Can't link against text-2.0.1 on Windows?

[tfausak]

I recently had a build fail for a confusing reason: https://github.com/tfausak/monadoc/runs/7532083433?check_suite_focus=true#step:15:277

Linking D:\\a\\monadoc\\monadoc\\dist-newstyle\\build\\x86_64-windows\\ghc-9.2.3\\monadoc-0.2022.7.27\\x\\monadoc\\opt\\build\\monadoc\\monadoc.exe ...
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000000.o:(.idata$2+0x0): multiple definition of `_head_libstdc___6_dll'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d000000.o):(.idata$2+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000001.o:(.idata$5+0x0): multiple definition of `__imp__ZTVN10__cxxabiv117__class_type_infoE'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006226.o):(.idata$5+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000001.o:(.idata$6+0x0): multiple definition of `__nm__ZTVN10__cxxabiv117__class_type_infoE'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006226.o):(.idata$6+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000002.o:(.idata$5+0x0): multiple definition of `__imp__ZTVN10__cxxabiv120__si_class_type_infoE'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006231.o):(.idata$5+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000002.o:(.idata$6+0x0): multiple definition of `__nm__ZTVN10__cxxabiv120__si_class_type_infoE'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006231.o):(.idata$6+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000003.o:(.idata$7+0x0): multiple definition of `libstdc___6_dll_iname'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006587.o):(.idata$7+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000000.o:(.idata$2+0x0): multiple definition of `_head_libstdc___6_dll'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d000000.o):(.idata$2+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000001.o:(.idata$5+0x0): multiple definition of `__imp__ZTVN10__cxxabiv117__class_type_infoE'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006226.o):(.idata$5+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000001.o:(.idata$6+0x0): multiple definition of `__nm__ZTVN10__cxxabiv117__class_type_infoE'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006226.o):(.idata$6+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000002.o:(.idata$5+0x0): multiple definition of `__imp__ZTVN10__cxxabiv120__si_class_type_infoE'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006231.o):(.idata$5+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000002.o:(.idata$6+0x0): multiple definition of `__nm__ZTVN10__cxxabiv120__si_class_type_infoE'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006231.o):(.idata$6+0x0): first defined here
C://tools//ghc-9.2.3//mingw//bin/ld.exe: d000003.o:(.idata$7+0x0): multiple definition of `libstdc___6_dll_iname'; C://tools//ghc-9.2.3//mingw//bin/../lib/gcc/x86_64-w64-mingw32/10.2.0/libstdc++.dll.a(d006587.o):(.idata$7+0x0): first defined here
collect2.exe: error: ld returned 1 exit status
`gcc.exe' failed in phase `Linker'. (Exit code: 1)
collect2.exe: error: ld returned 1 exit status
`gcc.exe' failed in phase `Linker'. (Exit code: 1)
cabal.exe: Failed to build exe:monadoc from monadoc-0.2022.7.27.
Failed to build test:monadoc-test-suite from monadoc-0.2022.7.27.

I'm not certain that this was caused by text-2.0.1. Compared to the last successful build, that's one of the few things that changed. Here's a diff of the freeze file from the last successful build with that failing build:

102c102
<              any.http-client ==0.7.11,
---
>              any.http-client ==0.7.12,
176c176
<              any.text ==2.0,
---
>              any.text ==2.0.1,
228c228
< index-state: hackage.haskell.org 2022-07-26T09:56:38Z
---
> index-state: hackage.haskell.org 2022-07-27T01:07:47Z

The changes in http-client-0.7.12 don't seem like they could cause this. However I'd be happy to be proven wrong and pointed elsewhere!

If this is indeed caused by upgrading from text-2.0 to text-2.0.1, then I think it's related to #453.

[Bodigrim]

This is likely because of text, but I cannot reproduce neither locally nor on CI.

As a short term solution you can disable -simdutf flag.

Is the solution reproducible with GHC 9.4-rc1? If it does, please forward this issue to GHC developers.

[tfausak]

Unfortunately I don't have a Windows machine to test with. But I can confirm that setting --flags=-simdutf for the text package fixed my problem: https://github.com/tfausak/monadoc/runs/7577924604?check_suite_focus=true#step:10:183

[Bodigrim]

OK, I managed to reproduce with GHC 9.2.4. The problem is that monadoc depends both on text and double-conversion. The latter requires

  if os(windows)
    if arch(x86_64)
      extra-libraries: stdc++-6 gcc_s_seh-1

text-2.0 requires stdc++-6 as well:

text/text.cabal

Lines 93 to 95 in 9075b05

	if os(windows)
	if arch(x86_64)
	extra-libraries: stdc++-6 gcc_s_seh-1

However, text-2.0.1 is more relaxed and does not ask for specific version of stdc++:

text/text.cabal

Lines 93 to 104 in fdb06ff

	if impl(ghc >= 9.4)
	build-depends: system-cxx-std-lib == 1.0
	elif os(darwin) || os(freebsd)
	extra-libraries: c++
	elif os(openbsd)
	extra-libraries: c++ c++abi pthread
	elif os(windows)
	-- GHC's Windows toolchain is based on clang/libc++ in GHC 9.4 and later
	if impl(ghc < 9.3)
	extra-libraries: stdc++
	else
	extra-libraries: c++ c++abi

But under the hood stdc++ and stdc++-6 are the same library, so GHC linker gets confused. This technically qualifies as a GHC bug, but a low-priority one, because GHC 9.4 comes with a virtual package system-cxx-std-lib to provide a portable solution for packages depending on C++ runtime.

double-conversion would have to adopt to system-cxx-std-lib anyway, so it can switch from stdc++-6 to stdc++ for older GHCs at the same time. CC @Haskell-mouse. In the meantime -simdutf will do.

[Bodigrim]

Closing as there is nothing we can fix on text side, but feel free to forward this bug to GHC devs.

[tfausak]

Thanks for investigating! I think this issue is already solved by https://gitlab.haskell.org/ghc/ghc/-/merge_requests/8086, so I won't open another issue against GHC.