Support VS 2017

[brson]

Fixes#38584

This replaces all the MSVC linker logic with that from the 'gcc' crate. The code looks the same, but there could be regressions.

I've only tested this with x86_64.

r? @alexcrichton
cc @vadimcn@retep998

[brson]

I'm seeing issues in cmake that i want to investigate still.

[brson]

Ok, I think this patch is sufficient to fix rustc at least. Both gcc-rs and cmake still need to be updated to be able to find msbuild correctly. I'll file separate issues on those.

[alexcrichton]

@bors: r+

[bors]

📌 Commit 88b9cf8 has been approved by alexcrichton

[brson]

The rustup patch failed on i686.

[brson]

@bors r- I'm still not confident about what's going on in i686

[brson]

@bors r+ I want to see what bors thinks.

[bors]

📌 Commit 88b9cf8 has been approved by brson

[ishitatsuyuki]

Rollup failure.

PS: @frewsxcv please be careful when rolling up non-rollup priority PRs, inspect the PR comments before adding it to list.

[brson]

I'm not sure how the appveyor images are set up, but I suspect that after this lands we will be building Rust releases with VS 2017 instead of VS 2015.

[est31]

The error from the rollup is:

error: linking with `link.exe` failed: exit code: 1171
|
= note: "C:\\Program Files (x86)\\Microsoft Visual Studio 14.0\\VC\\bin\\amd64_x86\\link.exe" "/NOLOGO" "/NXCOMPAT" "/LARGEADDRESSAWARE" "/SAFESEH" "/LIBPATH:C:\\projects\\rust\\build\\i686-pc-windows-msvc\\stage1\\lib\\rustlib\\i686-pc-windows-msvc\\lib" "C:\\projects\\rust\\build\\i686-pc-windows-msvc\\stage1-std\\i686-pc-windows-msvc\\release\\deps\\std-ceb42dcce624041e.0.o" "/OUT:C:\\projects\\rust\\build\\i686-pc-windows-msvc\\stage1-std\\i686-pc-windows-msvc\\release\\deps\\std-ceb42dcce624041e.dll" "/DEF:C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\lib.def" "C:\\projects\\rust\\build\\i686-pc-windows-msvc\\stage1-std\\i686-pc-windows-msvc\\release\\deps\\std-ceb42dcce624041e.crate.metadata.o" "/OPT:REF,ICF" "/DEBUG" "/LIBPATH:C:\\projects\\rust\\build\\i686-pc-windows-msvc\\stage1-std\\i686-pc-windows-msvc\\release\\deps" "/LIBPATH:C:\\projects\\rust\\build\\i686-pc-windows-msvc\\stage1-std\\release\\deps" "/LIBPATH:C:\\projects\\rust\\build\\i686-pc-windows-msvc\\native\\compiler-rt\\." "/LIBPATH:C:\\projects\\rust\\build\\i686-pc-windows-msvc\\stage1\\lib\\rustlib\\i686-pc-windows-msvc\\lib" "advapi32.lib" "ws2_32.lib" "userenv.lib" "shell32.lib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\librand-cbaabb050663bb1e.rlib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\libpanic_unwind-7a8f267e1f472275.rlib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\liblibc-913fac93272423bb.rlib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\libcollections-0e863adcb979d99b.rlib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\libunwind-f8f46edb99891d80.rlib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\libstd_unicode-72685d3c05b2de5b.rlib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\liballoc-2fe699c323a9fcd5.rlib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\liballoc_system-1a9ffce435c6813e.rlib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\libcore-6796d5cf3cb4eed3.rlib" "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\rustc.RkAu4DLaw1LX\\libcompiler_builtins-13752c636400e0ca.rlib" "libcmt.lib" "/DLL" "/IMPLIB:C:\\projects\\rust\\build\\i686-pc-windows-msvc\\stage1-std\\i686-pc-windows-msvc\\release\\deps\\std-ceb42dcce624041e.dll.lib"
= note: LINK : fatal error LNK1171: unable to load mspdbcore.dll (error code: 126) 

The version number 15.0 corresponds to 2015, not 2017. And the linker is the only thing from VS we are running, right?

[brson]

Huh. That's not the error I expected :-/

[retep998]

The version number 15.0 corresponds to 2015, not 2017. And the linker is the only thing from VS we are running, right?

No, 15 corresponds to 2017 and 14 corresponds to 2015.

note: LINK : fatal error LNK1171: unable to load mspdbcore.dll (error code: 126)

This error is caused when you're running a cross linker but you forgot to add the host toolchain to PATH (because the cross linker folder depends on some DLLs from the host toolchain folder). The existing code for 11 12 and 14 does this correctly. More specifically, rustc seems to set the PATH for the linker later on, which means whatever Command I returned would have its PATH overridden so I couldn't at the host toolchain folder that way. I fix this I hacked in some code to add the host toolchain folder to the PATH that was being set later on, and your new version shelling out to gcc doesn't do that correctly.

[retep998]

The issue is right here. path is not supposed to be the directory with the linker, but rather the directory of the host toolchain (aka the folder with the linker whose target and host are both the same as the host of the linker that you are using). You'll have to teach gcc how to provide that information so you can pull it out here. And yes, you'll have to do this for VS 2017 too.

[brson]

Thanks for the review @retep998

[est31]

No, 15 corresponds to 2017 and 14 corresponds to 2015.

Right. I have mistyped. The version number of the pasted text is 14.0 though, so it still corresponds to 2015.

[brson]

@bors r+

I want to see what bors thinks again.

This is currently using a git replacement of the gcc crate.

[bors]

📌 Commit 1d177d5 has been approved by brson

[brson]

@bors r+

I changed the travis config to fail so this won't land with the crate replacement.

[bors]

📌 Commit 0e845bf has been approved by brson

[brson]

@bors r+

[bors]

⌛ Testing commit 5ffc7bb with merge d390024...

[Mark-Simulacrum]

AppVeyor passed; I'm canceling this. https://ci.appveyor.com/project/rust-lang/rust/build/1.0.3454

@bors retry
@bors r-

[brson]

This is ready to go now.

r? @alexcrichton

[alexcrichton]

@bors: r+

[bors]

📌 Commit da100fe has been approved by alexcrichton

[Mark-Simulacrum]

@bors rollup

We know that this passes AppVeyor, and Travis PR passed (and I think we've passed regular travis before).