- Boost - lists.boost.org

[multi-index] Changing hash_index_node.hpp to double linked list
by brad higgins 29 Apr '09

29 Apr '09

I am using boost multi-index, with a hashed-non-unique lookup. In my data, I am expecting many hash collisions. When I erase the multi- index, I find very poor performance. On investigation, I found that the hash collision data structure is a single linked list, and that removing an element involves traversing the entire list. Removing all of the elements involves traversing the list many times. Since my data has many collisions, I am traversing a very long list very many times. I know that the hash table is not designed for use with many collisions, but it is still faster in my testing on insert and serialization than the ordered_non_unique index specifier. I may end up using ordered_non_unique to avoid the worst-case erase() performance, and it may be good enough for the insert() performance. However, I would like to investigate making the hash collisions double- linked. I tried editing boost/multi_index/detail/hash_index_node.hpp, to make it a double-linked list, but I am seeing one of the included tests fail (test_safe_mode). I can't spot any bugs in my hash_index_node.hpp changes. What, apart from hash_index_node.hpp, needs to be updated, to support a double- linked list? boost/multi_index/detail/bucket_array.hpp? Others? Thanks, Brad

3 3

[encrypted strings]
by Sid Sacek 29 Apr '09

29 Apr '09

Does boost have any compile-time classes for string encryption? Is it even possible? When a hacker dumps an executable, they can see all of the strings the program might use, and some of those strings may contain sensitive information. Does boost have any classes that can encode the strings at compile-time? Ideally, the third string in the code below would never compile the "secret" string into the final binary. Regards, -Sid void foo( void ) { char const * s1 = "abc"; string s2 = "xyz"; encrypted_string s3( "secret" ); }

11 28

Re: [boost] [multi-index] Changing hash_index_node.hpp to double linked list
by JOAQUIN M. LOPEZ MUÑOZ 28 Apr '09

28 Apr '09

________________________________________ De: boost-bounces(a)lists.boost.org [boost-bounces(a)lists.boost.org] En nombre de brad higgins [bhiggins(a)arbor.net] Enviado el: martes, 28 de abril de 2009 21:56 Para: boost(a)lists.boost.org Asunto: Re: [boost] [multi-index] Changing hash_index_node.hpp to double linked list > Hi Joaquín, > Thanks for the quick response. I have attached the files I edited. > > Thank you very much for the help, Please allow a day or two till I find time to delve into this. I'll get back to you. Joaquín M López Muñoz Telefónica, Investigación y Desarrollo

1 0

[interprocess] patch for AIX
by Etienne PIERRE 28 Apr '09

28 Apr '09

Hello, Currently interprocess doesn't work on AIX for two reasons : - detail/os_thread_functions.hpp uses ::pthread_equal but this fails because pthread_equal is a macro under AIX and it fails because of the :: prefix - in detail/atomic.hpp there is no implementation of PowerPC under AIX so I added one that uses the combination of lwarx and stwcx for each function to implement. I saw that under AIX, the detail/atomic_count.hpp and detail/sp_counted_base.hpp falls back to a pthread implementation, I could submit a patch for a real atomic implementation under AIX but I wonder if there are other places in boost where this should be done ? Thanks, Etienne PIERRE

2 2

[Boost.any] Nightly trunk testing failures
by K. Noel Belcourt 28 Apr '09

28 Apr '09

Hi, Most nightly trunk testing results are failing with this error message. Have their been recent changes to Boost.any that could have caused this? # Making "/var/scratch/boost/results" directory... # Starting tests ("/var/scratch/boost/tools_bjam/bin.linuxx86/bjam" "- sBOOST_BUILD_PATH=/var/scratch/boost:/var/scratch/boost/tools_bb" "- sBOOST_ROOT=/var/scratch/boost/boost" "--boost=/var/scratch/boost/ boost" "--boost-build=/var/scratch/boost/tools_bb" "--debug- configuration" -l300 intel-10.0 -d2 preserve-test-targets=off --dump- tests -j8 link=static "--build-dir=/var/scratch/boost/results" >>"/ var/scratch/boost/results/bjam.log" 2>&1)... # Running Boost.Build tests Traceback (most recent call last): File "run.py", line 60, in <module> runner(root) File "/var/scratch/boost/tools_regression_src/regression.py", line 193, in __init__ self.main() File "/var/scratch/boost/tools_regression_src/regression.py", line 551, in main getattr(self,action_m)() File "/var/scratch/boost/tools_regression_src/regression.py", line 516, in command_regression self.command_test() File "/var/scratch/boost/tools_regression_src/regression.py", line 317, in command_test self.command_test_boost_build() File "/var/scratch/boost/tools_regression_src/regression.py", line 354, in command_test_boost_build "boost/bin.v2/libs/any/test/any_test.test")); OSError: [Errno 2] No such file or directory: '/var/scratch/boost/ results/boost/bin.v2/libs/any/test/any_test.test' We get the exact same error message on all Sandia testers, and from the looks of it, most other testers are having the same problem. Any help in getting testing going again would be appreciated. Thanks. -- Noel

3 4

mpl::print and the sun compiler
by Dmitry Goncharov 28 Apr '09

28 Apr '09

The sun c++ compiler doesn't issue a warning about comparison between signed and unsigned integer expressions. The attached patch makes mpl::print produce "implicit int is not supported" warning. With the patch applied the compiler output looks like this: $ CC +w -c print.cpp "/home/dgoncharov/boost/trunk/boost/mpl/print.hpp", line 55: Warning: Implicit int is not supported in C++. "print.cpp", line 3: Where: While specializing "boost::mpl::print<int>". "print.cpp", line 3: Where: Specialized in non-template code. 1 Warning(s) detected. Br, Dmitry

1 0

Re: [boost] [encrypted strings]
by Kasra 28 Apr '09

28 Apr '09

Sid Sacek Wrote: >This sounds very cool. Let me know when you're done. I would definitely >use it. For your needs the library is fully implemented, you don't need to look into RC4 implementation since the library already has many algorithms. On your encrypted string problem, it is best to user during installation techniques. Create an encrypted archive where the user has to provide a product key for the installation to commence. Read decrypted contents from the archive, encrypt it using any encryption algorithm you like however, use a PRNG to generate the key. Seed the PRNG from the available hardware and host OS and etc. Then each time you want the plaintext you regenerate your PRNG then decrypt your ciphertext. -- Kasra

2 1

Re: [boost] [asio] send blocks with MSG_DONTWAIT (linux)
by Stefan Pledl 28 Apr '09

28 Apr '09

> You can switch your socket in the non-blocking mode with > basic_stream_socket::non_blocking_io(). > Though asio should handle MSG_DONTWAIT as well. > > BR, Dmitry Hi Dimitry, that's right, but i need a blocking read and a non-blocking wirte. With basic_stream_socket::non_blocking_io() both, read and write are non-blocking. When i use C-functions read is blocking and write is non-blocking (when passing MSG_DONTWAIT). Best regards Stefan

1 0

Help with VMS port of CMake
by Bill Hoffman 27 Apr '09

27 Apr '09

Hi, I am trying to port CMake to VMS. For the first pass I am going to use the GNV tools and gmake. I am very close, but I am having trouble with files that have more than one "." in the name. CMake uses files like that in several places, as do many CMake based projects. Assuming I use POSIX paths on VMS, does anyone on this list know what magic the GNV cat program uses to handle the following: This works: cat /idisk1/devel/Windows-cl.cmake.in I wrote a simple C++ program that uses fstream to open argv[1], and get the following results: These work: ./test. Windows-cl.cmake.in ./test. Windows-cl^.cmake.in ./test. idisk1:[devel]Windows-cl^.cmake.in This does not: ./test. /idisk1/devel/Windows-cl.cmake.in Any help would be appreciated. Sorry if this is not very boost related, but in previous CMake discussions on the list the topic of VMS came up, so I thought there might be someone that could help me out. Thanks. -Bill

1 0

[reference_wrapper] feature request
by Thorsten Ottosen 27 Apr '09

27 Apr '09

Hi, It came up again in the RangeEx review that we really need reference_wrapper<T> to implement a forwarding operator(). Personally I don't care much for more than one or two arguments, but I guess others will. Thanks -Thorsten

2 1