Build Rust with MinGW GCC 8.1.0

[mati865]

I've tested it locally on Windows 10 and Wine and it seems to fix #47048.
GDB also should be fixed: #40184 (comment)

This commit is not meant to be merged as is but to run tests with new toolchain.
If it succeeds it can be considered to push toolchain packages to AWS and begin using them.

[Mark-Simulacrum]

r? @alexcrichton

We can't use try here since that doesn't run AppVeyor... we'll probably want to test locally if we decide to go ahead with this.

[mati865]

I see. I've put [WIP] tags to avoid merging it by mistake because I was hoping try will work the same as r+.

In this case somebody with access to AWS could upload packages and open new PR superseding this one.

[Mark-Simulacrum]

@bors r+

Uploaded artifacts to S3 mirror and updated the PR

[bors]

📋 Looks like this PR is still in progress, ignoring approval.

Hint: Remove [WIP] from this PR's title when it is ready for review.

[Mark-Simulacrum]

@bors r+

[bors]

📌 Commit c3474d073a3e8a1aefa4bb3ca08e21ed31f6b992 has been approved by Mark-Simulacrum

[bors]

⌛ Testing commit c3474d073a3e8a1aefa4bb3ca08e21ed31f6b992 with merge 4103f528f03fcd2016e4fdabf4f3203a35d2cfc9...

[bors]

💔 Test failed - status-appveyor

[mati865]

Okay, I wrongly assumed it will work on 32bit if it works on 64bit...

I'll prepare 32bit environment over the weekend and see if I can fix it.

[mati865]

So I ran over dozen of builds locally with i686 GCC 8.1 from mingw-builds with different configure and build options and I was unable to reproduce this error.

I looked into llvm-emscripten and it's using very outdated llvm checkout.
The line in which the error was triggered was later changed to uint8_t instead of char llvm-mirror/llvm@30e9aa6#diff-ed4cb1e2f672825819ef46b07d866eb6L714. This is the only relevant thing I could find.

[alexcrichton]

@mati865 the Emscripten LLVM is based on LLVM 4 which is pretty old at this point, but if there's a patch to send we can push it to our branch and try to fix the compile?

[mati865]

@alexcrichton the thing is I cannot reproduce this error locally with the same compiler and configure options.

[alexcrichton]

@mati865 I can reproduce locally if I download the same compiler, enable emscripten, and then ./x.py build src/llvm-emscripten. Could you try that?

[mati865]

So it turns out running ./x.py changes PATH variable at some point which I used to test different compiler:

$ export PATH="/opt/gcc8/mingw32/bin/:$PATH"
$ gcc -v      # to see if everything is ok

When cmake is being ran from src/bootstrap/native.rs variable PATH is already modified to contain:
"D:\\msys64\\mingw32\\bin;D:\\msys64\\opt\\gcc8\\mingw32\\bin\\;D:\\msys64\\mingw32\\bin;...
Second occurrence of D:\\msys64\\mingw32\\bin is where original MSYS2 PATH starts.

I'll test everything again when I fix my PATH.

[alexcrichton]

Hm odd! I set a custom PATH myself and it worked (via ./x.py), but glad it was found for you at least!

[mati865]

Okay Travis didn't complain, I think it can be tried again.

[alexcrichton]

@bors: r+

[bors]

📌 Commit c1da82a20b548d5f1d21dff3e8da1f6e28547452 has been approved by alexcrichton

[bors]

⌛ Testing commit 3fdafc8 with merge 5be69306aa065f8106e91f42d101e88612335cba...

[bors]

💔 Test failed - status-travis

[rust-highfive]

The job arm-android of your PR failed on Travis (raw log). Through arcane magic we have determined that the following fragments from the build log may contain information about the problem.

Click to expand the log.

[00:17:54] [RUSTC-TIMING] proc_macro test:false 10.505
[00:17:54]    Compiling syntax_ext v0.0.0 (file:///checkout/src/libsyntax_ext)
[00:18:19] [RUSTC-TIMING] syntax_ext test:false 25.125

Broadcast message from root@travis-job-bfe1c554-21f1-4290-ae40-a3e7db5a44cc
 (unknown) at 7:57 ...
The system is going down for power off NOW!
[00:21:54] 
[00:21:54] Session terminated, terminating shell... ...terminated.

I'm a bot! I can only do what humans tell me to, so if this was not helpful or you have suggestions for improvements, please ping or otherwise contact @TimNN. (Feature Requests)

[pietroalbini]

Oh well, the machine got killed.

@bors retry

[bors]

⌛ Testing commit 3fdafc8 with merge 2b6cd0393bb5c5a3b57e6567beafafb5c1542ac4...

[bors]

💔 Test failed - status-appveyor

[pietroalbini]

Ok, this consistently runs out of memory.

[mati865]

There might be no easy fix for that, current AppVeyor configuration has 4 cores and 6 GiB memory + 1 GiB swap which should suffice.
Some of possible workarounds:

• bump llvm hoping unspecified commit reduced compiler memory footprint
• wait for GCC 8.2 hoping it's more optimised
• switch to clang which tends to use less memory. It used to be very problematic with MinGW in the past but version 7 (trunk) received many fixes.

[pietroalbini]

Hmm, @rust-lang/infra, what's the best way to handle this?

[alexcrichton]

Fow now AFAIK we don't have a pressing need to upgrade GCC, so it's unlikely to receive a lot of effort to help push it over the finish line. If no one can think of a solution off the top of their heads we should likely close this for now

[mati865]

While this bug is quite problematic I do agree there is no straightforward fix for now.
Close this PR whenever you want.

I'll reach out mingw-w64 developers to see if something can be done.

[alexcrichton]

Ok, thanks regardless for pushing on this though @mati865!