[PATCH] Fix the --enable-shared build
rafael.espindola at gmail.com
Fri Jul 25 06:15:20 PDT 2014
On 24 July 2014 17:50, Sanjoy Das <sanjoy at playingwithpointers.com> wrote:
>> Passing -fno-devirtualize to broken gcc versions.
> Won't that affect performance?
It would, but the bug is already fixed in gcc upstream, so is that
really a big issue?
> The in-source workaround only fixes
> the issue affecting us, and seems lesser impact that a global
> -fno-devirtualize. The source code impact is likely to be greater as
> Moreover, I could not come up with a test program that links
> successfully with clang++ / older g++ but breaks with g++ 4.9. The
> example on the gcc bug tracker doesn't successfully link with clang
> because of a missing definition for PassNameParser::PassNameParser and
> main, and adding those in passes the test for g++ 4.9 as well. This
> means we'd probably have to hard-code the g++ version number, and not
> use AC_LINK_IFELSE. We could also parse the error messages produced
> when linking, but that seems fragile.
Can you compile the file and check for an undefined reference?
More information about the llvm-commits