- Boost-users - lists.preview.boost.org

Environment variable library?
by Aaron W. LaFramboise 17 Jan '05

17 Jan '05

Is there a Boost library that supports manipulating process environment variables? I realize that accessing environment variables in the normal way is typically trivial, but it would be nice to have the entire thing wrapped up in a system-independent interface that is natural for C++. Aaron W. LaFramboise

4 4

shared_from_this and custom deleters
by Niklas Wiberg 17 Jan '05

17 Jan '05

Hi, I have the following situation: In the constructor body of class A, I need to call functions that take a shared_ptr<A> as a parameter. These functions do not store the shared_ptr<A>, but in some situations a weak_ptr<A>. There is not yet any shared_ptr owning the A instance, since this is during A construction... At first, I thought using a shared_ptr<A> tmpThis(this, NullDeleter()) was a good idea. But when I later on (when another shared_ptr actually owns the A instance) want to use shared_from_this on A, I get an exception since the use_count of the internal weak_ptr is based on the first temporary shared_ptr that I used during construction (with the NullDeleter). Am I correct in my assumption that the internal weak_ptr in enable_shared_from_this is initialized by the first shared_ptr and that will prevent any later use of shared_from_this when I have another shared_ptr (with a default deleter) owning the element? I guess that a static create function would solve the problem, but changing that would be quite a big undertaking in this particular case so I thought I'd check if someone knows a simple solution to this... I added a small example code snipped below... Best regards, Niklas Example: void foo(shared_ptr<A> a); void otherFunction(shared_ptr<A> a); class A : public enable_shared_from_this<A> { A() { // Can't use shared_from_this since no shared_ptr owns this yet shared_ptr<A> thisTmp(this, NullDeleter()); foo(thisTmp); } bar() { otherFunction(shared_from_this()); // Throws exception... } }; int main() { shared_ptr<A> a(new A()); a->otherFunction(); ... }

2 1

How to pass a member variable with boost::ref() in boost::bind()?
by tom gee 17 Jan '05

17 Jan '05

Here is the work around: class A { ... private: typedef struct _ONE_RECORD_ { string sGUID; string sID; string sName; string sSupport; string sModify; } ONE_RECORD; vector<ONE_RECORD> m_v; private: void FillOneLine(const ONE_RECORD& one_line); void FillAllLines(); }; void A::FillAllLines() { ... // method 1 for_each(m_v.begin(), m_v.end(), boost::bind(&A::FillOneLine, this,_1)); // OK // method 2 for_each(m_v.begin(), m_v.end(), boost::bind(&A::FillOneLine, this, ref(_1))); // error ... } Since ONE_RECORD can be large, I prefer to calling bind() as method 2, but, compiled under VC7.1 it generates such an error: e:\Boost\include\boost-1_32\boost\bind.hpp(285): error C2664: 'R boost::_mfi::mf1<R,T,A1>::operator ()<A>(U & ,A1) const' : cannot convert parameter 2 from 'boost::arg<I>' to 'A::ONE_RECORD' with [ R=void, T=A, A1=A::ONE_RECORD, U=A * ] and [ I=1 ] Can any body help to explain why and give a solution? Thanks.

2 1

Smart pointers
by Kercso Jozsef 17 Jan '05

17 Jan '05

Hi! I would have several questions regarding to smart pointers. They are completely new for me, and I'm not sure if my design is the best. Feel free to comment it. :) The "safe" pointer problem: There is a class Data, whom instances are used by many other Processor objects. It can happen, that a Manager object deletes the Data object, so the Processors have invalid pointers. It must be made such a design, where the Processors can check the Data pointers before using it, if it is still valid or not. My solution: class Handler; class Data { boost::shared_ptr< Data > m_sharedPtr; //shared pointer to myself public: Data(); ~Data() {}; Handler getHandler(); int aMethod(); .... }; class Handler { boost::weak_ptr< CTrain > m_weakPtr; //weak pointer to the handled Data public: Handler(boost::shared_ptr< Data > data) bool aMethod(int& result); ... }; Data::Data() : m_sharedPtr(this) {} Data::~Data() {} Handler Data::getHandler() { return Handler(m_sharedPtr); } int Data::aMethod() { return 1; } Handler::Handler(boost::shared_ptr< Data > data) : m_weakPtr(train) {} bool Handler::aMethod(int& result) { bool retVal = !m_weakPtr.expired(); if (retVal) { boost::shared_ptr< Data > p = m_weakPtr.lock(); result = p->aMethod(); } return retVal; } When somebody wants to access a Data object, it gets only a Handler to it. The Handler can check with its weak pointer, if the corresponding shared pointer is expired (data object deleted) or not. Thus it can be assured, that no invalid pointer of the Data object will be used by the Processors. If the Manager object (who has a container of Data* ) deletes a data object, the corresponding Handler gives false back, because m_weakPtr.expired() returns false. So everything seems to be OK, but there are several problems and questions: 1. the destructor of the Data object will be called twice, when the Manager calls 'delete data;': - Once, when by the delete. - Once by the m_sharedPtr. After this the use_count() method returns -1 (!!!) and the m_weakPtr.expired() returns true (!!!). To eliminate this problem I initialize the m_sharedPtr in the Data constructor in the following way: m_sharedPtr(this, emptyDeleter), where emptyDeleter is an empty method: void emptyDeleter(Data* data) { } 2. What happens, if one of the Processors locks its m_weakPtr to obtain a shared pointer to acces the Data, but after obtaining the shared pointer the manager deletes the corresponding data object? If the Processor tries to acces again the (deleted) data object --> core dump. I have found the following solution to this problem: - before deletion the Manager somehow checks if data->m_sharedPtr.unique() returns true. If not, it refuses to delete the data object. 3. The Data object will be accessed many times by many Processors, and by every access the following construction will be used: if (!m_weakPtr.expired()) { boost::shared_ptr< Data > p = m_weakPtr.lock(); // do job on p ... } My question is: how "expensive" is such a construction? How fast is to obtain a shared pointer from a weak pointer unsing lock()? If you already met the problem of these "safe" pointers and have an other solution, please share it with us. Or if you have a comment to my solution, then say it too. :) Thanks, Jozsef Kercso

1 0

Boost with mingw creates .lib not .a
by John Arbash Meinel 17 Jan '05

17 Jan '05

I was wondering if there is a specific reason still that when using mingw to build the boost libraries it defaults to using the .lib extension? I was only able to find the following post: http://article.gmane.org/gmane.comp.lib.boost.devel/79594 Which is where David Abrahams states that file extensions are based on platform and not based on toolset. I've been using mingw lately, and at least current versions are able to find a library with a .lib extension without any problem. However, visual studio can not use that file as a library, and mingw will look for a .a first. What I'm thinking is that at least for the undecorated versions of the libraries (boost_python.lib) it should use the common convention for the compiler, that way you could have both in the same directory, and have the compiler pick the right one. I realize I can force this with a '-sSUFLIB=.a', I was just wondering if there was a reason the extension is based on platform, and not on TOOLS. John =:->

2 1

boost::filesystem::path question
by matthew.j.polder＠itt.com 16 Jan '05

16 Jan '05

Hi, I typically work on Windows and Unix. Our network has a bunch of shared drives that I can access from both Windows and Unix by using the form: //filesystem/some/path/ It looks like boost::filesystem is getting rid of the double slash at the beginning and replacing it with a single slash. This will work on Unix but WIndows needs the double slashes. Any way around this? thanks, matthew ************************************ This e-mail and any files transmitted with it are proprietary and intended solely for the use of the individual or entity to whom they are addressed. If you have received this e-mail in error please notify the sender. Please note that any views or opinions presented in this e-mail are solely those of the author and do not necessarily represent those of ITT Industries, Inc. The recipient should check this e-mail and any attachments for the presence of viruses. ITT Industries accepts no liability for any damage caused by any virus transmitted by this e-mail. ************************************

5 8

boost 1.32 with gcc 3.3 for darwin
by Becky Bendick 15 Jan '05

15 Jan '05

To anyone who is interested: With the assistance of a huge number of very useful people, I've sorted out how to successfully install the boost 1.32 package on Mac Darwin with gcc 3.3. There are two problems to solve: 1. some libraries are made but not usable because of errors like: /sw/include/boost/numeric/ublas/vector_expression.hpp:1837: error: due > to a defect in the G++ 3.2 ABI, G++ has assigned the same mangled > name to > twodifferent types In file included from > /sw/include/boost/numeric/ublas/matrix.hpp:23, > from conftest.cc:100: > /sw/include/boost/numeric/ublas/matrix_expression.hpp:5156: error: > due to a > defect in the G++ 3.2 ABI, G++ has assigned the same mangled name > to two > different types > configure:23533: $? = 1 > configure: failed program was: > | /* confdefs.h. */ The solution is to put -fabi-version=0 into your CFLAGS and CXXFLAGS, rebuild bjam, and install boost with the rebuilt bjam. 2. Install hangs on xml_grammar. The solution is outlined in http://lists.boost.org/MailArchives/boost/msg76758.php Or include the following: <darwin><release><inline>off Thanks to everyone who helped out. ------------------------------------------------------------------------ ---------------------- Rebecca Bendick, Ph.D. Bullard Laboratories University of Cambridge Madingley Road CAMBRIDGE CB3 0EZ United Kingdom bendick(a)esc.cam.ac.uk

1 0

boost1.32
by Becky Bendick 14 Jan '05

14 Jan '05

I need the boost libraries in order to compile a new application. I installed the boost1.31 package from the fink release, using fink install boost 1.31 without any trouble, but when I then tried to configure the dependent application, it found that the boost/numeric/ublas/matrix.hpp library was unusable. On the advice of the boost-users list, I then installed boost 1.32, using the prebuilt executable for Darwin available on the boost web site. The install hung several times creating the serialization libraries, so I installed using the tag --without-serialization, and completed the install successfully that way. However, I still get an error when I configure the dependent package, saying that the matrix.hpp library is unusable. The config.log says: checking for boost/numeric/ublas/matrix.hpp usability... configure:22782: g++ -c -I/sw/include -O2 -ansi -Wall -Wno-unused -I/sw/include -I/usr/local/include /boost-1_32 conftest.cc >&5 In file included from /usr/local/include/boost-1_32/boost/numeric/ublas/traits.hpp:27, from /usr/local/include/boost-1_32/boost/numeric/ublas/storage.hpp:29, from /usr/local/include/boost-1_32/boost/numeric/ublas/matrix.hpp:21, from configure:22826: /usr/local/include/boost-1_32/boost/numeric/ublas/ returntype_deduction.hpp:53: warning: use of `long double' type; its size may change in a future release /usr/local/include/boost-1_32/boost/numeric/ublas/ returntype_deduction.hpp:53: warning: (Long double usage is reported only once for each file. /usr/local/include/boost-1_32/boost/numeric/ublas/ returntype_deduction.hpp:53: warning: To disable this warning, use -Wno-long-double.) In file included from /usr/local/include/boost-1_32/boost/numeric/ublas/vector.hpp:24, from /usr/local/include/boost-1_32/boost/numeric/ublas/matrix.hpp:22, from configure:22826: /usr/local/include/boost-1_32/boost/numeric/ublas/ vector_expression.hpp:1924: error: due to a defect in the G++ 3.2 ABI, G++ has assigned the same mangled name to two different types In file included from /usr/local/include/boost-1_32/boost/numeric/ublas/matrix.hpp:23, from configure:22826: /usr/local/include/boost-1_32/boost/numeric/ublas/ matrix_expression.hpp:5201: error: due to a defect in the G++ 3.2 ABI, G++ has assigned the same mangled name to two different types configure:22785: $? = 1 configure: failed program was: | #line 22760 "configure" | /* confdefs.h. */ Higher in the config.log is another error (unrelated?) that doesn't seem to cause any trouble: gcc -E -I/sw/include conftest.c configure:15096:28: ac_nonexistent.h: No such file or directory configure:15085: $? = 1 configure: failed program was: | #line 15070 "configure" | /* confdefs.h. */ Here's the info on my gcc. What is wrong with it ? And how can I fix it? gcc -v </dev/null >&5 Reading specs from /usr/libexec/gcc/darwin/ppc/3.3/specs Thread model: posix gcc version 3.3 20030304 (Apple Computer, Inc. build 1671) Thanks! Becky ------------------------------------------------------------------------ ---------------------- Rebecca Bendick, Ph.D. Bullard Laboratories University of Cambridge Madingley Road CAMBRIDGE CB3 0EZ United Kingdom bendick(a)esc.cam.ac.uk

2 2

xml_grammar
by Becky Bendick 14 Jan '05

14 Jan '05

I'm trying to install boost1.32 using the prebuilt bjam for mac darwin, but it seems to hang on darwin-C++-action bin/boost/libs/serialization/build/libboost_serialization.a/darwin/ release/xml_grammar.o or perhaps this just takes hours, which seems unlikely. What am I doing wrong, and how do I fix it? Thanks, Becky ------------------------------------------------------------------------ ---------------------- Rebecca Bendick, Ph.D. Bullard Laboratories University of Cambridge Madingley Road CAMBRIDGE CB3 0EZ United Kingdom bendick(a)esc.cam.ac.uk

3 2

Multithreaded Signal Library
by Stephen torri 14 Jan '05

14 Jan '05

Since it has been advise by many people in the past that retrofitting software is troublesome and expense. I am looking at the issues of redesigning the signals library from scratch to keep the same interface. Before providing a new solution to consider I have a few questions. Is the relationship of signals to slots 1 to N where N = 0 to some value? From what I have read I believe this is true. Since signals are not copyable? N processes desiring to signal the M slots (N = 0 to some value N, M = 0 to some value M) will have to do it through the same signal object if the processes are of the same type (class has an signal object as a variable). Right? The second question comes from my initial understanding of the signals library. I was thinking about this scenario: There is a class call Cart which has a signal boost::signal<void()> in_Parking_Lot in it. Since a store will have multiple Carts for customers to use the owner has programmed his carts to signal the front desk computer when they are moved out of the store. class Cart { private: boost::signal<void()> in_Parking_Lot; public: void init (FrontDesk& fd) { in_Store = true; in_Parking_Lot.connect(fd); void process () { for (;;) { Check location (If in store set in_Store = true vice versa if not). if ( ! in_Store ) { in_Parking_Lot(); } } }; So if I have a very small shop with 50 Carts, all 50 Carts be using the same signal object to alert the front desk. If I had a big computer with 51 cpus (50 Carts and 1 FrontDesk) to run each process I have a race condition. Which process gets to alert the front desk when? >From what I understand the relationship of signals to slots is (1:N) where N is 0 to some number. So I have to have some structure in place so that I keep things safe (e.g. no race conditions or starvation) while maintaining throughput when the 50 processes try to send a signal at the same tiem. So given my example I was wondering what are the properties that I need to maintain? For example should each call to the signal's operator() function be blocking (each signaling process must update all the slots listening for the signal before returning) or non-blocking (some other process handles updating the slots while the signaling process returns back to work right away). The final question is that if indeed the relationship of signals to slots is 1:N then the multi-threading problems begins to surface more predominately. That is T threads, where T is a Cart object, use 1 signal to notify N slots. We have a restricted resource, the 1 signal, to update N slots at various times during the lifetime of the T threads. Is this the problem of using the signal library with multiple threads? I will leave things there for now and await your response. Stephen

1 0