prefix x86 asm symbols with _ on Windows like on Apple

[jeremyd2019]

Fixes #60

I added i686 to the sha1.yml also, it passes on stable, but fails on 1.43.0 probably due to it not being compatible with the version of mingw-w64 and/or gcc provided by msys2. This also fails on x86_64. I don't know where it was finding x86_64-w64-mingw32-gcc before, probably something random installed on the runner. Now it is definitely using the msys2 version. @mati865

[jeremyd2019]

If this is desired, I could do the other x86 asm files too.

[tarcieri]

@jeremyd2019 you can test it by cloning https://github.com/RustCrypto/hashes and adding a directive to Cargo.toml like:

[patch.crates-io.sha1-asm]
path = "../asm-hashes/sha1-asm"

[mati865]

fails on 1.43.0 probably due to it not being compatible with the version of mingw-w64 and/or gcc provided by msys2

That's very likely. 1.48 was the release when I've implemented the refined version of external toolchain detection mechanism. Before that it was hit or miss.

[jeremyd2019]

Thanks. I am not all that familiar with rust...

jeremyd@XXXX MINGW32 /d/tmp/hashes/sha1
$ cargo test -F asm
   Compiling version_check v0.9.4
   Compiling typenum v1.16.0
   Compiling cc v1.0.79
   Compiling proc-macro-hack v0.5.20+deprecated
   Compiling generic-array v0.14.7
   Compiling sha1-asm v0.5.1 (D:\tmp\asm-hashes\sha1)
   Compiling blobby v0.3.1
   Compiling block-buffer v0.10.4
   Compiling crypto-common v0.1.6
   Compiling digest v0.10.7
   Compiling hex-literal-impl v0.2.3
   Compiling cpufeatures v0.2.9
   Compiling cfg-if v1.0.0
   Compiling sha1 v0.10.5 (D:\tmp\hashes\sha1)
   Compiling hex-literal v0.2.2
    Finished test [optimized + debuginfo] target(s) in 2m 10s
     Running unittests src\lib.rs (D:\tmp\hashes\target\debug\deps\sha1-84c50a79e1c3c373.exe)

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

     Running tests\mod.rs (D:\tmp\hashes\target\debug\deps\mod-8bc81ebefe5b1830.exe)

running 2 tests
test sha1_main ... ok
test sha1_rand ... ok

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.13s

   Doc-tests sha1

running 1 test
test src\lib.rs - (line 14) ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 1.80s

[jeremyd2019]

If this is desired, I could do the other x86 asm files too.

I went ahead and did this, and all hashes tests pass on i686-pc-windows-gnu with -F asm now.

[newpavlov]

Looks good! Can you fix the CI? Simply bumping MSRV should be fine. It's unfortunate and strictly speaking goes against our MSRV policy, but I guess the crates did not work on Windows properly either way and MacOS was broken as well.

[jeremyd2019]

I was going to suggest that I could remove the CI changes from this PR if that would help reviewability. I'm not much of a rust user/developer so I wouldn't know how to go about the proper way of doing this. I could just bump the version in the CI matrix (I'm guessing to 1.48.0 based on @mati865's comment). Also, if this is something you want, I could attempt to update the other CI workflows (I only did sha1 so far) too.

I don't know what version to use to make macos work, either.

[jeremyd2019]

I don't know what version to use to make macos work, either.

Trial and error found the lowest rust version that succeeds on macos is 1.54.0.

[jeremyd2019]

I guess the crates did not work on Windows properly either way and MacOS was broken as well.

The CI did pass with 1.43.0 on x86_64 when using whatever mingw-w64 toolchain it found on the runner (presumably that toolchain was suitably old to work with that older rust version). I switched to using msys2's toolchain because there was no i686 toolchain found on the runner otherwise, and because setup-msys2 was being used I think it was intended that msys2 would be used.

I imagine it was something similar on MacOS, where the toolchain being used on the runner was too new to work with older rust versions, but if an older os/toolchain version were used the crate itself would probably be fine.