February 2002 - Boost-users

Boost.build question: Adding a new unit-test rule
by Helge Penne 04 Mar '02

04 Mar '02

Hi. I'm working on a project where some of the unit test are very inconvenient to implement with CppUnit and Mock classes or stubs alone, but would be very simple to implement by having them print messages to standard output, pipe this to a log file and diff this against a reference file. I added the following to our Jamrules file (most of it is stolen from the standard unit-test rule and run action): rule unit-test-diff { local files = [ exe $(<) : $(2) : $(3) : $(4) : EXE ] ; type-DEPENDS test : $(files) ; local file ; for file in $(files) { run-and-diff $(file) ; } } actions run-and-diff { $(SHELL_SET)PATH=$(RUN_PATH) $(SHELL_EXPORT)PATH $(<) > $(<:D=:S=.log) diff $(<:S=.log) $(<:S=.ref) # diff $(<:D=:S=.log) $(<:D=:S=.ref) } This runs the unit test program, pipes the output to <targetname>.log and diffs agains <targetname>.ref. There is only one problem: I cannot get Boost.build to locate these files in the correct folder. The Jamfile for the unit test is located in the same folder as the unit test source code. This is the only natural place for the ref-file as well. The action above has three different alternatives for the actual diff operation. The first uses the full path of the executable, but replaces the last part of the file name with .log and .ref. This places them in the binary folder of the unit test program, which isn't not very convenient. The second strips away the path, but since the currect folder always is the project root (not the folder where the Jamfile is), this places the log and ref files on the project root, which is almost as bad as the first alternative. I also experimented briefly with :G but this has no effect as the file path did not seem to contain grist. One possible solution might be to add a path feature to the target and set this to the reference file name, but I'm not quite sure how to implement this for rules that are not part of a tools.jam-file I really need the unit test rule to place these files in the folder where the rest of the test source code is. Does anyone know how to do this? - Helge Penne

3 2

RE: [Boost-Users] Linking an Archive
by Tom Matelich 27 Feb '02

27 Feb '02

The first getting_started1.so contains all the symbols from all of the objects linked into it. The second one only contains the symbols from getting_started.o and the symbols from libboost_python.a which satisfy unsatified symbols of getting_started1.o. Linking shared library is like linking an executable, it only takes the symbols from archives that it needs to satisfy itself. The first question is why you want libboost_python.a. If it is a valid reason, then you need to find a way to get all the code you need out of it into the shared lib. The first would be to use ar to get all the objects out, then take those and put them into the shared library. The second would be to keep adding unsatified symbols to getting_started.o until getting_started1.so has everything it needs. Tom > -----Original Message----- > From: doak87508 [mailto:jdoak@lanl.gov] > Sent: Wednesday, February 27, 2002 10:21 AM > To: Boost-Users(a)yahoogroups.com > Subject: [Boost-Users] Linking an Archive > > > I am playing around with using an archive to created a shared library > (.so file) instead of linking all the individual objects. I have > success when I link individual objects: > > % g++ -shared classes.o conversions.o errors.o extension_class.o > functions.o init_function.o module_builder.o objects.o types.o > cross_module.o getting_started1.o -o getting_started1.so > > % python > 1;31mh[1] >>> import getting_started1 > 1;31mh[1] >>> getting_started1.greet() > 'hello, world' > > Now, I create an archive: > > % ar r libboost_python.a classes.o conversions.o errors.o > extension_class.o functions.o init_function.o module_builder.o > objects.o types.o cross_module.o > > Then, I link that archive into the shared library: > > % g++ -shared libboost_python.a getting_started1.o -o > getting_started1.so > > % python > 1;31mh[1] >>> import getting_started1.so > Traceback (most recent call last): > File "<stdin>", line 1, in ? > ImportError: ./getting_started1.so: undefined symbol: > add__Q35boost6python19module_builder_basePQ45boost > 6python6detail8functionPCc > > I have also tried linking as follows with no change in result: > > % g++ -shared -L. -lboost_python getting_started1.o -o > getting_started1.so > > Does someone know how to get linking working with an archive file? > > JD > > > > Info: <http://www.boost.org> > Wiki: <http://www.crystalclearsoftware.com/cgi-bin/boost_wiki/wiki.pl> > Unsubscribe: <mailto:boost-users-unsubscribe@yahoogroups.com> > > > Your use of Yahoo! Groups is subject to > http://docs.yahoo.com/info/terms/ > >

2 1

in-place set_union/difference/etc ...
by ivan_dobrianov 27 Feb '02

27 Feb '02

Does anyone know of any existing standard tool [an insert iteartor or something equivalent] that can allow one to perform union, difference, etc on sets and maps in-place. E.g. instead of using set<MyClass> s1, s2, temp; set_union( s1.begin(), s1.end(), s2.begin(), s2.end(), inserter(temp, temp.begin())); temp.swap(s2); i'd like to do, say: copy(s1.begin(), s1.end(), my_unioner(s2, s2.begin())); which will do s2 = s1 + s2 in LINEAR time. It's pretty simple to write an inserter like that, but I am just wondering if there is one already ... or something similar. Thanks!

1 0

Bug in boost_1_27_0.tar.gz threads library
by Martin Brown 27 Feb '02

27 Feb '02

Hi There seems to be a missing '}' in libs/thread/src/once.cpp when compiling on gcc-3.04 on Redhat Linux 6.2. I added one and it seems to work. Not sure if it is in the right place. Context diff: *** once.cpp Thu Feb 7 15:33:15 2002 --- once.cpp.new Wed Feb 27 15:53:02 2002 *************** *** 55,62 **** --- 55,63 ---- { once_callback* cb = reinterpret_cast<once_callback*>(pthread_getspecific(key)); (**cb)(); } + } #elif defined(BOOST_HAS_MPTASKS) void *remote_call_proxy(void *pData) { std::pair<void (*)(), boost::once_flag *> &rData(*reinterpret_cast<std::pair<void (*)(), boost::once_flag *> *>(pData)); Thanks for some great libraries! Martin __________________________________________________ Do You Yahoo!? Everything you'll ever need on one web page from News and Sport to Email and Music Charts http://uk.my.yahoo.com

2 1

Linking an Archive
by doak87508 27 Feb '02

27 Feb '02

I am playing around with using an archive to created a shared library (.so file) instead of linking all the individual objects. I have success when I link individual objects: % g++ -shared classes.o conversions.o errors.o extension_class.o functions.o init_function.o module_builder.o objects.o types.o cross_module.o getting_started1.o -o getting_started1.so % python 1;31mh[1] >>> import getting_started1 1;31mh[1] >>> getting_started1.greet() 'hello, world' Now, I create an archive: % ar r libboost_python.a classes.o conversions.o errors.o extension_class.o functions.o init_function.o module_builder.o objects.o types.o cross_module.o Then, I link that archive into the shared library: % g++ -shared libboost_python.a getting_started1.o -o getting_started1.so % python 1;31mh[1] >>> import getting_started1.so Traceback (most recent call last): File "<stdin>", line 1, in ? ImportError: ./getting_started1.so: undefined symbol: add__Q35boost6python19module_builder_basePQ45boost 6python6detail8functionPCc I have also tried linking as follows with no change in result: % g++ -shared -L. -lboost_python getting_started1.o -o getting_started1.so Does someone know how to get linking working with an archive file? JD

1 0

autoconf+make
by josefsyk 26 Feb '02

26 Feb '02

I've been struggling to port parts of the boost libraries to Linux using gnu's automake, just to hit the wall every time. Anyone out there who has done this already? The files needed are: boost/smart_ptr.hpp boost/pending/ct_if.hpp boost/random.hpp boost/bind.hpp boost/function.hpp Would be nice to have the Makefile.am for BGL too. cheers.

2 1

boost::condition question
by svanechel 26 Feb '02

26 Feb '02

Hi, I'm trying to create a mechanism that synchronizes access to shared data on a per-operation basis. The idea is that some operations (E.g. read operations) can proceed concurrently, while others (write operations) must block all other threads. I thought boost::condition would do the trick. Since I do not have that much experience in multithreaded programming it is entirely possible that I do not fully understand boost::condition or its allowable usage. I tried to put operations in an "Operation Group". All operations belonging to the same group are allowed to execute concurrently, while all others are blocked (E.g. all read operations are in the same group). It is also possible that an operation doesn't specify an Operation Group, in which case a temporary group is assigned to the operation for each thread that calls it. Write operations are constructed this way. Below is an example of what I mean (OperationLevelLockable is defined below): class SomeClass : private OperationLevelLockable { public: int ReadData() { Lock lock( this, READ_OP_GROUP ); return ( mData ); } void WriteData( int data ) { Lock lock( this ); // assign temp group to operation, // so each write // operation blocks all other accesses mData = data; } private: int mData; }; I'm developing on Windows 2000 on a multiprocessor machine. This is the code I use to test my idea: const int32 OPERATION_GROUP_1 = 1; const int32 OPERATION_GROUP_2 = 2; const int NUM_THREADS = 4; class TestOperationLevelLockable : public MTTestCase { public: TestOperationLevelLockable() : mShared( this ), mCurrentGroup( 0 ), mThreadCount( 0 ) { // Create two sets of threads, each performing a // different operation // When the TestOperationLevelLockable object is // constructed all // registered threads are created an start running RegisterThread( this, &TestOperationLevelLockable::Operation1Thread, NUM_THREADS ); RegisterThread( this, &TestOperationLevelLockable::Operation2Thread, NUM_THREADS ); } void CallFromGroup( int32 opGroup ) { boost::mutex::scoped_lock lock( mMutex ); if ( 0 == mCurrentGroup ) { mCurrentGroup = opGroup; } if ( ( mCurrentGroup != opGroup ) && ( mThreadCount < NUM_THREADS ) ) { ReportFailure( "OperationLevelLockable failed to lock out operations of different group", __FILE__, __LINE__ ); } ++mThreadCount; } protected: class MTObject : public OperationLevelLockable { public: MTObject( TestOperationLevelLockable* target ) : mTarget( target ) {} void CallFromGroup( int32 opGroup ) { mTarget->CallFromGroup( opGroup ); } private: TestOperationLevelLockable* mTarget; }; MTObject mShared; int32 mCurrentGroup; int mThreadCount; boost::mutex mMutex; void DoMTTest() { } void Operation1Thread() { MTObject::Lock lock( mShared, OPERATION_GROUP_1 ); mShared.CallFromGroup( OPERATION_GROUP_1 ); SleepThread( 30 ); // seconds, to keep the thread busy } void Operation2Thread() { MTObject::Lock lock( mShared, OPERATION_GROUP_2 ); mShared.CallFromGroup( OPERATION_GROUP_2 ); SleepThread( 30 ); // seconds, to keep the thread busy } }; And here is OperationLevelLockable: class OperationLevelLockable { class AccessCounter; public: OperationLevelLockable(){} typedef Internal::DefaultTMIntType IntType; template<typename T> struct MakeSharedType { typedef T volatile Type; }; class Lock; friend class Lock; class Lock { public: Lock( OperationLevelLockable& obj ) : mCounter( obj.mOpAccessCounter ) { mCounter.EnterOperation( Lock::TempOpGroup() ); } Lock( OperationLevelLockable& obj, int32 opGroup ) : mCounter( obj.mOpAccessCounter ) { mCounter.EnterOperation( opGroup ); } ~Lock() { mCounter.LeaveOperation(); } private: Lock( Lock const& ); Lock& operator=( Lock const& ); static IntType TempOpGroup() { static volatile IntType opGroup = Internal::kFIRST_AUTO_GROUP; IntType newGroup = OperationLevelLockable::AtomicIncrement( opGroup ); if ( newGroup >= Internal::kLAST_AUTO_GROUP ) { OperationLevelLockable::AtomicAssign( opGroup, Internal::kFIRST_AUTO_GROUP ); } return ( newGroup ); } AccessCounter& mCounter; }; static inline IntType AtomicIncrement( volatile IntType& lval ); static inline IntType AtomicDecrement( volatile IntType& lval ); static inline void AtomicAssign( IntType& lval, IntType rVal ); static inline void AtomicAssign( volatile IntType& lval, IntType rVal ); private: class AccessCounter : private boost::base_from_member<boost::recursive_mutex> { public: AccessCounter() : mLock( member ), mRefCnt( 0 ), mOpGroup( Internal::kANY_GROUP ) {} bool IsOpGroupActive( IntType opGroup ) const { return ( ( Internal::kANY_GROUP == mOpGroup ) || ( opGroup == mOpGroup ) ); } void LeaveOperation() { if ( 0 == OperationLevelLockable::AtomicDecrement( mRefCnt ) ) { SetOpGroup( Internal::kANY_GROUP ); mCanOpProceed.notify_all(); } } void EnterOperation( IntType opGroup ) { mCanOpProceed.wait( mLock, CanOpProceed( *this, opGroup ) ); SetOpGroup( opGroup ); OperationLevelLockable::AtomicIncrement( mRefCnt ); } private: AccessCounter( AccessCounter const& ); AccessCounter& operator=( AccessCounter const& ); struct CanOpProceed { CanOpProceed( AccessCounter const& counter, IntType opGroup ) : mOpGroup( opGroup ), mCounter( counter ) {} bool operator()() const { return ( mCounter.IsOpGroupActive( mOpGroup ) ); } private: CanOpProceed& operator=( CanOpProceed const& ); int32 mOpGroup; AccessCounter const& mCounter; }; void SetOpGroup( IntType opGroup ) { OperationLevelLockable::AtomicAssign( mOpGroup, opGroup ); } boost::recursive_mutex::scoped_lock mLock; boost::condition mCanOpProceed; volatile IntType mRefCnt; volatile IntType mOpGroup; }; OperationLevelLockable( OperationLevelLockable const& ); OperationLevelLockable& operator=( OperationLevelLockable const& ); AccessCounter mOpAccessCounter; }; The idea is that the first operation that is executed acquires the lock and blocks out all threads not belonging to the same Operation Group. The original blocking works fine. Only threads executing in the same Operation Group are allowed to continue. When the last thread of the group leaves the operation in OperationLevelLockable::AccessCounter::LeaveOperation it releases all waiting threads and a thread from the other group is allowed to run. However when this thread finishes no new threads are allowed to run anymore and the application hangs. Am I using boost::condition incorrectly. Should I use something else? It the test code wrong? Are there any other errors in the code that prevent it from behaving as expected. Should I use another technique to achieve the same? I hope there is someone out there who has the hart to read through the entire posting and help me. I hope I have given enough information. Thx. Sven Van Echelpoel.

2 4

Building regex with msvc6sp5 and stlport
by John Maddock 25 Feb '02

25 Feb '02

>I have succesfully built the regex library >under msvc6sp5, using the vc6-sltport.mak file. > >But I notice that the compiler gives me warnings >relating to the microsoft stl include directories: > >example: vc98\include\iterator: blah blah blah Then you aren't compiling / linking to STLport >Should it not be linking to the stlport include files? Yes. Did you either * update your INCLUDE and LIB environment variables to point to STLport. or * invoke the makefile using something like: nmake CXXFLAGS="-Ic:\STLport-4.5.1\stlport" -fvc6-stlport.mak - John Maddock http://ourworld.compuserve.com/homepages/john_maddock/

2 1

Thread support?
by John Maddock 24 Feb '02

24 Feb '02

>If I'm using Boost in a single threaded app, do I need to define >BOOST_DISABLE_THREADS to avoid Boost using locks, etc. internally? You don't need to, but it may remove some redundent code, if your app is single threaded *and* links to a multithreaded runtime. - John Maddock http://ourworld.compuserve.com/homepages/john_maddock/

4 4

Problem linking with regex on Win32
by John Maddock 23 Feb '02

23 Feb '02

>Run-time Library: Multithreaded DLL/Debug Multithreaded DLL > >I've got a couple of problems trying to get my app to link against regex. > >I've tried linking to the debug and release versions of both the DLL and >non-DLL versions of the regex library and in each case I get errors about >unresolved external symbols for w32_regex_traits. > >w32_regex_traits.obj has been produced by Jam. > >In addition the static version winges about lots of locally defined symbols. > >I've specified BOOST_REGEX_NO_LIB as the generated filenames are not what >Jam produces. There could be a number of problems: Where the libs compiled with STLport (even a small error in the STLport include path could have messed this up)? The debug STLport libs are probably compiled by jam with __STL_DEBUG defined, your source will need to set this also. If you want to check your jam compiled libs, get a dump of the symbols they contain - if they're using the Dinkumware std lib then you'll see them referencing std::basic_string, release STLport will reference _STL::basic_string, and debug STLport will reference _STLD::basic_string. Failing that just create a static lib project in your IDE and add the regex sources to that with the build setting you want (this is probably the quickest solution IMO). Finally define BOOST_REGEX_STATIC_LINK to suppress the import/export warnings - John Maddock http://ourworld.compuserve.com/homepages/john_maddock/

1 0