Boost C++ Libraries: Ticket #6139: error: "BOOST_CLANG" is not defined

Daniel James — Fri, 18 Nov 2011 11:32:46 GMT

(In [75544]) Detail: Revert [67484] in container_fwd. Refs #6139

owner, status, component changed

Daniel James — Fri, 18 Nov 2011 11:35:38 GMT

owner changed from nasonov to Daniel James
status new → assigned
component any → None

Sorry I'd meant to revert that change before, but forgot to.

attachment set

Jürgen Hunold — Sat, 19 Nov 2011 14:35:28 GMT

attachment → container_fwd_guarded.diff

attachment set

Jürgen Hunold — Sat, 19 Nov 2011 14:35:56 GMT

attachment → container_fwd_simple.diff

Jürgen Hunold — Sat, 19 Nov 2011 14:42:38 GMT

Revision https://svn.boost.org/trac/boost/changeset/75544 just broke my clang builds. The correct solution would be to either

use #if defined(BOOST_CLANG) see container_fwd_guarded.diff
change "class" to struct see containner_fwd_simple.diff

The interesting thing is that container_fwd.hpp contains a msvc-suppression for struct/class mismatch: line 104ff:

#if defined(BOOST_MSVC)
#pragma warning(push)
#pragma warning(disable:4099) // struct/class mismatch in fwd declarations
#endif

So the "correct" solution depends on how the corresponding standard library declares struct or class "complex". At least the guarded approach prevents clang from emitting a warning.

anonymous — Sat, 19 Nov 2011 16:17:56 GMT

Detecting clang is not the right thing to do, it should be based on the standard library not the compiler. How did it break your build?

Jürgen Hunold — Sat, 19 Nov 2011 20:53:26 GMT

clang issues a "-Wmismatched-tags" warning for struct/class mismatches. This is C4099 in msvc as mentioned before.

To be compatible with the msvc-settings, we have -Werror=mismatched-tags to reduce the noise in our sources.

The proper way seems to be do "struct complex;" for all and be happy.

If all standard libraries do it that way, of course.

Daniel James — Sun, 20 Nov 2011 11:02:13 GMT

I tried to fix this in [75561], although I'm not sure if it will work as I don't have clang on windows. If it doesn't, can you build and run config_info (from libs/config/test/config_info.cpp) and attach the output.

I might just change it to always disable forwarding for clang, clang parses the standard headers pretty quickly so it shouldn't make any real difference.

Jürgen Hunold — Sun, 20 Nov 2011 15:19:37 GMT

Does not work. You should #define BOOST_CONTAINER_FWD_COMPLEX_STRUCT for #if defined(BOOST_CLANG). And this is on Linux, my settings are just for reducing cross-plattform compiler/stdlib inconsistencies. I think I'll #define BOOST_DETAIL_NO_CONTAINER_FWD in the meantime.

anonymous — Sun, 20 Nov 2011 16:17:32 GMT

Sorry, I'm not going to do anything until I have a better idea of what your setup is. If you attach the output of config_info.cpp that would help. Or if that's too difficult, you could just tell me.

It'd also help if you explained why you think that the definition of complex is determined by the compiler and not the standard library. I'm not making a change just because it happens to work on one particular setup without understanding the reason behind wanting that change.

On my linux box clang uses gcc's standard library by default, and in that complex is defined as class complex. There must be some reason why it isn't on yours, but without more information I can only guess. At the moment, I don't know which standard library you're using or which version of clang you're using.

attachment set

Jürgen Hunold — Mon, 21 Nov 2011 15:28:55 GMT

attachment → clang_config.info

Output of config_info_threadedfor clang 3.1

attachment set

Jürgen Hunold — Mon, 21 Nov 2011 15:36:14 GMT

attachment → error.txt

clang error backtrace

Jürgen Hunold — Mon, 21 Nov 2011 15:41:36 GMT

Yes, makes sense. I've attached the output of config_info (threaded) and my full clang backtrace of the error. If you need more information or more tests, please ping me again.

Daniel James — Mon, 21 Nov 2011 23:21:34 GMT

(In [75600]) Detail: Fix BOOST_CONTAINER_FWD_COMPLEX_STRUCT. Refs #6139.

My last change was wrong, it isn't struct for Dinkumware. In GCC's standard library it's declared as a class, but later defined as a struct. This is so far untested as I'm building the latest trunk version of clang (the warning doesn't show for the current clang release).

status changed; resolution set

Daniel James — Sun, 11 Dec 2011 21:43:04 GMT

status assigned → closed
resolution → fixed

(In [75909]) Unordered: Fix complex type for gcc's standard library. Fixes #6139.

I haven't had any confirmation, but I assume this fixes the problem.

anonymous — Mon, 12 Dec 2011 13:20:47 GMT

Oops, the subject of that changeset should have started with 'Detail' not 'Unordered'.