libtorrent

On Sun, 15 Jan 2006, Arvid Norberg wrote:

> > 2) if a single file torrent has space(s) in the file name, I will get
> > boost::filesystem::path: invalid name "xxx yyy zzzz" in path: "xxx yyy
> > zzzz". (Note, the 2 printed out filenames have no differences).
>
> Are you sure the name checking is disabled in boost.filesystem?
> see http://www.rasterbar.com/products/libtorrent/docs.html#filename-
> checks
>

Just an important update on this issue:

Run through VC.NET 7.0 debugger. Confirmed that for boost 1.33.1,
boost::filesystem::path_posix_windows.cpp::m_path_append() (line 340), the
checker(name) returns true. The following codes fail to disable the
default name checking:

int main(int argc, char **argv)
{
    using namespace libtorrent;

	namespace fs = boost::filesystem;
	fs::path::default_name_check(fs::no_check);

    try
    {
        session s;
        s.listen_on(std::make_pair(6881, 6889));

....

For now, let me just comment out the blocks inside
boost::filesystem::path_posix_windows.cpp, that should fix my problem.

Best regards,

Allen Zhao

On Jan 17, 2006, at 04:55, Allen Zhao wrote:

>
> On Mon, 16 Jan 2006, Arvid Norberg wrote:
>
>> There's no way to make this as a singleton and maintain the guarantee
>> that WSAStartup() is called before any socket-operation. Because if a
>> socket operation is performed from another singleton (static/global
>> variable) it may be called first. The reason why there is a reference
>> count now is to make sure it only gets called once. If the reference
>> counter > 0 then it's already initialized. And also to know when
>> WSAShutdown() can be called. i.e. when the last socket/address object
>> is destructed.
>
> I see. In theory, this should be ok, as long as it will balance out.
>
> Just run the simple client inside .NET debugger. (Not a very pleasant
> experience, as there are too many assert failures alert I have to  
> click
> "ignore" - anyway to turn them off? Guru anyone?).

Are you sure the asserts can be ignored? are they failing inside  
libtorrent?

> This is my observation, each address() contructor will increment the
> reference count, and each ~address() destructor will decrement the  
> count.
> Most of the counting are due to the address() object
> allocation/deallocation, scattered around several places, even in some
> lookup (are we sure that this is not a redundent copy constructor's  
> on the
> stack? Just asking, as I have not digged into boost yet, I should not
> comment on that yet.) The session called socket only twice.

That is correct. I don't know what you mean by redundant copy  
constructor though.

> Anyway, the reference count is not really balanced. The last time when
> WSACleanup() is called, the debugger reports memory overblown  
> error. Some
> memory problems, that is my guts' feeling at the moment.

When you say they're not really balanced, you mean that WSACleanup()  
is called more than once then? so the reference counter drops below 0?

Now when I think of it, maybe it's a threading problem. You could try  
adding a mutex around that reference counter. I'm currently replacing  
the socket code with asio anyway, so it might not be worth too much  
effort.

> Well, I have less grab on MS debugger, has anyone run this through
> BounceChecker or Purify? In term of memory debugger, I found Sun  
> Studio's
> dbx/SPARC is the best in the industry (just run through the bcheck  
> will
> reveal a lot of things).  (Sun Studio 11 is free nowaday.) Just  
> asking to
> see if someone else already done it or not. (I may get into this if  
> no one
> has done it yet.)

I've run it through BoundChecker, but that won't find buffer overruns  
very easily. I've also run it through valgrind on linux without any  
problems. valgrind will find buffer over/underrunds immediately.


--
Arvid Norberg

Hi,

The assert happens in dbgheap.c (I believe it is part of the MSVC, could
be from many source, boost or libtorrent, I am not sure) The first alert
is "Unhandled exception at xxxxx in test.exe: user breakpoint", if I
continue, I got "Debug Assertion Failed" (inside free.c -- again inside
MSVC).

Ignore my comment on the copy contrustor part, as this is no STL, and I do
not know much yet about boost.

On Tue, 17 Jan 2006, Arvid Norberg wrote:

> > memory problems, that is my guts' feeling at the moment.
>
> When you say they're not really balanced, you mean that WSACleanup()
> is called more than once then? so the reference counter drops below 0?

Actually, I should not say that is not balanced. Strictly speaking, the
observation is that the first time when WSACleanup is called, some memory
issue kicks in, and application got aborted. This is in debug build. In
release build, I will get hundreds lines of WSAStartup error as reported
before.

>
> Now when I think of it, maybe it's a threading problem. You could try
> adding a mutex around that reference counter. I'm currently replacing
> the socket code with asio anyway, so it might not be worth too much
> effort.

Some type of racing condition? Could be. Well, I have not studied the
codes yet, so just a wild guess.


> > Well, I have less grab on MS debugger, has anyone run this through
> > BounceChecker or Purify? In term of memory debugger, I found Sun
> > Studio's
> > dbx/SPARC is the best in the industry (just run through the bcheck
> > will
> > reveal a lot of things).  (Sun Studio 11 is free nowaday.) Just
> > asking to
> > see if someone else already done it or not. (I may get into this if
> > no one
> > has done it yet.)
>
> I've run it through BoundChecker, but that won't find buffer overruns
> very easily. I've also run it through valgrind on linux without any
> problems. valgrind will find buffer over/underrunds immediately.

OK, that is good. On Solaris/SPARC (not the x86) dbx or bcheck with check
access turned on, memory access error can be reported very precisely.
Anyway, I will save it for another day (after the client work on Win32, I
will move on to server side on both Solaris and Linux anyway.)

Best regards,

Allen Zhao

I made simple console application with VC7.1  like this.

#include "stdafx.h"
#include <libtorrent\session.hpp>
#include <utility>

int _tmain(int argc, _TCHAR* argv[])
{
   using namespace libtorrent;

	namespace fs = boost::filesystem;
	fs::path::default_name_check(fs::no_check);

   try
    {
        session s;
        s.listen_on(std::make_pair(6881, 6889));

        Sleep(100000);
    }
   catch (...)
   {}
   return 0;
}

This did exact as it is supposed to do. No assertion nor exception occurs. It starts 
listening on port 6881.

Maybe you should post your whole code that does not work.


boost 1.34 was used

MassaRoddel

Hi, there,

The code I am using is the simple_client in the example, I used the HEAD,
not the one from 0.9.1. It is almost the same as yours, except the int
main(....) part.

If I choose to break and debug, It looks like the stack trace is at the
some string deletion. Well, I did not do a systematic study, just some
random checking.

I am using boost 1.33.1. But maybe I will have to look into switching to
v7.1 now.

Best regards,

Allen Zhao

On Tue, 17 Jan 2006 MooPolice@... wrote:

> I made simple console application with VC7.1  like this.
>
> #include "stdafx.h"
> #include <libtorrent\session.hpp>
> #include <utility>
>
> int _tmain(int argc, _TCHAR* argv[])
> {
>    using namespace libtorrent;
>
> 	namespace fs = boost::filesystem;
> 	fs::path::default_name_check(fs::no_check);
>
>    try
>     {
>         session s;
>         s.listen_on(std::make_pair(6881, 6889));
>
>         Sleep(100000);
>     }
>    catch (...)
>    {}
>    return 0;
> }
>
> This did exact as it is supposed to do. No assertion nor exception occurs. It starts
> listening on port 6881.
>
> Maybe you should post your whole code that does not work.
>
>
> boost 1.34 was used
>
> MassaRoddel
>
>
>
> -------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
> for problems?  Stop!  Download the new AJAX search engine that makes
> searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
> http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642
> _______________________________________________
> Libtorrent-discuss mailing list
> Libtorrent-discuss@...
> https://lists.sourceforge.net/lists/listinfo/libtorrent-discuss
>

Hello, i have never had a problem getting libtorrent to compile before 
but now i do :(
I know i am missing something stupid but for the life of me i just can't 
seem to figure it out, any help would be much appreciated.
I'm getting these errors..


1>.\src\storage.cpp(115) : error C2065: 'not_directory_error' : 
undeclared identifier
1>.\src\storage.cpp(155) : error C2039: 'system_error_code' : is not a 
member of 'boost::filesystem::detail'
1>.\src\storage.cpp(155) : error C3861: 'system_error_code': identifier 
not found
1>.\src\storage.cpp(167) : error C2039: 'system_error_code' : is not a 
member of 'boost::filesystem::detail'
1>.\src\storage.cpp(167) : error C3861: 'system_error_code': identifier 
not found

It's ok sorted it, was a boost problem :o
> Hello, i have never had a problem getting libtorrent to compile before 
> but now i do :(
> I know i am missing something stupid but for the life of me i just 
> can't seem to figure it out, any help would be much appreciated.
> I'm getting these errors..
>
>
> 1>.\src\storage.cpp(115) : error C2065: 'not_directory_error' : 
> undeclared identifier
> 1>.\src\storage.cpp(155) : error C2039: 'system_error_code' : is not a 
> member of 'boost::filesystem::detail'
> 1>.\src\storage.cpp(155) : error C3861: 'system_error_code': 
> identifier not found
> 1>.\src\storage.cpp(167) : error C2039: 'system_error_code' : is not a 
> member of 'boost::filesystem::detail'
> 1>.\src\storage.cpp(167) : error C3861: 'system_error_code': 
> identifier not found