited font-size property, so it is difficult to use these units as there is no associated font-size property. Another issue relating to this is how to resolve metric values between units that are not directly convertable. I have suggested using a device::resolve method that takes metric, point, size and rect types, but it may not always be possible to get access to a device. A further complication is the location of the origin. If we treat the origin as being in the top left corner (which is true for Windows, HTML/CSS and other document models such as text editors) this will simplify the design and for platforms that have a different model (e.g. Mac) then the implementation on that platform will need to map between the different representations. I personally like the CSS box model (http://www.w3.org/TR/css3-box/) which can be defined as: struct css::box_model { rect margin; rect border; rect padding; size dimensions; // width and height // T content; }; The problem with supporting it by default is that we would need: class component: public css::box_model; class graphics::object: public css::box_model; This clearly violates the don't pay for what you don't need rule of C++, so for a css_block_layout, it needs to contain the box model properties and a pointer to the component/graphical object. Obviously some level of support for the CSS box model should be provided, allowing an implementation of it to be written. There has been some discussion and disagreement as to what representation a rectangle takes. It is a good idea to hide the representation of a rectangle and provide accessors to the various values. This is because different people will want different things from it. However, for efficiency, it is a good idea to expose the members of point and size. In supporting the CSS box model and adoption of top-left as (0, 0) the representation becomes: class rect { point tl; point br; }; with: top() = tl.y; left() = tl.x; bottom() = br.y; right() = br.x; width() = br.x - tl.x; width( w ): br.x = tl.x + w; size() = ::size( width(), height()) center() = tl + ( size() / 2 ) An issue with this is: do we keep the width of the rectangle when setting left/right? In order to keep things simple, the answer has to be no. Construction of a rect should have the following variants: rect() = rect( 0, 0, 0, 0 ) rect( n ) = rect( n, n, n, n ) // This models the CSS "margin: 2px;" behaviour. rect( t, r, b, l ): tl( t, l ), br( b, r ) // This models "margin: 1pt 2pt 3pt 4pt;" rect( point p, size s ) = rect( p.x, p.y, p.x + s.dx, p.y + s.dy ) // This allows for support for traditional GUI representations Note that this model will change the nature of a program that uses the current model. We also have inflate/deflate operations that should take size *and* rect variants. This will allow easier box-model calculations: rect disp( point( 0, 0 ), size( 500, 700 )); rect margin( 1, 2, 3, 4 ); rect box = disp; box.deflate( margin ); // apply margin adjustments // ... Regards, Reece

3 8