[LLVMdev] Link time inlining vs compile time inlining (was RE: gold linker failing to resolve weak undefined symbol)

[Balaram Makam]

Changing the subject line and reposting the question.

Bumping up to make sure that this post was not lost during the holidays ;)

Appreciate any feedback.

 

Thanks,

Balaram

 

From: llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at cs.uiuc.edu] On
Behalf Of Balaram Makam
Sent: Saturday, May 24, 2014 10:04 AM
To: llvmdev at cs.uiuc.edu
Subject: [LLVMdev] gold linker failing to resolve weak undefined symbol

 

Hi,

 

I am experimenting with an inline strategy wherein I postpone inlining to
link time rather than inlining at compile time, but I am facing runtime
failure and wonder if anyone else saw this.

Here is my sample test case:

 

$ cat test.cpp

#include <sstream>

int main() {

  std::stringstream ss;

  ss << "hello";

  ss.str();

}

 

To test my inline strategy, I compile at O1 and use flto with gold plugin as
below:

clang -O1 -flto -Wl,-fuse-ld=gold -static test.cpp -o test.exe

 

But the executable segfaults due to a weak undefined symbol:

 

Program received signal SIGSEGV, Segmentation fault.

0x00000000 in ?? ()

(gdb) bt

#0  0x00000000 in ?? ()

#1  0x000221e6 in std::basic_stringbuf<char, std::char_traits<char>,
std::allocator<char> >::str() const ()

#2  0x00009020 in main ()

 

$ nm test.exe|grep
_ZNSs12_S_constructIPcEES0_T_S1_RKSaIcESt20forward_iterator_tag

         w _ZNSs12_S_constructIPcEES0_T_S1_RKSaIcESt20forward_iterator_tag

 

Note that the symbol value is null for the symbol
_ZNSs12_S_constructIPcEES0_T_S1_RKSaIcESt20forward_iterator_tag in the
symbol table.

 

However, when I compile at O3 and use flto with gold plugin, the linker is
able to resolve the symbol correctly and there is no runtime failure:

clang -O3 -flto -Wl,-fuse-ld=gold -static test.cpp -o test.exe

 

$ nm test.exe|grep
_ZNSs12_S_constructIPcEES0_T_S1_RKSaIcESt20forward_iterator_tag

00030139 W _ZNSs12_S_constructIPcEES0_T_S1_RKSaIcESt20forward_iterator_tag

 

Has anyone seen this issue before? Could this be an issue with the gold
linker? 

Is it a bad design strategy to postpone inlining to link stage?

 

Thanks,

Balaram

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140527/b0dbbd3d/attachment.html>