[libcxxabi] r228408 - unwind: use -fno-rtti -fno-exceptions -funwind-tables

[Saleem Abdulrasool]

On Fri, Feb 13, 2015 at 11:02 AM, Sergey Dmitrouk <
sdmitrouk at accesssoftek.com> wrote:

> There is no need to revert it completely, not adding -fno-exceptions if
> -funwind-tables is not available should be enough.  This is a workaround
> for CMake issue, but is also a correct thing to do by itself.
> Specifying one option without another literally breaks libunwind, these
> options dependent on each other and should either be both applied or
> none of them should be applied.


How about doing that along with a check that if you are building a DSO
aborting the build with an error that it is unsupported?


>
> --
> Sergey
>
> On Fri, Feb 13, 2015 at 10:48:14AM -0800, Anton Korobeynikov wrote:
> > Since the stuff is heavily broken at this moment, shall we revert the
> > detection stuff for a moment?
> >
> > On Fri, Feb 13, 2015 at 7:36 PM, Saleem Abdulrasool
> > <compnerd at compnerd.org> wrote:
> > > On Thu, Feb 12, 2015 at 1:13 PM, Sergey Dmitrouk
> > > <sdmitrouk at accesssoftek.com> wrote:
> > >>
> > >> How about the attached patch?  Seems to work for me (it's trivial, but
> > >> syntax is weird, so I wasn't sure).
> > >
> > >
> > > It would prevent building unwind as a DSO without libc++abi linked in,
> so,
> > > this isn't the way to handle this.  I think you need to tweak the
> detection
> > > of the -funwind-tables option.
> > >
> > >>
> > >>
> > >> --
> > >> Sergey
> > >>
> > >> On Thu, Feb 12, 2015 at 07:36:14AM -0800, Sergey Dmitrouk wrote:
> > >> > On Thu, Feb 12, 2015 at 07:13:09AM -0800, Saleem Abdulrasool wrote:
> > >> > >    I've been successful with clang as well.
> > >> > >    Reproduction steps would help in identifying what is happening.
> > >> >
> > >> > Looks like I found the culprit, it's CMake.  I'm cross compiling
> > >> > libcxxabi for ARM on x86_64 host, compiler targets ARM only (so ARM
> is
> > >> > also the default target).  cmake/config-ix.cmake contains (also note
> > >> > "_c_"
> > >> > instead of "_cxx_", but it doesn't affect results):
> > >> >
> > >> >     check_c_compiler_flag(-funwind-tables
> LIBCXXABI_HAS_FUNWIND_TABLES)
> > >> >
> > >> > This test fails on linking stage:
> > >> >
> > >> >
> > >> >
> CMakeFiles/cmTryCompileExec3829915952.dir/src.cxx.o:(.ARM.exidx+0x0):
> > >> >     undefined reference to `__aeabi_unwind_cpp_pr0'
> > >> >
> > >> > because successful linking requires symbol defined in libunwind (and
> > >> > that's
> > >> > what is being built).  Now here:
> > >> >
> > >> >     append_if(LIBUNWIND_COMPILE_FLAGS
> LIBCXXABI_HAS_NO_EXCEPTIONS_FLAG
> > >> > -fno-exceptions)
> > >> >     append_if(LIBUNWIND_COMPILE_FLAGS LIBCXXABI_HAS_NO_RTTI_FLAG
> > >> > -fno-rtti)
> > >> >     append_if(LIBUNWIND_COMPILE_FLAGS LIBCXXABI_HAS_FUNWIND_TABLES
> > >> > -funwind-tables)
> > >> >
> > >> > `-fno-exceptions` is added, but not `-funwind-tables`.  As a result
> > >> > libunwind sees "can't unwind" marker in exidx records and gives up.
> > >> >
> > >> > I found this bug report, which seems to be related:
> > >> >     http://www.cmake.org/Bug/view.php?id=15264
> > >> >
> > >> > Not adding `-fno-exceptions` flag if `-funwind-tables` isn't
> available
> > >> > seems
> > >> > to be reasonable by itself, but will also work around this issue
> with
> > >> > CMake.
> > >> >
> > >> > --
> > >> > Sergey
> > >
> > > --
> > > Saleem Abdulrasool
> > > compnerd (at) compnerd (dot) org
> > >
> > > ____________
> >
> > --
> > With best regards, Anton Korobeynikov
> > Faculty of Mathematics and Mechanics, Saint Petersburg State University
>



-- 
Saleem Abdulrasool
compnerd (at) compnerd (dot) org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150213/25b76427/attachment.html>