[lld] r268178 - ELF: New symbol table design.

[Davide Italiano via llvm-commits]

I'm sorry to bring bad news but this patch breaks (or exposes an
existing bug) with --lto-jobs=N, with N > 1.
If I try to link llvm-tblgen, I get the following:

[...]

duplicate symbol:
llvm::Twine::toNullTerminatedStringRef(llvm::SmallVectorImpl<char>&)
const in lib/libLLVMSupport.a(Twine.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol:
llvm::Record::getValueAsListOfStrings(llvm::StringRef) const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::Record::dump() const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::DefInit::getAsString() const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::DefInit::getFieldInit(llvm::Record&,
llvm::RecordVal const*, std::__cxx11::basic_string<char,
std::char_traits<char>, std::allocator<char> > const&) const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::IntInit::getAsString() const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol:
llvm::VarInit::getFieldType(std::__cxx11::basic_string<char,
std::char_traits<char>, std::allocator<char> > const&) const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol:
llvm::VarInit::resolveListElementReference(llvm::Record&,
llvm::RecordVal const*, unsigned int) const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::BitRecTy::typeIsConvertibleTo(llvm::RecTy
const*) const in lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::BitsInit::convertInitializerTo(llvm::RecTy*)
const in lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::DagRecTy::getAsString() const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::UnOpInit::resolveReferences(llvm::Record&,
llvm::RecordVal const*) const in lib/libLLVMTableGen.a(Record.cpp.o)
and LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::BitsRecTy::getAsString() const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol:
llvm::FieldInit::resolveListElementReference(llvm::Record&,
llvm::RecordVal const*, unsigned int) const in
lib/libLLVMTableGen.a(Record.cpp.o) and
LLD-INTERNAL-combined-lto-object
duplicate symbol: llvm::SourceMgr::PrintMessage(llvm::SMLoc,
llvm::SourceMgr::DiagKind, llvm::Twine const&,
llvm::ArrayRef<llvm::SMRange>, llvm::ArrayRef<llvm::SMFixIt>, bool)
const in lib/libLLVMSupport.a(SourceMgr.cpp.o) and
LLD-INTERNAL-combined-lto-object

[...]

The cmake invocation looks pretty much like this one:

$ cmake -GNinja -DCMAKE_C_COMPILER="/home/davide/llvm/build/bin/clang"
-DCMAKE_AR="/home/davide/llvm/build/bin/llvm-ar"
-DCMAKE_RANLIB="/usr/bin/true"
-DCMAKE_CXX_COMPILER="/home/davide/llvm/build/bin/clang++"
-DCMAKE_C_FLAGS="-flto -fuse-ld=lld -Wl,--lto-jobs=4"
-DCMAKE_CXX_FLAGS="-flto -fuse-ld=lld -Wl,--lto-jobs=4"
-DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=OFF ../

If I revert the patch, llvm-tblgen links successfully.

Peter, can you please take a look when you get a chance?
As a general consideration, we might want to set up a bot (or enhance
existing bots) to test this configuration (e.g. --lto-jobs=2).

Thanks!


On Mon, May 2, 2016 at 6:45 PM, Rui Ueyama via llvm-commits
<llvm-commits at lists.llvm.org> wrote:
> I sync'ed to the head today, and I saw the improvement on my machine as
> well. This is one of the best patches committed to LLD/ELF. Thank you for
> doing this!
>
> On Mon, May 2, 2016 at 6:41 PM, Peter Collingbourne via llvm-commits
> <llvm-commits at lists.llvm.org> wrote:
>>
>> http://reviews.llvm.org/D19836 is the fix for Rafael's reproducer. Ed,
>> please let me know if it fixes the issue for you.
>>
>> Peter
>>
>> On Mon, May 2, 2016 at 1:02 PM, Peter Collingbourne <peter at pcc.me.uk>
>> wrote:
>>>
>>> On Mon, May 2, 2016 at 12:55 PM, Rafael Espíndola
>>> <rafael.espindola at gmail.com> wrote:
>>>>
>>>> To reproduce the problem:
>>>>
>>>> clang -c test.s
>>>> clang -c test2.s
>>>> llvm-ar rc test2.a test2.o
>>>> ld.lld -shared -o test.so  test2.a test.o
>>>>
>>>> with
>>>>
>>>> test.s:
>>>>         addq    a at GOTTPOFF(%rip), %rax
>>>> test2.s:
>>>>         .section        .tdata,"awT", at progbits
>>>>         .globl  a
>>>> a:
>>>>         .long   42
>>>
>>>
>>> Was able to reproduce, looking.
>>>
>>> Thanks,
>>> --
>>> --
>>> Peter
>>
>>
>>
>>
>> --
>> --
>> Peter
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>

-- 
Davide

"There are no solved problems; there are only problems that are more
or less solved" -- Henri Poincare