[Rob Stewart]

Can you tell us what change can be made to the code to effect what had been gained by using that macro?

It's unconditionally: #define _FPOSOFF(fp) ((long long)(fp)) So you should be able to replace it with a static_cast, or eliminate it entirely. STL

[STL]

It's unconditionally: #define _FPOSOFF(fp) ((long long)(fp))

[Gareth Sylvester-Bradley]

These days I'm sure that's true, but formerly don't I remember on 32-bit platforms it could be something like this? typedef struct fpos_t { unsigned int lopart; int hipart; } fpos_t; #define _FPOSOFF(fp) ((long)(fp).lopart) How long in the tooth am I?

I've never seen that. VS 2012+ have the long long cast. In VS 2010 (and going back to 2005, IIRC), it was: #define _FPOSOFF(fp) ((long)(fp)) which was hostile to large files, which was why we changed it. STL

On 9/6/2016 2:49 PM, Daniel James wrote:

On 6 September 2016 at 19:24, Stephan T. Lavavej wrote:

...

[Rob Stewart]

...

Can you tell us what change can be made to the code to effect what had been gained by using that macro?

It's unconditionally:

#define _FPOSOFF(fp) ((long long)(fp))

So you should be able to replace it with a static_cast, or eliminate it entirely.

That might break older versions, or other uses of Dinkumware. I would try changing this configuration file so that it only applies to old versions:

https://github.com/boostorg/iostreams/blob/develop/include/boost/iostreams/d...

I don't have access to Visual C++ so I can't do it myself. If anyone wants to, all the relevant code is here:

https://github.com/boostorg/iostreams/blob/develop/include/boost/iostreams/p...

The best way for iostreams seems to be to create and use its own macro, equivalent to _FPOSOFF(fp), for older versions of VC++ but hardcodes as '((long long)(fp))' for VC++14 on up. This would allow Microsoft to remove it in future versions of VC++ without affecting iostreams. I assume Microsoft will not be retroactively removing for older versions of VC++ prior to VC++14, even if it is an implementation detail. If iostreams is not currently under CMT, if you add it as a CMT library I will make the appropriate change and test it out. This won't get in 1.62 but should be in the next release after that.

On 6 September 2016 at 21:32, Edward Diener wrote:

The best way for iostreams seems to be to create and use its own macro, equivalent to _FPOSOFF(fp), for older versions of VC++ but hardcodes as '((long long)(fp))' for VC++14 on up. This would allow Microsoft to remove it in future versions of VC++ without affecting iostreams.

IMO it would be better if all new compilers/libraries used the same implementation, and the special Dinkumware implementation used only for older compilers/librarires. Although that might not be possible, and there seem to be some functions that are only defined for Dinkumware, so they might need to be supported. Anyway, it's up to whoever does this.

Thanks :D Billy3 From: Edward Dienermailto:eldiener@tropicsoft.com Sent: Monday, September 19, 2016 5:19 AM To: boost@lists.boost.orgmailto:boost@lists.boost.org Subject: Re: [boost] Boost should not be saying "_FPOSOFF" On 9/2/2016 2:06 PM, Billy O'Neal (VC LIBS) wrote:

Hello, Boost folks. :)

I tried to remove this undocumented internal macro from our STL because we used it in only one place, but this broke Boost:

#define _FPOSOFF(fp) ((long long)(fp))

...patience... ...found 722 targets... ...updating 18 targets... common.mkdir build\boost\bin.v2\libs common.mkdir build\boost\bin.v2\libs\iostreams common.mkdir build\boost\bin.v2\libs\iostreams\build common.mkdir build\boost\bin.v2\libs\iostreams\build\msvc-99 common.mkdir build\boost\bin.v2\libs\iostreams\build\msvc-99\debug common.mkdir build\boost\bin.v2\libs\iostreams\build\msvc-99\debug\link-static common.mkdir build\boost\bin.v2\libs\iostreams\build\msvc-99\debug\link-static\threading-multi compile-c-c++ build\boost\bin.v2\libs\iostreams\build\msvc-99\debug\link-static\threading-multi\file_descriptor.obj file_descriptor.cpp d:\src\vc\wcfb01\qa\vc\FE\3rdpartycomp\boost_1_60_0\boost/iostreams/positioning.hpp(107): error C3861: '_FPOSOFF': identifier not found libs\iostreams\src\file_descriptor.cpp(265): warning C4244: 'argument': conversion from 'std::streamsize' to 'DWORD', possible loss of data libs\iostreams\src\file_descriptor.cpp(289): warning C4244: 'argument': conversion from 'std::streamsize' to 'DWORD', possible loss of data

cl /Zm800 -nologo @"build\boost\bin.v2\libs\iostreams\build\msvc-99\debug\link-static\threading-multi\file_descriptor.obj.rsp"

...failed compile-c-c++ build\boost\bin.v2\libs\iostreams\build\msvc-99\debug\link-static\threading-multi\file_descriptor.obj... compile-c-c++ build\boost\bin.v2\libs\iostreams\build\msvc-99\debug\link-static\threading-multi\mapped_file.obj mapped_file.cpp d:\src\vc\wcfb01\qa\vc\FE\3rdpartycomp\boost_1_60_0\boost/iostreams/positioning.hpp(107): error C3861: '_FPOSOFF': identifier not found libs\iostreams\src\mapped_file.cpp(59): warning C4244: 'return': conversion from 'const boost::iostreams::stream_offset' to '::size_t', possible loss of data libs\iostreams\src\mapped_file.cpp(325): warning C4244: 'argument': conversion from 'boost::iostreams::stream_offset' to 'SIZE_T', possible loss of data

cl /Zm800 -nologo @"build\boost\bin.v2\libs\iostreams\build\msvc-99\debug\link-static\threading-multi\mapped_file.obj.rsp"

...failed compile-c-c++ build\boost\bin.v2\libs\iostreams\build\msvc-99\debug\link-static\threading-multi\mapped_file.obj... ...skipped libboost_iostreams-vc-mt-gd-1_60.lib for lack of file_descriptor.obj... ...skipped libboost_iostreams-vc-mt-gd-1_60.lib for lack of libboost_iostreams-vc-mt-gd-1_60.lib... common.mkdir build\boost\bin.v2\libs\iostreams\build\msvc-99\release common.mkdir build\boost\bin.v2\libs\iostreams\build\msvc-99\release\link-static common.mkdir build\boost\bin.v2\libs\iostreams\build\msvc-99\release\link-static\threading-multi compile-c-c++ build\boost\bin.v2\libs\iostreams\build\msvc-99\release\link-static\threading-multi\file_descriptor.obj file_descriptor.cpp .\boost/iostreams/positioning.hpp(107): error C3861: '_FPOSOFF': identifier not found libs\iostreams\src\file_descriptor.cpp(265): warning C4244: 'argument': conversion from 'std::streamsize' to 'DWORD', possible loss of data libs\iostreams\src\file_descriptor.cpp(289): warning C4244: 'argument': conversion from 'std::streamsize' to 'DWORD', possible loss of data

cl /Zm800 -nologo @"build\boost\bin.v2\libs\iostreams\build\msvc-99\release\link-static\threading-multi\file_descriptor.obj.rsp"

...failed compile-c-c++ build\boost\bin.v2\libs\iostreams\build\msvc-99\release\link-static\threading-multi\file_descriptor.obj... compile-c-c++ build\boost\bin.v2\libs\iostreams\build\msvc-99\release\link-static\threading-multi\mapped_file.obj mapped_file.cpp .\boost/iostreams/positioning.hpp(107): error C3861: '_FPOSOFF': identifier not found libs\iostreams\src\mapped_file.cpp(59): warning C4244: 'return': conversion from 'const boost::iostreams::stream_offset' to '::size_t', possible loss of data libs\iostreams\src\mapped_file.cpp(325): warning C4244: 'argument': conversion from 'boost::iostreams::stream_offset' to 'SIZE_T', possible loss of data

cl /Zm800 -nologo @"build\boost\bin.v2\libs\iostreams\build\msvc-99\release\link-static\threading-multi\mapped_file.obj.rsp"

...failed compile-c-c++ build\boost\bin.v2\libs\iostreams\build\msvc-99\release\link-static\threading-multi\mapped_file.obj... ...skipped libboost_iostreams-vc-mt-1_60.lib for lack of file_descriptor.obj... ...skipped libboost_iostreams-vc-mt-1_60.lib for lack of libboost_iostreams-vc-mt-1_60.lib... ...failed updating 4 targets... ...skipped 4 targets... ...updated 10 targets... failed

I have pushed a fix for this on the 'develop' branch of iostreams. In the fix basically iostreams uses _FPOSOFF for versions of VC++ prior to VC++14 but uses '((long long)(fp))' for VC++14 on up. This won't get in the upcoming release but it should get in the subsequent Boost release.

Billy3

_______________________________________________ Unsubscribe & other changes: https://na01.safelinks.protection.outlook.com/?url=http%3a%2f%2flists.boost.org%2fmailman%2flistinfo.cgi%2fboost&data=02%7c01%7cbion%40microsoft.com%7c9a62884203044a54548708d3e0873faf%7c72f988bf86f141af91ab2d7cd011db47%7c1%7c0%7c636098843906180217&sdata=aGAHNfX6BMrM2WyPwb9fCjMLoAjfCv9Ub5aAWiCaM4c%3d

On 9/19/2016 1:32 PM, Rene Rivera wrote:

On Mon, Sep 19, 2016 at 7:18 AM, Edward Diener wrote:

...

I have pushed a fix for this on the 'develop' branch of iostreams. In the fix basically iostreams uses _FPOSOFF for versions of VC++ prior to VC++14 but uses '((long long)(fp))' for VC++14 on up.

That seems like a fragile fix to me. How do you know which version of VC is not going to have that def? Wouldn't it be better to use _FPOSOFF if it's defined, and the cast otherwise? That way whenever the macro goes away it will work, regardless of the specific version when that happens.

I understand what you are saying but _FPOSOFF is "undocumented" in the sense that it should not be relied on. Since I was told in this thread that for VC++14 on up Microsoft plans to "remove" it would it not simply be better not to rely on it for those releases of VC++ ? I could of course easily enough test for it and, if it exists, still use _FPOSOFF, which is what you are suggesting.

some context to the use of _FPOSOFF in boost might be useful. up to and including VS 2015 std::streamoff(std::streampos()) returned _Myoff + _FPOSOFF(_Fpos). there is no other way to retrieve the value of _Myoff from std::streampos. until VS 2012 _FPOSOFF was a cast to long. therefore the only way to safely get the stream offset from a std::streampos was to do: pos.seekpos + std::streamoff( pos ) - _FPOSOFF( pos.seekpos ) == _Fpos + _Myoff + _FPOSOFF(_Fpos) - _FPOSOFF( _Fpos ) == _Fpos + _Myoff after VS 2012 the _FPOSOFF macro is a cast to long long so calling std::streamoff(std::streampos()) returns the correct value and the above workaround is no longer necessary -- View this message in context: http://boost.2283326.n4.nabble.com/Boost-should-not-be-saying-FPOSOFF-tp4687... Sent from the Boost - Dev mailing list archive at Nabble.com.

OK, in more detail: the actual offset in a std::streampos is "_Fpos + _Myoff". std::streampos::seekpos returns "_Fpos" std::streamoff(std::streampos) returns "_Myoff + _FPOSOFF(_Fpos)" to get to the desired "_Fpos + _Myoff" we run: std::streampos::seekpos() + std::streamoff(std::streampos) - _FPOSOFF( std::streampos::seekpos() ) which is equivalent to: _Fpos + std::streamoff(std::streampos) - _FPOSOFF( _Fpos ) which is equivalent to: _Fpos + _Myoff + _FPOSOFF(_Fpos) - _FPOSOFF( _Fpos ) the two _FPOSOFF(_Fpos) cancel out to leave the desired: _Fpos + _Myoff -- View this message in context: http://boost.2283326.n4.nabble.com/Boost-should-not-be-saying-FPOSOFF-tp4687... Sent from the Boost - Dev mailing list archive at Nabble.com.

On 9/19/2016 1:32 PM, Rene Rivera wrote:

On Mon, Sep 19, 2016 at 7:18 AM, Edward Diener wrote:

...

I have pushed a fix for this on the 'develop' branch of iostreams. In the fix basically iostreams uses _FPOSOFF for versions of VC++ prior to VC++14 but uses '((long long)(fp))' for VC++14 on up.

That seems like a fragile fix to me. How do you know which version of VC is not going to have that def? Wouldn't it be better to use _FPOSOFF if it's defined, and the cast otherwise? That way whenever the macro goes away it will work, regardless of the specific version when that happens.

I have changed the fix based on your objection above, since I think you are right.